BOOK 101002 Ver1

See
discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/262918393
Modeling in Mechanical and Civil Engineering:

Collection of papers from Prof. Nguyen-Dang
Hung's former students
BOOK OCTOBER 2010
READS
144
1 AUTHOR:
H. Nguyen-Dang
University of Lige
176 PUBLICATIONS 970 CITATIONS
SEE PROFILE
Available from: H. Nguyen-Dang

Retrieved on: 17 October 2015
EDITORIAL BOARD
Honorary Editor
Nguyen-Dang Hung, Prof.
LTAS - Fracture Mechanics, Aerospace and Mechanical Department,
University of Lige, Belgium
EMMC Bureau, Ho Chi Minh University of Technology, HCM City, Vietnam
Co-Editors in Chief
Gry de Saxc, Prof.
Laboratory of Mechanics, University of Sciences and Technologies of Lille, France
Nicolas Mos, Prof.
Institut GeM, Ecole Centrale de Nantes, France
International editorial advisory board

Stphane Bordas, Prof.
Department of Civil Engineering,
University of Glasgow, Glasgow, UK
Doan-Duc Bao, Dr.

Tire-Vehicle Engineering Technology Dept.
Goodyear, Luxemburg
Michel Bruyneel, Dr.

R&D Engineer, Analysis Group,
SAMTECH SA, Lige, Belgium
Matthieu Domaszewski, Prof.

Department of Mechanical Engineering,
University of Technology of BelfordMonbliard, Belfort, France
Marc Duflot, Dr.

Multi-scale Materials Modeling Group,
CENAERO, Gosselies, Belgium
Pierre Duysinx, Prof.

Laboratory of Vehicle Engineering,
Aerospace and Mechanical Department,
Mohammed Hjiaj, Prof.

Department of Civil Engineering and Urbanism,
INSA de Rennes, France
Reinaldo Jacques Jospin, Dr.

IEN (Instituto de Engenharia Nuclear),
Rio de Janeiro, Brasil
Patrick Morelle, Dr.

SAMTECH SA, Lige, Belgium
Jean-Philippe Ponthot, Dr.

LTAS - Continuum Mechanics and
Thermomechanics,
Marc Scibetta, Dr.

SCK-CEN, Mechanical Testing Group,
Reactor Materials Research Unit,
Boeretang, Mol, Belgium
Yan Ai Min, Dr.

LTAS - Structural Dynamics,
i
Vietnamese editorial advisory board

Ms. Mai Chi, Deputy-Director
Department of Agriculture and Rural
Development,
Phan Thiet, Binh Thuan, Vietnam
Vu-Duc Khoi, Dr.

Institute of Applied Mechanics,
University of Kaiserslautern, Kaiserslautern,
Germany
Nguyen-Tien Duong, Dr.

Welding Engineering and Metal Technology,
Hanoi University of Technology,
Hanoi, Vietnam
Phan-Hong Quang, Ms.

Deputy-Director of HUNG VIET Technology,
HCM City, Vietnam
Nguyen-Hoai Son, Dr.

Department of Mechanical Engineering,
Ho Chi Minh City University of Technical
Education,
HCM City, Vietnam
Bui-Cong Thanh, Prof.

Faculty of Civil Engineering,
Ho Chi Minh City University of Technology,
HCM City, Vietnam
Chau-Dinh Thanh, Ms.

EMMC Bureau,
HCM City, Vietnam
Le-Dinh Tuan, Dr.

Naval Architecture & Marine Engineering
Department,
HCM City, Vietnam
ii
FOREWORDS
In 1984 I was invited to evaluate documentation submitted to the University of Lige for the
attainment of the degree of Spcial doctorat- Agrg de lEnseignement Suprieur. This invitation
was a pleasure for me since internationally acknowledged research leaders such as Charles
Massonnet and B. Fraeijs de Veubeke had made Lige an attractive irradiation centre of both
structural and computational mechanics.
A candidate who had applied for that degree was a young colleague whom I vaguely knew through
some conferences and publications: Nguyen-Dang Hung.
I was impressed very positively by his thesis on plasticity in view of its comprehensive and, to
some extent, interdisciplinary approach, because it evidenced that the author mastered that broad
area from thermodynamical and microstructural foundations to the computer codes for the
numerical solution of problems occurring in the engineering practice (citations from my report
dated January 30, 1984).
In the subsequent two decades international conferences, papers on journals and meetings gave me
several opportunities to strengthen the above evaluation, but concerning Hung's academic activities
in broader and broader scientific areas, also distinct from plasticity; and there were opportunities to
establish, and to strengthening in time, personal friendship.
At present Hungs leading role as researcher and as educator, both in Belgium and in his original
country Vietnam, is well known in the international community of theoretical and applied
mechanics. On the occasion of his 65th anniversary many students, research associates and
colleagues now celebrate Hungs remarkable career. This commemorative volume represents a
well-deserved tribute through selected papers by former students; most of them are now mature
professors and researchers.
I am glad to join with this brief foreword, but I must say that I am deeply impressed, and so will be
most readers, by unexpected information, provided in the book, on Hungs poetical works in
Vietnamese, and also on his commitments in ethical, educational and political issues through
articles and interviews. Both the interdisciplinary approach to research I had appreciated in his
thesis a quarter of century ago, and his most active appurtenance to two so different countries are
well documented by this book. Moreover they turn out to be associated with, and enriched by,
Hung's rare talent to bridge the gap between the two cultures, the sciences and the humanities: a
gap whose depth in modern times was the subject of the famous memorable complaints and
warnings of Charles Percy Snow.
Also in view of Hungs gift for poetry and for versatility and internationality, this short preface ends
with a wish in Latin (language which was called by Clifford Truesdell res omnium et nullius, i.e.
language of everybody and of nobody): ad multos annos, Hung, ad multos annos!
Giulio Maier
Technical University (Politecnico) of Milan, Italy
iii
PREFACE
It has been more than 45 years since Prof. Nguyen-Dang Hung came to Belgium for university
studying, scientific researching and finally academic lecturing at the University of Lige, Belgium.
It will be also 15 years since he had oriented a part of his teaching tasks to Vietnam and created the
Excellence Centres for postgraduate training in Vietnam, EMMC at Ho Chi Minh University of
Technology (since 1995) and MCMC at Hanoi University of Technology (since 1998).
During that long period where he devoted himself to the international post-graduate education in
Vietnam, in the field of computational Mechanics and Civil Engineering, an important number of
students have successfully defended their Master and Ph.D. theses both in Vietnam and in Belgium.
His retirement is an opportunity for his former students in different countries to pay tribute to his
life and work. A collection of selected scientific papers written to that effect is gathered in the
present book. The Editorial Advisory Board is composed with Prof. Nguyen-Dang Hungs elder
students who currently are university professors, eminent members in research centres and
enterprises in both Europe and Vietnam. All accepted papers have been reviewed.
As my own field of specialty is Numerical Analysis and Optimisation, which is not so far from him,
I appreciate very much all that he has done for training young Vietnamese researchers in
Computational Mechanics and Civil Engineering. Under his supervision, two international
postgraduate training centres have graduated about four hundred Vietnamese young scientists both
in Master and Ph.D. programmes. So far these are the only programmes in Vietnam, where modest
students are offered standard international level technological studies programmes and can obtain
European official diploma, after successfully defending their theses. This new idea of delocalised
international university centres helps to substantially reduce the expenses and to effectively limit
the brain drain in developing countries like Vietnam.
I also appreciate Prof. Nguyen-Dang Hungs interest and devotion to education revival and reform
in Vietnam. I have read a lot of his papers in Vietnamese mass media and I am very glad to see that
we share the same opinions, the same values concerning the urgent necessity for fundamental
changes in the Vietnamese university system.
I am very pleased that he was among 23 outstanding local and overseas Vietnamese scientists and
university professors who approved and signed the Propositions for the revival, modernisation
and reform in the Vietnamese education system which constituted an important event in Vietnam
two years ago and still are a priority subject for the new Vietnamese government.
It is with regret that I learned about his retirement, which entails the cessation of his postgraduate
programmes for Vietnam. In fact, it would be very difficult to find somebody else to replace him in
that job, somebody with both the academic and scientific experiences in advanced countries and the
spirit of devotion and self-sacrificing he has so far demonstrated.
Nevertheless, it is also my hope that Prof. Nguyen-Dang Hung will take the opportunity of his
retirement to come back more frequently to his native country and pursue his efforts in supporting
the modernisation of our university education system.
On the behalf of Prof. Nguyen-Dang Hung, I would like to thank all his former students for their
kind contribution to the edition of the present book, particularly the two principal editors Prof. Gry
de Saxc, Prof. Nicolas Mos and Prof. Stphane Bordas who had the nice gesture to join the team.
It is important to mention also his close collaborators Phan-Hong Quang who had the initial idea
and Chau-Dinh Thanh who accepted the coordination job.
Hanoi, September 12, 2006
Prof. Dr. Hoang Tuy, Institute of Mathematics,
Vietnam National Centre for Science and Technology
vi
Dedicated to Professor Nguyen-Dang Hung and his family
CONTENTS
Editorial Board
Forewords
i
iii
Preface
Contents
ix
Nguyen-Dang Hung: the Scientist and the Man
Gry de Saxc and Duong-Thuy Phuong-Khanh
Prof. Nguyen-Dang Hungs Curriculum Vitae
11
International Editorial Advisory Members Papers

A Mtis Finite Element for Singular Stress Analysis of Junctions
and Delamination Cracks and Composite Structures
43
Gry de Saxc, D. Crpin and M. Pyrz
Image-based Micromechanics Analysis using Level Sets

and the Extended Finite Elements Method
55
Ionescu, Nicolas Mos, P. Cartraud and M. Bringhier
Smooth Strain Finite Elements: Selective Integration
62
Nguyen-Xuan Hung, Stphane Bordas and Nguyen-Dang Hung
Topology Optimization of Membranes Made of Orthotropic Material
78
Vincent Rion and Michel Bruyneel
Limit Analysis Theorems for Materials with a Non-Associated Flow Rule
88
Mohammed Hjiaj
Cold Roll Forming and Metal Cutting Simulation

using a 3D Arbitrary Lagrangian Eulerian Formulation
104
R. Boman, L. Papeleux and Jean-Philippe Ponthot
Fracture Toughness Measurement using CircumferentiallyCracked Round Bar
118
Marc Scibetta, R. Chaouadi and E. Van. Walle
Analytic and Numeric Shakedown Solutions of Cylinder

and Sphere Structures under Thermal Loading
141
Yan Ai Min
EMMC & MCMC Students Papers

Metis Element Method for Delamination in Composite Laminate
156
Nguyen-Tien Duong and Nguyen-Dang Hung
A United Algorithm for Limit State Determination of Frame with Semi-Rigid Connections
168
Nguyen-Dang Hung and Hoang-Van Long
Modified 8-Node Quarter-Point Element in Calculating Stress Intensity Factors

of 2D Crack Problems
184
Le-Hoai Long and Nguyen-Dang Hung
Dual Analysis for Finite Element Solutions of Kirchhoff Plates
194
J.F. Debongnie, Nguyen-Xuan Hung and Nguyen-Huy Cung
ix
Equilibrium Model in the Element Free Galerkin Method
207
Bui-Quoc Tinh, Marc Duflot, Stphane Bordas and Nguyen-Dang Hung
Active Balancing of Monorotor during Operation
218
Le-Dinh Tuan
A Refinement Criterion with p-Version in the Error Estimate

and the Convergence Rate of the Finite Element Analysis
229
Nguyen-Hoai Son
A Hybrid Stress Approach for High-Accuracy of Stresses

and Low Sensitivity to Mesh Distortion
237
Dang-Dinh Thi and Nguyen-Dang Hung
Puncture Mechanics of Rubber Membranes
255
Nguyen-Chien Thang, Vu-Khanh Toan and Jaime Lara
Prediction of Delamination Strength at Interface

between Thin Film and Substrate by Cohesive Zone Model
268
Do-Van Truong, Hiroyuki Hirakata, Takayuki Kitamura
Creating Macro Commands by using Bacon Commands to Simplify Meshing

of 3-Dimensional Cracks and Calculate the Lifetime of Cracked Structures due to Fatigue
280
Nguyen-Dang Hung, Tran-Duc Han, Chau-Dinh Thanh and Nguyen-Van Hieu
Influence of Moment-Shear Interaction to T-stub Effect in Bolted Column-Beam Joints
294
Nguyen-Thanh Son, Nguyen-Van Yen and Bui-Cong Thanh
Determination of the Fatigue Characteristics of Cracks by an Inverse Analysis
304
Hang-Minh Dinh, Nguyen-Dang Hung
Limit Analysis of 2D Structure using Gliding Line Mechanism

Generated by Rigid Finite Elements
318
Phan-Hong Quang and Nguyen-Dang Hung
An Application of Homogenizations Method for the Behaviour Elastic of Polymer Concrete
329
Nguyen-Huy Gia, E. Ghorbel, S. Ortola
Dual Limit Analysis of Plate Bending
340
Le-Van Canh, Nguyen-Xuan Hung, Nguyen-Dang Hung
Limit Analysis of Cracked Structures by Primal-Dual Optimization
354
Vu-Duc Khoi
Initiation of Interface Crack at Free Edge between Thin Films with Weak Stress Singularity
371
Effect of Frequency on Fatigue Crack Growth along Interface

between Copper Film and Silicon Substrate
382
Two Simple Flat Shell Elements for Elastic Analysis of Thin Shell Structures
393
Nguyen-Ngoc Duong, Nguyen-Xuan Hung, Nguyen-Van Hieu and Nguyen-Dang Hung
Index
405
Collection of papers from Prof. Nguyen-Dang Hung's former students
NGUYEN-DANG HUNG: THE SCIENTIST AND THE MAN

Gry de Saxc* and Duong-Thuy Phuong-Khanh+
*
Laboratory of Mechanics of Lille, UMR CNRS 8107,
University of Lille 1, France
+
Communication Manager Sanofi-Aventis, Ho Chi Minh City, Vietnam
The researcher and the professor
Nguyen-Dang Hung was born in 1941 in Vietnam. He spends his youth in a poor village in
central Quang Nam D Nang province. Being an outstanding student at Ptrus Truong Vinh Ky
College, he has the chance to obtain a Belgian scholarship to go to Belgium in 1960 and to
undertake engineering studies at the University of Lige. The poverty of his native country, his
mother killed during the dark period of the resistance war against the French occupation and above
all, his fathers love drive him to succeed brilliantly in his studies. In 1966, he graduates as
Engineer Physicist from the Laboratory of Aerospace Techniques (L.T.A.S.), a centre founded by
Professor Fraeijs de Veubeke and of worldwide reputation in the scientific community.
In 1966, he is hired in the L.T.A.S. as research engineer. Midst of this top level team, he is
hooked on the research passion that would not leave him yet. Fully immersed in Fraeijs de Veubeke
variational and numerical techniques, he dedicates his first works to the duality between
displacement and equilibrium methods. It will be a leading thread in the whole of his scientific
work. Nguyen-Dang Hung says sometimes the scope of duality goes beyond the science field and
may be understood in a large sense as a law of the nature and the life, which can be compared to
similar ideas in the oriental philosophy such as the yin and the yang. Besides his career takes soon a
new and important bend when in 1968 he moves to the Department of Strength of Materials and
Stability of Constructions headed by Professor Charles Massonnet, another impressive personality
of the Engineering Faculty of Lige in the closed domain of Civil Engineering.
Nevertheless, he is still preoccupied by the duality topic and he proposes what will undeniably
be one of its noteworthy and seminal contributions, the family of mongrel finite elements.
Mongrel or mtis elements are two-field hybrid elements exhibiting monotonic and very fast
convergence with respect to the corresponding pure elements. The mongrel elements are
intermediate between the standard hybrid elements, which they share a good convergence rate with,
and the pure ones of which the convergence is monotonic too.
But it will not be long before Charles Massonnet with his great energy and enthusiasm attracts
him into another exciting topic, the plastic limit analysis. Already trained in structural computation,
Nguyen-Dang Hung develops the computing package CEPAO for the plastic limit analysis and
optimisation of typical civil engineering structures, the metallic plane frames. Certainly the main
originality of his code with respect to other existing ones is its ability to automatically generate the
collapse mechanism equations.
Coupled with a simplex solver, it is a powerful numerical tool to compute the limit load of such
structures. This pioneering work is the starting point of fruitful research in various connected topics.
The numerical code will soon be enhanced with a plastic optimisation module. A breakthrough
improvement consists in introducing additional controls on the stability rules according to the
ECCS European recommendations, converting an academic code into a practical tool for engineers
in research consultancy. Without endeavouring to mention all the features of CEPAO, we can
1
emphasize among them the extension to reinforced concrete frames and the shakedown optimisation
module accounting for variable repeated loads.
At 30 years old
In his LTAS-Fracture Mechanics office at 60

years old
Another computing package developed during the 70s and 80s decades is ADELEF. NguyenDang Hung glimpses the power of the mathematical programming solvers combined with the finite
element method. It will be the centre of various research activities. The use of stress field
approximations seems especially suitable in plastic analysis but enforcing the yield condition
anywhere was a hard task. He proposes an original method based on the criterion of yield in the
mean leading to a relaxed variational principle allowing to compute the limit load of twodimensional structures. Limit analysis is an elegant method to bypass the complete loading history
before the collapse but it is drastically limited to problems with proportional loading, what is often
rather unrealistic in many applications.
The method reveals all its strength when applying to variable repeated loads: this is the
shakedown approach already mentioned. The spreading of knowledge is today widely provided by
the scientific literature in the journals and proceedings but it is not the unique way. In researchers
life, there are events that can be decisive. Such was his meeting with Jan Knig who explains him
the keystone ideas of the shakedown approach. Together, they write a paper about how to use the
finite element formulation for shakedown problems that was published in 1976 and will be the
starting point of works covering a large scope of engineer structures, two- and three-dimensional
structures, rotational plate and shells, elbows. A first token of the top level in this research field is
the participation to the European Union sponsored BRITE-EURAM Project LISA devoted to the
"FEM Based limit and shakedown analysis for design and integrity assessment in European
industry" (1988-2002).
Around the core of ADELEF package initially dedicates to the limit state problems, other
modules are developed to perform visco-plastic and contact analysis. All these works are achieved
by a team of young researchers supervised by Nguyen-Dang Hung and among whom can be
2
mentioned P. Desir, P. Detroux, P. Falla, D. Frangopol, R. Jospin, P. Morelle, L. Palgen and Gry
de Saxc. All this wealth of scientific results is gathered and structured in his Special Ph.D.
dissertation that conferred him the title of "Agrg de Facult". In 1982, Nguyen-Dang Hung
organises in Lige a workshop on the "Applications of the Mathematical Programming Method to
Structural Analysis and Design" and drawing forty or so participants from European countries. In
parallel with his research activities, he is going up through the academic grades: Assistant in 1968,
Assistant Professor in 1972, Lecturer in 1978, Associate Professor in 1984 and full Professor in
1991.
Shortly afterwards, his research work is changing for a new domain that will modify deeply his
career: the fracture mechanics, a discipline strongly used by Aerospace and for which new
numerical tools in the commercial computational mechanics packages are needed by such
industries. SAMCEF, the finite element code of the L.T.A.S. has become a reference code in this
industrial sector and is commercialized by the SAMTECH Company. In 1987, Nguyen-Dang Hung
fulfils a one-year contract with SAMTECH to implement fracture mechanics tools into SAMCEF
for the S.E.P. firm. This first contract will be followed by other ones financed by the
AEROSPATIALE firm.
As a natural consequence of this important evolution, Nguyen-Dang Hung gets back in 1991 to
the L.T.A.S. as Professor and Head of the Section "Fracture Mechanics of Solids" and he will
remain there until now. This last period of his career is marked by an intensive contractual activity
in fracture mechanics with the companies and the regional and European institutions, in the spirit of
a high level theoretical and numerical research turned to the industrial and economical
development. From 1991 to 1995, he gets himself involved in the program "Multimaterials"
sponsored by the Walloon Region and devoted to the "development and assessment of the damage
phenomenon of composites". It is followed by the European COST action on the "modelling and
simulation of damage growth and crack extension in metallic and composites structures subjected
to fatigue with the presence of residual stresses" from 1995 to 1999, and the COMDY project
sponsored by the Walloon Region on the "delamination and damage of laminated composites under
dynamic loadings" from 2002 to 2006. Among these far-reaching investigations, we only would
like to mention the development of the computation package BECOME bases on the boundary
element method which turns out to be particularly efficient modelling crack problems. In this
matter, we can perhaps perceive the influence of Charles Massonnet whose the pioneering work on
the solving of the integral equations is often considered as precursor of this numerical method.
This main activity on fracture mechanics does not prevent him from pursuing its previous
researches on the shakedown load computation but, from then on, in a contractual way. ELSA, a
specialised version of ADELEF to elbows, is developed and used in the frame of a contract with
FRAMATOME Company. New contracts financed by the European community in the domain of
the "Nuclear Science and Technology" are concerned by the shakedown analysis of pipes and
elbows, in collaboration with the Polytechnic Faculty of Mons and the British company NNC.
During this second L.T.A.S. period, the list of students and collaborators of Nguyen-Dang Hung
is long: Nicolas Mos, B. Borle, Bui-Cong Thanh, Kang Chi Hang, Bui-Quoc Viet, Yan Ai Min,
Doan-Duc Bao, M. Duflot, S. Bordas, E. Marchal, Jiang Yansheng, Nguyn-Dai Quy, NguyenTien Duong, Nguyen-Van Hieu, P. Schils, M. Scibetta, Vu-Duc Khoi, Do-Quang Khanh, NguyenDinh Giang, Dang-Dinh Thi, Tran-Duc Han, Chau-Dinh Thanh, Tran-Thanh Ngoc, Do-Van
Truong, Phan-Hong Quang, Bui-Quoc Tinh, Nguyen-Vinh Phu, Nguyen-Xuan Hung, Hoang-Van
Long
Nguyen-Dang Hung is the author of more than 20 books and courses including a dictionary, and
more than 200 scientific works, most of them published by prestigious magazines throughout the
world. He was awarded a medal by the Belgian Royal Academy of Sciences, Literature and Arts
3
and was decorated by the Belgian Government Grand Officer of King Leopold II 's orders in 1999
and is proposing to be decorated Grand Officer of the Crowns order in 2006. It is worthwhile to
mention that in July 1999 he was selected by "Le Vif lExpress" Belgium weekly magazine as one
of the ten active foreigners "who made Belgium move" but it is another facet of his personality that
we shall present in details in the following.
Developing the human resources for Vietnam
There were three waves of immigration from Vietnam: the first one date from the French
colonization, the second one which Nguyen-Dang Hung belongs to, consists of students, some of
them being scholarship holders. The latter wave is the one of the "boat people", a more diversified
population that is found in all the jobs, especially in the restaurants.
With Binh Thuan province officials. Behind them

is the station built in Mui N, Vietnam, dedicated to
coastal measures, ANTIERO project supported by
Brussels-Walloon Region (July, 2006).
With Congolese students in Kinshasa (1983)
Nguyen-Dang Hung has played an active part in the activities of the Association of Patriotic
Overseas Vietnamese in Belgium. He uses to join demonstrations in Brussels or Lige against the
US war and in favor of the recovery of peace in Vietnam. In 1976, after the national reunification,
Nguyen-Dang Hung returns to Vietnam as a guest of the State Committee for Sciences and
Technology, the former name of the Ministry of Sciences and Technology, and gives lectures
during two weeks at the Institute of Transportation of Hanoi where the auditors are researchers
gathering from various Institutes of Technologies of Hanoi: Institute of Technology, Civil
Engineering, Water Resources, Mechanics
He maintains his Vietnamese nationality and would like to prepare his return to participate to the
development of Vietnam. Unhappily, the after war Vietnams situation was not as he could imagine.
One may find in his famous article: The overseas Vietnamese's and the scientific co-operation with
Vietnam (DOAN KET N 418, Paris, November 1989), all his deception about the former
economic and social orientation in his beloved Vietnam during the first ten years of the
reunification.
But during this short stay, he also notes that: Vietnam has abundant human resources and the
young people possess a traditional willingness to learn. Because of the long war, young Vietnamese
could not have the opportunity to go abroad and to receive an international education". Willing to
develop the human resources of Vietnam, Dr. Nguyen-Dang Hung understands that this objective
could only be achieved on the ground of a solid partnership embedded in an official program.
However, only until after 1989, with the "doi moi" policy, he may find the suitable economic and
politic opening to realize his projects. To reach this objective, he weaves a relationship network
thanks to several small projects taking off from 1990 and financed by various national and
international institutions (University of Lige, C.G.R.I., F.I.C.U.-A.U.P.E.L.F.-U.R.E.F now AUF,
Agence Universitaire Francophone).
With his own prestige and patience, he tries to persuade the foreign sponsors and the European
partners to finance the program. He also does his best to persuade the Government of Vietnam to
approve the program. This is not a simple action in that period. Thanks to his patience, tenacity and
efforts, in 1995, the program "European master in Mechanics of Construction, EMMC" for postgraduate inter-university training program has been launched. Another program: "European Master
in Computational Mechanics of Continua, MCMC" is set up in Hanoi, three years later. At the
beginning, this project is supported by the National Agency of International Cooperation for
Development of Belgium. The post-graduate program is built on the European standards. In case of
success, the University of Lige delivers a diploma. Extra fellowships are friendly attributed to the
best graduates to continue their studies in Europe, USA, Canada, Japan and Australia for higher
level, mostly doctorate degree.
So far, both centres, which have invited renowned professors from various Belgian and later
European universities to lecture, have graduate more than 350 masters, of whom 20 are working
towards Ph.D. degrees around the world, mostly in Belgium. Of those graduated masters, 20% go
to abroad for Ph.D. degree, 30% are lecturing at universities in the country while the remaining
50% are working for state agencies, enterprises and businesses. Using his prestige, Nguyen-Dang
Hung has helped the Vietnamese Ministry of Education and Training coordinating with universities
in Europe to carry out a new Ph.D. training program for Vietnam's students. Under this program,
trainees will spend one-third of their study time in Vietnam and three-third in European countries.
Nguyen-Dang Hung says: "This kind of graduate program helps students gain international
standard knowledge while saving training expenses". Moreover, in a social viewpoint, it allows
avoiding the dramatic damaging effects of the brain drain to the developed countries. In a practical
viewpoint, it makes more efficient the influence of the university scientific research on the technical
and economical problems of the Vietnam. There is no doubt that the philosophy of these projects is
partially inspirited by his first experience of Belgian cooperation as Associate Professor at the
University of Kinshasa in Zaire from 1982 to 1986.
In 2001, EU helped Vietnam developing information technology, including improving quality of
doing research and teaching. The first program is done in Ho Chi Minh City, which involved in
universities of Technology in Ho Chi Minh City, Lige, French Montpellier II and Dutch TU Delft.
The second program is implemented in Hanoi with the participation of Universities of Lige,
French Aix-Marseille, Sweden TU Lule, and five Vietnamese Universities of Technologies in
Hanoi. In 2003, an agreement was signed by Belgium Lige University and Vietnamese Civil
Engineering University to graduate 50 Ph.D. students during 10 years. This is the cooperation
between ten Vietnamese technology universities and ten famous ones in Europe.
There are more than 100 articles publishing in the Vietnamese newspapers speaking about the
heart and the enthusiasm of Prof. Hung dedicated to the Vietnamese post-students or for on-the-job
trainees. But few persons know his worrying for them when they first arrived in Belgium.
As most of them went abroad for the first time, there were several difficulties for them to get
used to the local daily life. Nguyen-Dang Hung has to arrange accommodation for the new students
and to explain to them the necessary procedure and formalities, the cultural shock, the working
method.
Opening ceremony of the EMMC programme

(With Belgian Consul and Rector Truong Minh Ve of Ho Chi Minh University of Technology, 1995)
With Rector of the University of Lige, Bernard Rentier at the ceremony of delivering diploma to the students of the
EMMC-2 class
He also has to solve many problems causing by the clumsiness of the Vietnamese students to the
Belgium people such as secretaries in office or managers in dormitory. He is not only interested in
developing the knowledge to the Vietnamese students, but he also wants that Vietnamese could
have self-confidence and could overcome the timidity. Thereby, he encourages Vietnamese to
participate in the cultural activities. In the occasion of Lunar New Year party organised by the
Embassy of Vietnam in Brussels, most of Vietnamese students in Liege perform several songs for
the Vietnamese and international audience under the leading of their enthusiast bandmaster:
Professor Nguyen-Dang Hung. Consequently, after the show, there were a number of Europeans
investors asks him about the Vietnamese students trained in Belgium and they contact those
Vietnamese immediately.
6
Inauguration of EU-EMMD programme and diploma distribution of promotion MCMC-I

in Hanoi, 2001
Vietnamese newspaper cutting

dedicated to EMMC and MCMC program
There are about three million Viet kieu (overseas Vietnamese) living and working in foreign
countries, and more than 300,000 of them are graduated of post-secondary education. On January
27, 2005, the mobilization committee of the Overseas Vietnamese's Science and Technology Club
was inaugurated. Nguyen-Dang Hung belongs to the Provisory Executive Committee and the action
of which among the others is expected creating breakthroughs in attracting overseas Vietnamese
back to Vietnam. Nowadays, the Vietnamese Government becomes aware of the need encouraging
active contributions by overseas Vietnamese. On February 18, 2005, nineteen Viet kieu, among
them Professor Nguyen-Dang Hung were decorated for their contributions to the nations
development. The awards ceremony, titled Vinh danh nuoc Viet (Viet Nams Glory), is the first
of its kind organised by VietnamNet online newspaper, with support from the Viet Nam Fatherland
7
Front. Besides, he is awarded by the Foreign Ministry of Vietnam in 2001 and two times (2002, 2006)
by the Popular Committee of Ho Chi Minh City.
With his LTAS - Fracture Mechanics staff,

University of Lige, 1999
With wife and youngest son at WCCM VI Sept. 5-10, 2004, Beijing
Recently (2004), in an interview, the journalist asks him about his future. He says: I would like
using the rest of my cycle of times to continue some little useful things which give my life a sense. Then,
I would like to be back to my childhood, passing the night near a source murmuring with transparent
water current or to be close to a sea coast gathering the eternal music of waves coming from
eternity
All of us, we hope he could realise these modest wishes during his actual retirement, surrounding by
his crowded and united family with four children and more than eight grand children.
The poet
Far from the scientist creativity, Nguyen-Dang Hungs talent is many-sided: he is also the
author of poem collections in Vietnamese. It seems to us worthwhile to illustrate this facet of his
personality by reproducing this typical one translated from Vietnamese by himself in French.
8
With LTAS (Aerospace Department) colleagues,

University of Lige, June 2006
With colleagues and graduated students,

Aerospace section, June, academic year 2005-2006
Une lettre de l't

(Bc th ma h, Translation for my son Thierry)
Il m'est arriv de me promener dans le Grand Nord,

Et j'attendais en vain la fin du jour
Car le rayon du soleil ne s'teignait jamais, l't est de retour
Sur les collines des pins bercs par le vent
J'entendais comme si l'espace se tordait
Et tout le pass me revenait, comme un tourment
Et m'loignait du sommeil
Je pensais aux temps o tu n'tais qu'un petit enfant
Tu restais seul la maison dans cette ville bante
O est maman? Demandais-tu d'une voix sans pareil!
Je pensais ces sentiers de notre village, inonds de soleil
J'ai suivi grand-mre visiter les tombes des disparues
Je pensais ces vies dcousues
A ces existences simples et phmres
A ces pluies, ces vents, ces boues autour de cette chaumire
Et mon coeur se serrait comme une blessure mal soigne
Je pensais ce petit orphelin lev par ces gens
Et par miracle, il a grandi sans trop de mal
Et quand il commence comprendre que la vie a un sens
Ils sont tous partis sans laisser le moindre bruit
Maintenant la paix est revenue et la vie continue
Mais les proches se sont disperss travers les quatre continents
Et les amis sont loin et on ne sait pas s'ils sont vivants
Oui la guerre est triste, une tristesse dconcertante
Mais qui peut dcrire toutes ces dtresses immenses?
En tout cas, moi je les porte comme un destin
Et mme par chance, la vie parfois me sourit
Elles surgissent un jour comme aujourd'hui, un beau matin
Ne pouvant point dormir, je compose ces lignes

Et je note sur cette lettre : ici c'est le Grand Nord, le jour y est sans fin!
(Lule, Sweden, June 1998, Traduction Lige, March 9, 2004)
With his crowded and united family,

Herstal, Lige, Belgium, May 2006
With G. de Saxc at the Eighth Int. Conf. Comp. Struct. Techn.,
Las Palmas, 12-15 September 2006
Colleague witness accounts
Gry de Saxc
It was a privilege and a pleasure to know Professor Nguyen-Dang Hung, first as teacher and
supervisor of my Master Degree stage, next as my Ph.D. supervisor and colleague. He is a
passionate and busy man whose the enthusiasm in research is communicative. It often occurred that
we wrangle together about scientific or politic subjects but these discussions were always enriching
and friendly. He is an outstanding developer of computation codes and one of the essential thinks
that he taught me is being always aware of the underlying theoretical bases. He is a nationalist in
the noble sense of this word: he is a man who is able of mobilising the energies of his fellow
countrymen to build together a better future and life.
At last, as good Vietnamese, he is a great jasmine tea drinker!
Duong-Thuy Phuong-Khanh
Recently, I visit his office located in Ho Chi Minh University of Technology and I have seen
again this very busy man who has just arrived in Ho Chi Minh City from Belgium. I ask him about
the motivation of his constant enthusiasm... He tell me possibly this is due to the fact he had so
much suffered during the French resistance war when he was a child living in Quang Nam
province... He had the chance to be alive, another chance to obtain Belgian scholarship to go
Belgian university... He seems better to do something useful for young Vietnamese and this feeling
pursue him all the times during these long years devoting for this education action...
10
Prof. NGUYEN-DANG HUNGS CURRICULUM VITAE
Personal coordinates
Name:
NGUYN- NG HNG
Date and place of birth: 1st January 1941, Dien Ban, Quang Nam, Vietnam.
Nationality:
Belgian and Vietnamese
Civil status:
Married with Dng-Th Hunh-Mai,
Children:
Two sons and two daughters, Nguyn-ng Thi (Thierry, 1965), Nguynng Kiu-Mai (Delphine, 1973), Nguyn-ng Kiu-Anh (Ccile, 1974)
and Nguyn-ng Hng-Nguyn (Constant, 1994).
Grand-children:
Four grand-sons and four grand-daughters (2006)
Homepage:
http://www.ltas-rup.ulg.ac.be/dang
http://www.ulg.ac.be/ltas-rup/dang
Address:
Institut de Mcanique et Gnie Civil, Bt. B52/3,
Chemin des Chevreuils 1, B-4000 Lige, Belgique
Tel : 32-4-3669 240 , 32-4-3669 241,
Fax : 32-4-3669 311 , 84-8-8655351
E-mail:
H.NguyenDang@ulg.ac.be
Education / Professional studies
Secondary school:
Ptrus Truong Vinh Ky, Saigon, Vietnam (1954-1959)

and Chu Van An, Saigon, Vietnam (1959-1960).
University Institution: University of Lige, Belgium
Degree/Diploma:
Engineer in Physic-Aero-Space Engineering (1961-1966)
Degree/Diploma:
Special Ph.D. (Agrg de l'Enseignement Suprieur, 1984)
Membership of professional bodies
Member of European Mechanics Society (EUROMECH).

Member of the "Association des Ingnieurs sortis de l'Universit de Lige" (A.I.Lg.).
Member of Canadian Society for Mechanical Engineering.
Member of I.A.S.S. (International Association for Shell and Spatial Structures).
Member of the Reviewer Corps of the Applied Mechanics Reviews.
Member of the CEE-Working Group AG2 (Structural analysis of nuclear components).
Member of the Management Committee of COST Action 512 (MMSP)
Vice-President of the Provisory Management Committee of the Overseas Vietnamese Scientits
Club.
Honours and awards
Awarded by Belgian Royal Academy of Sciences, Literature and Arts (Louis BAES prize) (1984).
Decorated by the Belgian Government: Labour medal of first class (1998).
Chosen by LE VIF-EXPRESS (16/7/1999) to be among the immigrated Belgians who participate to
the change of Belgium.
Decorated by the Belgian Government: Grand Officer of King Leopold IIs order (1999).
11
Awarded by the Foreign Ministry of Vietnam (2001).

Awarded by the Popular Committee of Ho Chi Minh City, Vietnam (2002).
Awarded by the Fatherland Front's Central Committee and the VIETNAMNET magazine: Glory
of Vietnam (2004).
Decorated by the Belgian Government: Belgian Civic Cross of first class (2005).
Awarded by the Popular Committee of Ho Chi Minh City, Vietnam (2006).
Proposed to be decorated by the Belgian Government: Grand Officer of the Crowns order
(2006).
Present position within the organizations
Member of the Executive Manager Comity of LTAS (Laboratory of Aero-Spatial Techniques:

L.T.A.S).
Full Professor and Head of the Division LTAS-Fracture Mechanics of Solids (1991-2006),
University of Liege, Belgium.
Years with the organizations: 40 years
Research Engineer in the Department of Prof. B. FRAEIJS DE VEUBEKE (1966-1968),

(Laboratory of Aerospace Techniques (L.T.A.S.), University of Lige, Belgium.
Assistant of Prof. Ch. MASSONNET (1968-1972), Department of Strength of Materials and
Stability of Constructions, University of Lige, Belgium.
Assistant Professor in the same Department (1972-1976).
Lecturer (Chef des Travaux) in the same Department (1976-1982).
Associate Professor, University of Kinshasa, Zaire (1982-1986).
"Agrg de Facult" (1984-1991), Department of Mechanics of Materials, Stability of
Constructions and Mechanics of Structures, Univ. of LIEGE.
Associate Professor, University of Lige, (charg de cours, 1984-1991). Depart. of Strength of
Materials and Stability of Constructions, University of Lige.
Professor and Head of the Section LTAS-Fracture Mechanics of Solids (1991-2006).
Professor Emeritus (2006-20**), University of Lige.
Key qualifications
7.1
Regular academic activities
In charge of the following lectures:

"Complments de mcanique des solides (Advanced Solid Mechanics) (30 h + 30 h), CSMMC
(option), EMAAE, 2MP.
Mcanique de la rupture, (Fracture Mechanics, 30 h. + 30h), 3MP, 3EMA.
Mcanique des cramiques, (Mechanic of Ceramic, 15 h + 15h), 3 EM (option)
Aspect thermodynamique des comportements des solides, (Thermodynamic Aspect of the
Behaviour of Solids, 20 h + 20 h), 3 EM (option).
Modlisation des solides par la mthode des lments aux frontires, (Boundary Element
Method in Solid Mechanics, 15 h + 15 h), 3 EM (option), 3 MP (option)
7.2
Research activities
12
7.2.1 Main topics

Computational Solid Mechanics (finite element method, boundary element method, analytical
method, mathematical programming techniques)
Non-linear Behaviour of Solids (plasticity, visco-plasticity, visco-elasticity, fatigue, damage)
Limit State Calculations (limit analysis, shakedown analysis, and second order effects)
Advanced Solid Mechanics (fracture mechanic, contact of solids, mechanic of ceramic)
Optimisation of Structures (limit design, shakedown design)
Structural Analysis (beams, plates in extension and in bending, shells)
Fracture Mechanics (linear, non-linear)
7.2.2 Main computing packages
Author of the following computing packages:
ADELEF, Analyse et Dimensionnement aux Etats Limites par la Mthode des Elments Finis Analysis and Design in Limit State by Finite Element Method.
This package may perform:
Elastic analysis: plane structures, rotational shells
Limit analysis: plane structures, rotational shells
Visco-plastic analysis: plane structures
Shakedown analysis: plane structures, rotational plates and shells
Contact analysis: plane structures
Fracture analysis: plane structures
CEPAO, Calcul Elastique-Plastique, Analyse et Optimisation - Analysis and Optimisation of
frame structures.
This package may perform:
Limit Analysis, Limit Design, Shakedown Analysis, Shakedown Design, Step-by-step ElasticPlastic Analysis, Second Order Effects, Plastic Deformation Direct Calculations, Stability
Control, Optimal Design of Reinforced Concrete Structures...
SAMCEF, Systme d'Analyse des Milieux Continus par la Mthode des Elements Finis:
Contribution to the computation in the domains of Fracture Mechanic and Delamination of
Composites.
ELSA (1991-1993, Elbows Limit and Shakedown Analysis).
BECOME (1992-1994, Boundary Element Computational Method).
7.2.3 Main joint-researches with industries
C.M.I. - contract: "Shakedown analysis of nuclear-reactor structures", (3 years, 1983-1985)
S.E.P. - SAMTECH - contract: "3-D fracture Mechanic by Finite Element Method" (1 year, 19871988)
CARAT-DUCHATELET - contract: Design of a lengthened and reinforced Mercedes" (1988).
AEROSPATIALE (Marignane) - SAMTECH - contract: "Equivalent Domain Integral of VCE
Method (1989).
CEE-MONS - contract: Numerical Analysis of Pressured Elbows" (1989).
AEROSPATIALE (Marignane) - SAMTECH - contract: "Collage Algorithm", (1990).
AEROSPATIALE (Toulouse) - SAMTECH contract: "Calculation of the integral J by the method
of Equivalent Domain Integral in SAMCEF under nonlinear conditions". (1991).
13
WALLOON REGION - Program Multimaterials: "Development and assessment of the damage

phenomenon of composites" (1991-1995).
CEE-AG2 Contract N ETNUT-0110-B (1993) (in collaboration with Faculty of Technologies
of Mons, Belgium and the British society NNC): "Shakedown analysis of pipes and elbows" (19931995).
CEE-AG2 Contract N 0235 UK, (major contactor: NNC): "Application of shakedown methods to
piping" (1994-1995).
FRAMATOME (Paris) - Contract "Analyse limite des coudes avec le programme ELSA: Limit
analysis of elbows (1994-1995).
CEE- COST 512- EUROPEAN CONCERTED ACTION ON MODELLING IN MATERIAL
SCIENCE AND PROCESSING (MMSP) MATERIALS. Convention N 2760, sponsored by
Walloon Region. Project: "Modelling and simulation of damage growth and crack extension in
metallic and composites structures subjected to fatigue with the presence of residual stresses",
(1995-1999).
CEE, BRITE&ERAM: Project N BE97-4547: LISA: FEM Based limit and shakedown analysis
for design and integrity assessment in European industry, (1988-2002).
WALLOON REGION - Project COMDY, Convention 114696, "Delamination and damage of
laminated composites under dynamic loadings" (2002-2005).
7.3
General scientific and academic activities
7.3.1 Manager of scientific meetings

-Manager and chairman of the Lige-Workshop on theme "Applications of the Mathematical
Programming Method to Structural Analysis and Design", Lige, June 1, 1982 (40 participants
from European countries).
-Manager of the Milan-Lige day on theme "Boundary Finite Element Method", Lige, January 26,
1988.
-Manager of the contact group meeting on the theme: "Applications of the mathematical
programming method in optimisation and analysis of structures", Lige, May 6, 1988.
-Manager and chairman of the "One day Seminar on Boundary Element Method", Lige December
2 1991", organised by FNRS-BEM Contact Group with the participation of Ch. Massonnet, C.A.
Brebia, P. Pereira, M. Gradin, P. Beckers, M. Hogge, J. Etienne...
-Member of the international scientific meeting "Computational Methods in Contact Mechanics,
organized by Wessex Institute of Technology, Southampton, UK, July 1993.
7.3.2 Promoter of European research groups
-President of the FNRS contact group "Boundary Element Method".
-Promoter and Secretary of the FNRS contact group "Mathematical Programming Techniques and
their Applications in Structural Analysis and Design".
-President of the FNRS contact group "Fracture and Damage Mechanics" (1992).
-Promoter and co-ordinator in Lige of the 3rd cycle Advanced Studies ERASMUS teaching
program on "Computational Mechanics of Solids and Structures", with the participation of 16
European Universities or High Schools.
14
-Administrator of the COMETT program named "European Network for Computer Science Applied
to Mechanics in Industry" (CSAMI, F. Muller, Place Jussieu, 4, University of Paris IV, France).
-Promoter and co-ordinator in Lige of the TEMPUS program named "Computer aide design of
structures made of material with complex behaviour" with the participation of 20 European
universities or High Schools including Warsaw, Cracow, Poznan, Budapest, Bratislava, Prague,
Miskolc, Nis, Ljubljana...
-Chairman of working group IV (Simulation material properties), COST 512 of E.C.: Modelling of
Properties COST 512 MMSP (95-96-97).
International activities
8.1
Member of cooperation agency
-Member of the Expert Committee (Aupelf&Uref, French Speaking Agency for Universities) for
the revival of "Institute of Technology of Phnom-Penh, Cambodia (1992)
-Member of the Expert Committee (Aupelf&Uref, French Speaking Agency for Universities) for
the creation of the Institute of Computer Sciences, IFI in Ha Noi (1992).
-Member of the consortium "Engineering Sciences" (Aupelf&Uref, French Speaking Agency for
Universities) for the engineering training co-operation with Indochina (1994).
-Director of the Training Centre: "European Master in Mechanics of Constructions, EMMC", Ho
Chi Minh University of Technology. (1995-2008).
-Director of the "Modelling and Computation in Mechanics of Continuum, MCMC" Centre, Hanoi
University of Technology (1998-2008).
-Vice-President and Belgian Co-Coordinator of the Management Committee of the project
"Coopration en matire de formation de Doctorat en Sciences Appliques en partenariat,
Financement national du Vietnam", (2003-2010).
8.2
Promoter and co-ordinator of North-South projects
-Promoter and Co-ordinator of a North-South Multilateral Project, 1993 (Lige - Compige Ho

Chi Minh-City) sponsored by AUPELF: "l'aide aux institutions universitaires vietnamiennes en vue
de la formation de chercheurs dans le domaine du calcul des structures via l'ordinateurs et de la
mcanique des solides".
-Promoter and Co-ordinator of the project (1993-1995): "Training of Vietnamese researchers in
computational mechanics" with the participation of the Technical Faculty of Mons, Polytechnic
University, Institute of Mechanics, University of Ho Chi Minh city. This project is supported by
French speaking community of Belgium via the CGRI.
-Promoter and Co-ordinator of the project (1995-2008): "European Master in Mechanics of
Constructions, EMMC" with the participation of the Technical Faculty of Mons, Free University of
Brussels, Catholic University of Louvain and University of Technology of Ho Chi Minh City .
15
-Promoter and Co-ordinator of the project (1998-2006):"Master in Computation and Modeling of

Continuum, MCMC" with the participation of the Technical Faculty of Mons, Free University of
Brussels, Catholic University of Louvain and Institute of Technology, Institute of Civil
Engineering, National University and Institute of Mechanics of Hanoi
-Promoter and chairman of the organising committee of the International Colloquium in
Mechanics of Solids, Fluids, Structures and Interactions", Nha Trang, Vietnam, August 14 -18,
2000.
-Promoter and co-ordinator of the project of transfer of know-how to National University of Ho Chi
Minh City (University of Technology), in the domain computer aided engineering design by the
creation of a complete training programme: "European Master in Modeling of Continuum", 20012003, in collaboration with the Universities of Delft (Holland) and Montpellier II (France).
-Promoter and co-ordinator of the project of transfer of know-how Institute of Technology of
Hanoi, Vietnam in the domain computer aided engineering design by the creation of a complete
training programme: "European Master in Modeling and Design of Engineering Sciences, EMMD,
2001-2003, in collaboration with the Universities of Lulea (Sweden) and Marseille II (France).
-Promoter and co-ordinator of the project Feasibility study on and design of an inter-locking
blocks for the protection again erosion in the Binh Thuan Province, ANTIERO, 2001-2007, in
collaboration with the Water Resource University of Hanoi, Vietnam.
-Promoter and co-ordinator of the project "Coopration en matire de formation de Doctorat en
Sciences Appliques en partenariat", (National funding from Vietnamese government), in
collaboration with the University of Civil Engineering, Hanoi, 2003-2010.
8.3
Lectures and seminars
-During the period of August 3 to September 23, 1977. Invited by the Committee of Sciences and
Techniques of Vietnam and with aids of the CECODEL the following courses are given by
Nguyen-Dang Hung in Hanoi (Hydraulic Institute): - Implementations of the Finite Element Method
in Structural calculations (30 h + 15 h).
-During the period of May from 2 to 21.1991 Nguyen-Dang Hung is invited by the Polytechnic
University of Ho Chi Minh City (under contract with French speaking University Association
(AUPELF) to give a cycle of lectures on the topic: "Advanced Computational Solid Mechanics".
-23rd September 1977, Hydraulic Institute of Hanoi (invited by State Committee of Sciences and
Technologies of Vietnam). The topic: Limit analysis of shells of revolution by F.E.M. and
mathematical programming.-2nd February 1989 at National Institute of Polytechnic Engineering,
Ho Chi Minh City, Vietnam. Topic: "The Development of Computational Mechanics in the
University of Liege".
-15th February 1990 at National Institute of Polytechnic Engineering, Ho Chi Minh City, Vietnam.
Topic: "The Variational Principles and the Finite Element Method".
-During the period of July 15 to August 15.1991 Nguyen-Dang Hung is invited by the two above
Vietnamese universities to continue a cycle of lectures on the topic: "Introduction to finite element
method".
16
-11th February 1992 at National Centre for Scientific Research of Vietnam, Institute of Mechanics
(N224 Doi Can, Hanoi, Profs Nguyen-Van Dao & Nguyen-Van Diep). Topic: "On the
computational research in the domain of fracture mechanics".
-17th February 1992 at Institute of Mechanics of Ho Chi Minh City, Vietnam (Prof. Nguyen-Xuan
Hung). Topic: "A synthetic vision of fracture mechanics"
-18th February 1992 at Institute of Mechanics of Ho Chi Minh City, Vietnam (Prof. Nguyen-Xuan
Hung). Topic: "The use of linear programming method in CEPAO, a compact package for plastic
analysis and design of frame structures".
-During the period 2 to 23 January 1993, Nguyen-Dang Hung is invited by the Institute of
Mechanics and University of Technology of Ho Chi Minh City, Vietnam, to give lessons on the
topic: "Analyse et dimensionnement limite des structures dans les conditions de chargements
complexes".
-25th June 1993, at Technical University of Budapest, Department of Metallic Structures (Prof.
Ivanyi Miklos, Final meeting of TEMPUS project). Topic (in French): "Dimensionnement des
structures barres dans les conditions de chargement fixes et variables".
-19th September 1994, at University of Aix-Marseille II, Department UNIMECA, Technople de
Chteau-Gombert (Prof. Daniel Dufresne). Topic (in French) "Les nouveaux dveloppements au
service LTAS-Mcanique de la rupture des solides concernant les recherches sur les mthodes
numriques en mcanique des solides".
-15th March 1995, AG2- Meeting (European Commission, Environment, nuclear safety and civil
protection) Brussels. Topic: "Comparison and testing of numerical techniques for shakedown
analysis of nuclear plant components, Research progress of the project ETNUT-0110-B (1993).
-14th February 1996, Laboratoire de Mcanique et Acoustique, CNRS, Marseille, (Prof. P.
Suquet and Dr. P. Chabrand). Topic (in French): Analyse limite des tuyaux et des coudes par la
mthode des lments finis et la programmation mathmatique".
-15th May 1996, at AEROSPATIALE Avions building (Toulouse-Blagnac), Department of
Structures (B. Laschaud). Topic (in French): "Les rcents progrs de recherche sur la
modlisation des composites au LTAS, Lige".
-31st May 1996, at FRAMATOME building, Paris-la-Dfense, Dpartement Bloc Racteur et
Boucles Primaires (M. Ph. Gilles). Topic (in French): Analyse limite des coudes utiles sans les
racteurs nuclaires".
-23rd February 2000, Midterm meeting of Brite&Eram LISA project, Brussels (Profs. M. Staat
and E.F. Hicken). Topic: Developments of algorithms for shakedown analysis.
-17th May 2000, Laboratoire de Gnie Civil et Mcanique, LGCM, Universit de Montpellier II,
(Prof. Ch. Licht). Topic (in French): "Analyse limite et calcul dadaptation plastique des coudes et
des coques."
-19th November 2000, at University of New South of Wales, Department of Civil Engineering,
(Prof. Francis Tin-Loi). Topic: About equilibrium finite elements.
17
-7th December 2000, at Institute of Architecture of Hanoi (Prof. Nguyen Tram). Topic:
"Introduction to fracture mechanics".
-10th January 2001, at University of Lige, Journey of CEIM (Prof. Ren Maquoi). Topic (in
French): Le Service LTAS-Mcanique de la Rupture et les recherches dans le domaine de
lanalyse limite et du calcul dadaptation plastique des structures mtalliques.
-August 2002, at University of Water Resource of Hanoi (Prof. Nguyen-Van Mao). Topic:
"Fracture mechanics and applications on dam constructions".
-November 2002, at University of Lige (Chaired by Prof. Willy Legros, Rector of University of
Lige and Prof. Pham-Sy Tien, Head of the Department of Postgraduate Training, Ministry of
Education and Training of Vietnam). Topic: "Bilan de dix ans la coopration Universit de Lige Universits Vietnamiennes dans la formation doctorale".
-June 2004, Tuesday 1st, at "Stakeholder meeting" organised by EC Delegation in Hanoi. Topic:
"The programme EU-EMMD and the Inter-University cooperation with a consortium of 5
Universities of Technology in Hanoi".
-June 2004, Thursday 3rd, at "Stakeholder meeting" organised by EC Delegation in Ho Chi Minh
City : Topic : "The programme EU-EMMC and the Inter-University cooperation with the Ho Chi
Minh University of Technology".
-Visiting lecture from 20/12/2003 to 15/2/2004, Ho Chi Minh University of Technology and Hanoi
University of Technology, Vietnam.
-Visiting lecture from 28/12/2004 to 21/2/2005, Ho Chi Minh University of Technology and Hanoi
University of Technology, Vietnam.
-March 2005, Friday 11 at "NidaGroup and Henri Tudor Research Centre organise by Dr Salim
Belouettar, CRP Henri Tudor, and Luxembourg: Topic: "Direct and indirect metis finite element
method for edge delamination in laminates".
Professional experience record in other countries
-During academic years, (1982-1986) on contracts with A.G.C.D. (Belgian Ministry of Cooperation), Nguyen-Dang Hung is a permanent professor at Polytechnic Faculty, University of
Kinshasa Zaire. He gives the following courses:
Applied Static (30 h + 30 h, First year of Engineering study)
Strength of material and elasticity (45 h + 45 h, 2nd year)
Stability (45 h + 45 h, 3rd year)
Metallic constructions (60 h + 60 h, 4th year)
Applied Mechanics (30 h + 30 h, 4th year)
Advanced studies of stability (30 h + 30 h, 5th year)
-During the period from 15th to 19th July 1985, Nguyen-Dang Hung is invited by the International
Centre of Mechanical Sciences, Undine, Italy to give four lectures in the topic of "Applications of
mathematical Programming Technique to Plastic analysis and Design".
18
-During the period from 26th October to 26 January, Nguyen-Dang Hung is invited by l'A.G.C.D.
(Belgian Ministry of Co-operation) to give the following lectures in Polytechnic Faculty, University
of Kinshasa, Congo:
Elasticity (15 h + 15 h)
Stability (45 h + 45 h)
Applied Mechanics (30 h + 30 h)
Advanced Studies of Stability (30 h + 30 h)
-During the period from 15th to 17th November 1988, Nguyen-Dang Hung is invited by the Society
SAMTECH (Lige) and Aerospatiale-Division Hlicoptres (Marignane, France) to give a course
on the topic "Fracture Mechanics and the Finite Element Approaches".
-During the period from 17th to 20th January 1989, Nguyen-Dang Hung is invited by the Society
SAMTECH (Lige) and Aerospatiale, Service de calcul des structures, (Toulouse, France) to give a
course about "Finite Element Method in Fracture Mechanics".
-During the period from 19th to 21st June, 1990, Nguyen-Dang Hung is invited by the society
SAMTECH (Lige) to give a course on the topic "The SAMCEF Package and the Computational
Fracture Mechanics".
-During the period from October from 8 to 12.1991, Nguyen-Dang Hung is invited by the Technical
University of POZNAN, Institute of technological constructions, Poland to give a cycle of lecture
on the topic: "Limit analysis and limit design of frame structures".
-During the period of October from 14 to 18.1991, Nguyen-Dang Hung is invited by the Technical
University of KRAKOW, Institute of Civil Engineering, Poland to give a cycle of lecture on the
topic: "Computation of limit and shakedown design of frame structures".
-During the period from May 30 to June 4. 1992, Nguyen-Dang Hung is invited by the Technical
University of Budapest, Hungary, Department for Steel Structures to give a cycle of lecture on the
topic: "Direct plastic limit analysis and limit design of frame structures under proportional and
variable loadings".
-During the period from 2nd to 23rd January 1993, he is invited by Institute of Applied Mechanics,
and Institute of Technology of Ho Chi Minh City, Vietnam, to give a cycle lecture on (in
Vietnamese): "Limit Analysis and design of structures under complex loadings".
-During the period from 17 to 21 June 1993, Nguyen-Dang Hung is invited by the "Istituto
Superior Tecnico, Departemento de Engenharia Civil" (Prof. Teixeira de Freitas), Portugal, to give
lessons on the topic: "The monotonous convergence of the pure and metis finite element models".
-During the period from 17th to 21st June 1993, he is invited by Istituto Superior Tecnico,
Departemento de Engenharia Civil (Prof. Teixeira de Freitas), Portugal, to give two lessons on the
topic: "The monotony of the convergence of the pure and metis finite element" and On the
double singularity of a metis finite element useful in crack problems.
-During the period from 24th to 28th January 1994, he is invited by Institute of Applied Mechanics,
and Institute of Technology of Ho Chi Minh city, Vietnam, to give a cycle lecture on "Advanced
mechanics of solids".
19
-During the period from 11th to 15th July 1994, he is invited by Institute of Applied Mechanics, and
Institute of Technology of Ho Chi Minh City, Vietnam, to give a cycle lecture on "Introduction to
boundary elements.
-During the period from 8th February to 8th March the from the first to 30th July 1996, he is invited
by University of Aix-Marseille II, Department UNIMECA (Prof. D. Dufresne) to give a cycle of
lectures on (in French): "La fissuration et le calcul de tnacit par les mthodes numriques dans le
SAMCEF".
-During the period from 21st to 22nd October 1996, he is invited by Prof. Tran-Ich Thinh, Institute
of Technology of Hanoi to give two seminars on: Introduction to fracture mechanics and The
finite element method and the fracture mechanics.
-During the period from 24th to 25th October 1996, he is invited by Prof. Phan- Ky Phung, Rector
of the University of Da Nang, Vietnam to give two seminars on: Introduction to fracture
mechanics and The finite element method and the fracture mechanics.
-During the period of 24th February to first March 1997, he is invited by University of AixMarseille II, Department UNIMECA (Prof. D. Dufresne) to give a cycle of lectures on (in French):
"Introduction de la mthode des lments aux frontires"
-From September 1995 to December 2000, he is invited by Institute of Technology of Ho Chi Minh
City under the auspice of the EMMC programme to give ten lectures on the topics (in Vietnamese):
Modeling in solids and structures under non linear conditions, Fracture
mechanics,Boundary element method.
-During the period from 10th April to 16th June 2000, he is invited by University Montpellier II,
Laboratory LGCM, and (Profs Ren Motro et Christian Licht) to give cycle of lectures on (in
French): "Comportements non linaires des solides".
-During the period of 9th to 14th October 2000, he is invited by University of New South of
Wales (Prof. Francis Tin-Loi) to give a seminar on the theme: "On the equilibrium elements and
their role in structural mechanics".
10
Publications
Books and courses: 23,

Papers (individual or collective), articles in conference proceedings and scientific reports: 203,
several general papers, travel reports and interviews
Poem individual or collective collections: 12 (in Vietnamese).
10.1 Books and courses

1. "Non-Linear Behaviour of Materials", (in French) 140 pages, Department of Mechanics of
Materials, Stability of Constructions and Mechanics of Structures, University of Lige, (1980).
2. Plasticity, viscoelasticity, thermoelasticity (in French and in collaboration with Ch.
MASSONNET), Lecture edited by the "Laboratoire de Mcanique des Matriaux et de Stabilit des
Constructions", (1981), 340 pages.
3. "Applied Static" (in French) - Edition PUZ, Kinshasa, Zaire (1983), 220 pages
20
4. "Elements of Elasticity" (in French) - Edition PUZ, Kinshasa, Zaire (1983), 120 pages
5. "Strength of Materials" (in French) Tome I, Department of Constructions, Polytechnic Faculty,
170 pages, 1983.
6. "Strength of Materials" (in French) Tome II, Department of Constructions, Polytechnic Faculty,
185 pages, 1983;
7. Applications of the mathematical programming method to structural analysis and design,
Engineering Structures, Butterworth, vol. 6, n 1, January 1984.
8. "On the Plasticity theory and the plastic limit calculations of Structures" (in French), Publication
n 98, Faculty of Applied Sciences, University of Lige (1985), 480 pages.
9. "A course of Elasticity" (in French), Division of "Numerical Modeling in Mechanics and
Structures", Department of Mechanics, Technology University of Compigne, France (1986), 204
pages.
10. "Complements of Mechanics of Solids: plasticity, viscoelasticity, Fracture Mechanics" (in
French), Lecture edited by the Laboratoire de Mcanique des Matriaux et de Stabilit des
Constructions, University of Lige, (1989), 340 pages.
11. "Fracture Mechanics: Numerical approaches" (in French), edited by SAMTECH (1989), 250
pages.
12. "Introduction to Boundary Element Method", a course in French, LTAS Fracture Mechanics,
1990, 120 pages.
13. "A course of Fracture Mechanics" (in French), edited by LTAS-Mcanique de la Rupture des
Solides, University of Lige, 1990, 350 pages.
14. "Inelastic Analysis of Structures under Variable Loads, Theory and Engineering Applications",
published by Kluwer Academic Publishers, 2000, the chapter: Direct finite element kinematical
approaches in limit and shakedown analysis of shells and elbows (in collaboration with YAN AM), pp. 233-254.
15. "Introduction to Solid Mechanics", (in French and in Vietnamese), edited by Polytechnic
University of Ho Chi Minh City, Vietnam, May 1991, 250 pages.
16. "Advanced fracture mechanics", syllabus in French used during stay I in University of Marseille
II (UNIMECA), 1996, 26 pages.
17. "Fracture mechanics and finite element method", syllabus in French used during stay I in
University of Marseille II (UNIMECA), 1996, 30 pages.
18. "Introduction to the boundary element method", syllabus in French used during stay I in
University of Marseille II (UNIMECA), 1996, 60 pages.
19. "Non linear mechanics of solids, syllabus in French used during stay in University of
Montpellier II, 2000, 250 pages.
21
20. "Dictionnaire de l'architecture et du btiment, franais-Vietnamien". (in collaboration with

Doan-Nhu Kim and Doan-Dinh Kien), book prepared under the sponsoring of l'Agence
Universitaire de la Francophonie" Van Hoa publisher, 1999.
21. "Proceedings of the International colloquium in Mechanics of solids, fluids, structures and
interaction", August 14-18 Nha Trang, edited by LTAS-University de Lige (co-editors: F. Ronday
and Nguyen-Van Diep), 886 pages, 2000.
22. Limit Analysis of Defects, (in collaboration with Staat, M.; Heitzer, M.; Yan, A.M.; Vu-Duc
Khoi; Voldoire, F. and Lahousse, ZENTRALINSTITUT FR ANGEWANDTE MATHEMATIK
(ZAM), Berichte des Forschungszentrums Jlich 3746, Jl-3746, 2000.
23. "Complementary, duality and symmetry in nonlinear mechanics", proceedings of Shanghai
IUTAM Symposium, Edited by David Gao, Kluwer Ac. Publ., Chapter: Duality in kinematic
approaches of Limit and Shakedown Analysis of Structures. (in collaboration with Yan Ai Min and
Vu-Duc Khoi), pp 128-148, 2004.
10.2 Journal papers (chronological order)

1. Displacement and equilibrium methods in matrix analysis of trapezoidal structures. Collection
des Publications de la Facult des Sciences Appliques, n 21, 1970.
2. Duality in the analysis of shells by the finite element method. International Journal of Solids and
Structures, March 1971, vol. 7, issue 3 pp. 281-299.
3. Limit analysis of framed structures. (in French), Construction mtallique, magazine edited by
CTICM, Nr 4, 1972.
4. Limit analysis of rigid-plastic structure by a quasi-direct method (in French) presented at
Sminaire de Plasticit et Viscolasticit, Ecole Polytechnique de Paris, 27-29 September 1972.
Published in Sciences et Techniques de l'Armement. Mmorial de l'Artillerie Franaise, 2nd section,
1973.
5. Application of the plasticity in the metal working problems (in French). Revue de la Socit
Royale des Ingnieurs et des Industriels, Nr 1/2; January-February 1974, pp. 14-32.
6. Introduction to Finite element method (in French). Annales des Travaux Publics de Belgique, N
1, 1973/1974.
7. Quasi-static approach of limit analysis problems (in French). Buletinul Scientific al Institutului
de Constructi, Bucharest, Anul 17, 1974.
8. Displacement potentials for membranes on stretching and stress-function potentials for plate on
bending. Ingenieur-Archiv 45 (1976), pp. 41-53.
9. Theoretical formulation and numerical efficiency of a bending finite element. (in French and in
collaboration with P. Detroux), Collection des Publications de la Facult des Sciences Appliques,
University of Lige, N5, 1976, pp. 357-376.
10. Direct Limit Analysis via Rigid-Plastic Finite Element, Computer in Applied Mechanics and
Engineering, vol. 8, Issue 1, April-May 1976, pp. 81-116.
22
11. A finite element formulation for shakedown problems using a yield criterion of the mean. (in
collaboration with Jan Konig). Computer in Applied Mechanics and Engineering, vol. 8, Issue 2,
June 1976, pp. 179-192.
12. Finite element discretization and the numerical method in applied mathematics, (in French),
Annales des Travaux Publics de Belgique, nr 5, 1976, and pp. 357-376.
13. Two field variational principles for Saint-Venant-von Mises material (in French) Bulletin de
l'Acadmie Polonaise des Sciences (Warsaw). Srie des sciences techniques, vol. 24, n 7-8, pp.
325-331, 1976.
14. Plastic collapse of shells of revolution by non-linear programming and finite element
techniques. Journal of Engineering Mechanics Division, ASCE, vol. 104, Nr 3, June 1978, pp. 707712.
15. Probabilistic plastic analysis of framed structures. (in collaboration with D. Frangopol),
Construction mtallique, magazine edited by CTICM, Paris, Nr 1, 1977, pp. 19-43.
16. The numerical performance of a plate bending hybrid finite element. (in French and in
collaboration with P. Desir). Collection des Publications de la Facult des Sciences Appliques,
University of Lige, Nr 66, 1977, pp. 65-123.
17. Quasi-lower bounds and upper bounds of the plastic ultimate pressure in rotationally shells by
finite element methods and mathematical programming techniques (in French, in collaboration with
Marc Trapletti and Daniel Ransart). International Journal of Non-linear Mechanics, edited by Prof.
A. Nash, vol. 13, Issue 2, pp. 79-102, 1978.
18. Plastic strength of a reinforced double T beam subjected to bending and shearing (in French
and in collaboration with Ch. Massonnet and J. Janss), Centre de Recherches Scientifiques et
Techniques de l'Industrie des Fabrications Mtalliques, MT 126, Brussels, Belgium, September
1978.
19. New numerical experiences in shell limit analysis of revolution (in French). (in collaboration
with D. Frangopol). Science and Technique Journal of Rumania-Series of Applied Mechanics, vol.
23, Nr 4, 1978.
20. Shakedown analysis by displacement method and equilibrium finite element (in collaboration
with L. Palgen). Transactions of the CSME, vol. 6, Issue 1, pp. 32-39, 1980-81.
21. Frictionless contact of elastic bodies by finite element method and mathematical programming
technique. (in collaboration with Gry de Saxc). Computers and Structures, vol. 11, Issue 1-2, pp.
55-67, 1980.
22. Contact of elastic solids by finite element and mathematical programming methods. (in French
and in collaboration with Gry de Saxc). Archiwum Inzynierii Ladowej, Polish Academy of
Sciences, Warsaw, 26, Nr 1, pp. 95-124, 1980.
23. on the variational principles in plasticity (in French). Collection des Publications de la Facult
des Sciences Appliques, Lige, Belgium, Nr 84, 1980, pp. 59-76.
24. Finite element equilibrium analysis of creep using the mean value of the equivalent shear
modulus. Meccanica (AIMETA, Italy), Nr 4, December 1980, vol. 15, pp. 234-245.
23
25. Analysis and optimisation of framed structures in the stable conditions. (in French).
Construction mtallique (FRANCE), Nr 3, 1981, pp. 15-38.
26. Shakedown analysis by finite element method and linear programming techniques. Journal de
mcanique applique, Paris, France, vol. 2, no 4, pp. 587-599, 1983.
27. Dual analysis of frictionless contact problems by displacement and equilibrium finite element
models (in collaboration with Gry de Saxc). Engineering Structures, Butterworth, Special number
edited by Nguyen-Dang Hung, vol. 6, Nr 1, January (1984), pp. 26-32.
28. CEPAO, an Automatic Program for Plastic Analysis and Optimisation of Frame Structures
Under Stability Conditions, Engineering Structures, Special Number, edited by Nguyen-Dang
Hung, vol. 6, Nr 1, January (1984), pp. 33-51.
29. Applications of the mathematical programming method to structural analysis and design.
Proceedings of the Workshop of Lige, 1 June 1982, Engineering Structures, Butterworth, vol. 6, Nr
1, January 1984.
30. Aspect of Analysis and Optimisation of Structures under Proportional and Variable Loadings.
Engineering Optimisation, vol. 7, Nr 1, pp. 35-77 (1983).
31. An Automatic Method for the Displacement Elastic-Plastic Calculation at Plastic Collapse
State-Method Using Linear Programming Technique (in French). Construction mtallique, Nr 1,
1983, pp. 27-36.
32. A Direct Method for Information Analysis at Plastic Collapse State (in French). Annales de la
Facult Polytechnique, University of Kinshasa, Zaire, vol. III, IV, V, June 83/84/85, pp. 37-54.
33. Further Aspects of Analysis and Optimisation of Structures Under Proportional and Variable
Loading. Annales de la Facult Polytechnique, University of Kinshasa, Zaire, vol. III, IV, V, June
83/84/85, pp. 1-6.
34. Dual analysis of frictionless problems by displacement and equilibrium finite elements,
Engineering Structures, (in collaboration with Gry de Saxc), Volume 6, Issue 1 , January 1984,
Pages 26-32.
35. CEPAOan automatic program for rigid-plastic and elastic-plastic analysis and optimization
of frame structures, Engineering Structures , Volume 6, Issue 1 , January 1984, pp 33-51.
36. The geometric nature of plasticity laws (in collaboration with Gry de Saxc), Engineering
Fracture Mechanics, Volume 21, Issue 4 , 1985, Pages 781-798
37. Introduction la mcanique de la rupture (in French). Annales des Travaux Publics de
Belgique, N 4, 26 pages (1988).
38. Modified Formula for three-dimensional Direct Stress-Intensity Factor Computation (in
collaboration with Gry de Saxc). Communication. in Appl. Num. Meth.", Vol. 6, pp 223-230
(1989).
39. The computational of 2-D stress intensity factors using hybrid-mongrel displacement finite
element. (in collaboration with Gry de Saxc and Kang Chi Hang), Engineering Fracture
Mechanics Volume 38, Issues 2-3 , 1991, pp 197-205.
24
40. Dual boundary element analysis of the multiple-cracked body. (in collaboration with Yan Ai
Min and Nicolas Mos), book edited by C.A. Brebbia - A.S. Kassab, p. 235-242, 1994.
41. Stress intensity factors and crack extension in a cracked pressurised cylinder. (in collaboration
with Yan Ai Min), Engineering Failure Analysis, Vol. 1, Issue 4, pp 307-315, December 1994.
42. Elastic analysis of the multiple-cracked structure by DBEM. (in collaboration with Yan Ai
Min). Compte rendu du 3me Congrs national belge de Mcanique thorique et appliqu. Lige,
May, pp 1-4, 1994.
43. Multiple-cracked fatigue crack growth by BEM, (in collaboration with Yan Ai Min),
Computational Mechanics, Vol. 16, No. 5, pp273-280, (1995).
44. Simulation du dlaminage des stratifis par des lments dinterface (in collaboration with BuiQuoc Viet). Journey of FNRS, Louvain-la-Neuve, March 1997.
45. Analyse limite des structures fissures (in collaboration with Yan Ai Min). Journey of FNRS,
Louvain-la-Neuve, March 1997.
46. Limit analysis and shakedown computation of shells and elbows (in collaboration with Borlee
B). Journey of FNRS, Louvain-la-Neuve, March 1997.
47. An enhanced elbow element application in limit state analysis. (in collaboration with Yan Ai
Min and Jospin R.J). Int. Journal of Num. Method. Engng. Vol.46, 409-431, 1999.
48. Practical estimation of the plastic collapse limit of curved pipe subjected to complex loading (in
collaboration with Yan Ai Min and Gilles Ph). Structural Engineering and Mechanics, pages 241438, Vol.8, N4, 1999.
49. Limit analysis of cracked structures by mathematical programming and finite element technique
(in collaboration with Yan A-M). Computational Mechanics, Springer-Verlag, pages 319-333, Vol.
23, 1999.
50. Imperfect interlaminar interfaces in laminated composites: bending buckling and transient
response (in collaboration with Bui-Quoc Viet and E. Marchal) Comp Sci Techol, 1999; 59, pp:
2269-77.
51. Imperfect interlaminar interfaces in laminated composites: interlaminar stresses and strainenergy release rates, (in collaboration with Bui-Quoc Viet and E. Marchal), Composites Science
and Technology, Volume 60, Issue 1 , 1 January 2000, pp 131-143.
52. Direct finite element kinematical approaches in limit and shakedown analysis of shells and
elbows (in collaboration with Yan Ai Min), pp. 233-254, Inelastic Analysis of Structures under
Variable Loads, Theory and Engineering Applications, published by Kluwer Academic Publishers,
2000.
53. Imperfect interlaminar interfaces in laminated composites: delamination with the R-curve
effect, (in collaboration with Bui-Quoc Viet and E. Marchal), Composites Science and
Technology Volume 60, Issue 14 , November 2000, pp 2619-2630;
25
54. A finite-element model of mixed-mode delamination in laminated composites with an R-curve

effect (in collaboration with Yan Ai Min and E. Marchal). Composites Science and Technology 61,
2001, pp 1413-1427.
55. Kinematical shakedown analysis with temperature-dependant yield stress (in collaboration with
Yan Ai Min) Int. J. Num. Meth. Engng., 2001; 50:pp1145-1168.
56. A truly meshless Galerkin method based on moving least squares quadrature (in collaboration
with Duflot Marc). Communication in International Journal for Numerical Methods in Engineering,
N18, pp 441-449 (2002)
57. Dual analysis by a meshless method (in collaboration with Duflot Marc). Communications in
Numerical Methods in Engineering, N18, pp 621-631 (2002).
58. Analysis of cracked plate bending by using a metis element. (in Vietnamese and in collaboration
with Tran-Thanh Ngoc). Revue Construction, Vol. 7, 2002, p 22-25.
59. Analysis of cracked plate membrane problem using a metis element model (in collaboration with
Tran-Thanh Ngoc). Vietnam Journal of Mechanics, vol. 24, 2002, N4, pp 249-256.
60. Kinematical formulation of limit and shakedown analysis (in collaboration with Yan A. M., VuDuc Khoi), NIC Series Volume 15: Numerical Methods for Limit and Shakedown Analysis, John
von Neumann Institute for Computing, Jlich, Edited by Manfred Staat and Michael Heitzer, 2003,
pp 85-103.
61. A dual form for discretized kinematical formulation in shakedown analysis. (in collaboration
with Vu-Duc Khoi and Yan Ai Min). Int. Journal of Solids and Structures, Vol. 41, Issue 1 pp 267277, Pergamon, 2003.
62. On the construction of a protection structure against the coastal erosion of Ham Tien Mui Ne,
Province of Binh Thuan, Vietnam (in Vietnamese, collaboration with Nguyen-Van Mao, Water
Resources University of Hanoi). Journal of Resources & Environment Engineering Nr 1, pp 25-36,
June/2003.
63. Analysis of cracked plates and shells using "metis" finite element model. (in collaboration with
Tran-Thanh Ngoc). Finite Elements in Analysis & Design (Elsevier), Volume 40, Issue 8, pp. 855878, May 2004.
64. A primal-dual algorithm for shakedown analysis of structures (in collaboration with Vu-Duc
Khoi and Yan Ai Min). Computer Methods in Applied Mechanics and Engineering (Elsevier), Vol.
193/42-44 pp. 4663-4674, May 2004.
65. Fatigue crack growth analysis by an enriched meshless method (in collaboration with Marc
Duflot), Journal of Computational and Applied Mathematics , Volume 168, Issues 1-2 , 1 July
2004, pp 155-164.
66. Regular and singular metis finite element models for delamination in composite laminates (in
collaboration with Nguyen-Tien Duong), Finite Elements in Analysis and Design, 42 (2006) 650659.
26
67. An extended finite element library (in collaboration with Bordas S., Nguyen-Vinh Phu, Dunant
C., and Guidoum A), International Journal for Numerical Methods in Engineering, to appear in
2006.
68. Interlaninar stresses and delamination of composite laminates under bending and extension (in
collaboration with Nguyen-Tien Duong), Structural Engineering & Mechanics, to be appeared in
2006.
69. Computation of interlaminar stresses using Metis elements, (in collaboration with Nguyen-Tien
Duong), Technical Note in the Structural Engineering & Mechanics, to be appeared in 2006.
70. Using metis model to solve torsion problem of prismatic bar, (in collaboration with NguyenTien Duong), Vietnam Journal of Mechanics (VAST), Vol. 27, No. 3, pp. 149-157, 2005.
71. An object-oriented extended finite element module with embedded automatic mesh database, (in
collaboration with Bordas S., Nguyen-Vinh Phu, Dunant C., Bonnaz R., and Guidoum A),
International Journal for Numerical Methods in Engineering, submitted September 2005.
72. Architecture trade-offs of embedded meshers for extended finite element methods, (in
collaboration with Dunant C., Bordas S., Nguyen-Vinh Phu, and Guidoum A.), Journal Europen de
Mcanique Numrique, submitted, 2006.
73. A posteriori error estimates for the extended finite element method, (in collaboration with
Bordas S., Nguyen-Vinh Phu, Dunant C. and Guidoum A), Computer Methods in Applied
Mechanics and Engineering, in preparation, 2006.
74. Nodal integration in finite elements, (in collaboration with Nguyen-Xuan Hung, Stphane
Bordas), International Journal for Numerical Methods in Engineering, under review, 2006.
75. On the convergence of equilibrium element free Galerkin methods, (in collaboration with BuiQuoc Tinh, Duflot Marc, Bordas S.), International Journal for Numerical Methods in Engineering,
submitted.
76. Accuracy estimation for smoothed finite element method in elasticity, (in collaboration with
Nguyen-Xuan Hung and S. Bordas), in preparation, 2006.
77. Finite element method with conforming nodal integration: convergence, accuracy, performance,
(in collaboration with Nguyen-Xuan Hung and S. Bordas), in preparation, 2006.
78. On the connection between displacement finite element method via strain smoothing technique,
(in collaboration with Nguyen-Xuan Hung and S. Bordas), in preparation, 2006.
10.3 Conference proceedings (chronological order)

1. Generalization of the dual approach in the analysis of shells by the finite element method.
Presented at NATO seminar LINEC, Lisbon, Portugal, September 1971.
2. Limit analysis of shells of revolution by F.E.M. (in collaboration with D. Ransart), WCOSE
Proceeding (World Congress of IASS on enclosure structures), July 1976, Montreal, Canada, pp.
471-479.
27
3. Duality and monotony of the two-field finite elements in elastic stress calculations. (in
collaboration with P. Desir, P. Detroux and D. Frangopol). Proceedings of 6th Canadian Congress
on Applied Mechanics, Vancouver, Canada, pp. 1025-1026, 1977.
4. Application of the F.E.M. and the programming techniques in limit analysis of shells of
revolution, (in French), Colloquium of Numerical methods in Engineering Mechanics, Ecole
Centrale des Arts et Manufactures, 92290 CHATENAY-MALABRY, France.
5. A probabilistic approach for checking safety of centrally loaded steel columns. Preliminary
report, Colloquium on the stability of steel structures, Lige, Belgium, pp. 113-118, 1977.
6. Dual and direct limit analysis by F.E.M. Proceedings of the 6th Canadian Congress on Applied
Mechanics, Vancouver, Canada, pp. 981-982, 1977.
7. Reliability analysis and optimum probability based design of plastic structures. (in collaboration
with D. Frangopol and J. Rondal). Paper presented at 14th International Congress on Theoretical
and Applied Mechanics, 30 August- 4 September 1976, and Delft, NETHERLANDS. Published in
Bulletin de l'Acadmie Polonaise des Sciences, (Warsaw), srie des sciences techniques, vol. 26, Nr
5, 1978.
8. On a special class of hybrid finite elements: the "metis" (in French). Proceedings of the 1st
International Congress of GAMNI, (E. ABSI and R. GLOWINSKI), Dunod technique, 1979, pp.
53-63.
9. Implementation of the duality in the finite element analysis of shells: a mixed-metis planar shell
element. (in collaboration with P. Detroux, Ph. Falla and G. Fonder). Proceedings of World
Congress on Shell and Spatial Structures, Theme 4, pp. 4.1. -4.19. Madrid, September 1979.
10. Shakedown analysis by displacement method and equilibrium finite element. (in collaboration
with L. Palgen). Proceedings of the 5th International 1979, Div. L, Section L3, Methods 1.
11. On the monotony and the convergence of a special class of hybrid finite elements. The Metis
elements. Variational Methods in the mechanics of solids. Proceedings of the IUTAM symposium
held at Northwestern University, Evanston, Illinois, U.S.A., edited by S. Nemat Nasser, Pergamon
Press, 1980 (pp. 208-213).
12. Shakedown analysis by F.E.M. - Force method and Displacement method. (in collaboration with
L. Palgen). Proceedings of the colloquium on the topic Materials and Structures under Cyclic
loadings, Ecole Nationale des Ponts et Chausses, Paris, France, July 1979, pp. 250-264.
13. Numerical shakedown analysis of plates and shells of revolution. (in collaboration with P.
Morelle). 6th International Conference on Structural Mechanics in Reactor Technology, 17-21
August 1981, Paris, Session L10/4, Proceedings of the 3rd congress and exhibition of finite element
methods. Beverly Hills, California, 12-16 October 1981.
14. Finite element analysis of contact problems, based on the unilateral constraints formulation. (in
collaboration with Gry de Saxc and G.M.L. Gladwell). Proceedings of the International
Symposium of Structural Control. University of Waterloo, Ontario, Canada, (4 July 1979), edited
by H.H.E. LEIPHOLZ, pp. 341-373, North-Holland, 1980.
15. Efficient Solution of Plastic Collapse Analysis and Optimum Design of Steel Frame Structures
with Taking Account of Instability. (in collaboration with Ch. Massonnet). The Michael R. Horne
28
Conference (20-22 Sept. 1983) on the Instability and Plastic Collapse of Steel Structures, 98 p.
(1983).
16. On the Stress intensity Factor Calculations Using the Quarter-Point Isoparametric Element...
(in collaboration with Gry de Saxc). Transactions of the 10th Int. Conf. Struct. Mech. Reactor
Technology, Anaheim, California, August 14-18, (1989) page 203 vol. G.
17. Une Deduction Complte du Comportement Plastique partir des Considrations
thermodynamiques. Belgian National Congress of Theoretical and Applied Mechanics, Brussels,
1990.
18. Analyse Limite des Coudes par une Formulation Cinmatiquement Admissible. (in collaboration
with R. Jospin). Belgian National Congress of Theoretical and Applied Mechanics, Brussels, 1990.
19. Plastic Shakedown Analysis. (in collaboration with P. Morelle). CISM courses and LECTURES
Nr 299: Mathematical Programming Methods in Structural Plasticity edited by D.L. Smith, PringerVerlag (1990), pp 181-205.
20. Optimal Plastic Design and the Development of Practical Software. (in collaboration with P.
Morelle). CISM courses and LECTURES Nr 299: Mathematical Programming Methods in
Structural Plasticity edited by D.L. Smith, Pringer-Verlag (1990), pp 207-229.
21. Direct limit analysis of elbows by finite element and mathematical programming. (in
collaboration with R.J. Jospin and Gry de Saxc), Anisotropy and localisation of plastic
deformation, Proceed. Symp. Plasticity 91, Grenoble, France, Edited by Boehler J.P. and Khan A.S,
pp. 623-626, 1991.
22. Shakedown analysis of structures, Proceed. Int. Conf. Steel & Aluminium (Addenum),
Singapore May 1991, Elsevier Appl. Sc., Edited by S. L. Lee and N. E. Shanmungam. (1991).
23. Crack closure studies of bending plate using a mongrel singular finite element. (in collaboration
with Kang Chi Hang, Z. Q. Feng and Gry de Saxc). Fracture of engineering materials and
structures, Proceed. Edited by S.H. Teoh and K. H. Lee, Elsevier, pp 443-449 (1991).
24. A double singular finite element for cracked structures. (in collaboration with Gry de Saxc
and Kang Chi Hang). Proceeding of the "Congrs International sur l'ducation de la pratique et la
promotion des mthodes de calcul de l'ingnieur utilisant les micro-ordinateurs". Dalian, Chine,
July 1992.
25. Limit and shakedown minimum design of the plane steel frames. (in collaboration with Gry de
Saxc). Proceed. of the first European Conference on Numerical Methods in Engineering, 7th to
11th September 1992, Numerical Method in Engineering '92, Edited by Ch. Hirsch, O.C.
Zienliewicz and E. Onate, 1992, pp 773-778.
26. Thermodynamic aspects of the plasticity. Proceedings of Int. Conf. Ho Chi Minh City 27-29
December 1995, Part II, Ho Chi Minh City Mathematical Society, vol. 5, 1995, 470-496.
27. Sur quelques expriences de coopration universitaire avec le Vietnam (in French). Workshop
on Engineering Education in Developing Countries, Proceedings of the Conference of European
Schools for Advanced Engineering Education and Research (CESAER), Lige 29-30 March, 1993.
29
28. Automatic design of frame structures under variable loadings, Proceedings of the TEMPUS
Workshop and International advances Course of Budapest on Computer Aided Design of
Structures, June 24-25, (1993).
29. Crack Extension of internally pressurized multiple cracked cylinder by boundary element
method (in collaboration with Yan Ai Min). Proceedings of the Saclay International Seminar on
Structural Integrity (SISSI), Tome 2 - Posters, pp 69-74 April (1994).
30. Comparative study of virtual crack extension method and equivalent domain integral method
applied to cracked structures with thermal loadings. (in collaboration with Bui-Cong Thanh).
Compte rendu du 3me Congrs national belge de Mcanique thorique et applique, Lige. May,
pp 5-9, (1994)
31. Fatigue crack growth studies using dual boundary element method. Proceedings of the Conf.
Advances in Computational mechanics, edited by M. Papadrakakis and B.H.V. Topping, CivilComp Ltd, Edinburgh, Scotland, pp 301-310, September (1994).
32. Limit and shakedown analysis of axisymetrical shell by Mathematical programming. (in
collaboration with Bui-Cong Thanh), LTAS- Fracture Mechanics of Solids, December, 1994.
33. Numerical comparative studies of the Rice and Bui integrals and the stress intensity factors
using a double singular metis finite element. (in collaboration with Nguyen D.T.), LTAS- Fracture
Mechanics of Solids, March, 1995.
34. Effet du facteur caractristique sur le moment limite du coude soumis la flexion plane. (in
collaboration with Yan Ai Min), 2nd report, FRAMATOME-LTAS, LTAS- Fracture Mechanics of
Solids, April 1995.
35. Multiple-cracked fatigue crack growth by Boundary Element Method. (in collaboration with
Yan Ai Min). Proceeding of the Third Word Congress on Computational Mechanics, Chiba, Japan,
August 1-5, volume I, pp 839-840, 1994, published in Computational Mechanics, edited by S.N.
Atluri, 16, pp 273-280 Springer International, August, 1995.
36. Determination of singularity order and development if a metis finite element for the phenomena
of delamination in composite laminates. (in collaboration with P. Schils, Gery De Saxc, and S.
Babric). Proceeding of the International Conference on Composite Materials and Energy, Fracture
II, pp 148-155, Montral, Canada, 8-10, May, 1995.
37. Simulation of the Crack Growth of 3D Structures under Fatigue in Taking Account of Residual
Stresses, (in collaboration with M. Scibetta and B. Borlee) Proceedings of the Eleventh Biennial
European Conference on Fracture, held at ENSMA Poitiers, Futuroscope, France, September 3-6,
1996. Editor: J. Petit, Co-Editors: J. de Fouquet, G. Henaff, P. Villechaise and A. Dragon, ECF 11 Mechanisms and Mechanics of Damage and Failure,
38. Estimation Scheme for Surface Cracked Piping Under Complex Loading: Part II: Complex
Shaped Elbow Solutions (in collaboration with P. Gilles and C. Bois). Proceedings of the Eleventh
Biennial European Conference on Fracture, held at ENSMA Poitiers, Futuroscope, France,
September 3-6, 1996. Editor: J. Petit, Co-Editors: J. de Fouquet, G. Henaff, P. Villechaise and A.
Dragon
39. Alternative estimations of interlaminar stresses using generalised laminate finite element. (in
collaboration with Putz Philippe and Lambert Eric). MMSP96 general Worshop, Proceed. of
30
general COST 512 Worshop on modelling in Materials science and processing, Davos 29.9 to
2.10.1996, pp 366.
40. Risk for cracking in multi-pass weldments. Edited by L.E. Lingren, H. Runnemalm, (in
collaboration with B. Borle). MMSP96 general Worshop, Proceed. of general COST 512
workshop on modelling in Materials science and processing, Davos, October, 1996.
41. Limit analysis and shakedown computation of shells and elbows. (in collaboration with Yan Ai
Min) Proceeding of the International conference EMT97, Engineering mechanics today, Hanoi,
August 1-3, 1997.
42. A damage model for mode I delamination in composites. (in collaboration with Bui-Quoc Viet).
Proceeding of the International conference EMT97, Engineering mechanics today, Hanoi, 1-3
August 97.
43. A finite element formulation for shakedown analysis using the weighed average criterion. (in
collaboration with Bui-Cong Thanh). Proceeding of the International conference EMT97,
Engineering mechanics today, Ho Chi Minh City, 14 August, 97.
44. Plastic collapsed of curved pipe and shell by finite element method and mathematical
programming technique. (in collaboration with Yan Ai Min) IASS International Colloquium on
Computation of Shell and Spatial Structures, Taipei, Taiwan, November 5-7, 1997.
45. Internal effect on the limit load of elbows. Proceeding of International Symposium of IASS,
Singapore, 10-14 November 97.
46. Shakedown of structures by improved Koiters theorem. (in collaboration with Yan Ai Min).
Proceeding of 4th National Congress on Theoretical & Applied Mechanics, Leuven, May 22-23, pp
449-452, 1997.
47. A modified Koiters shakedown formulation. (in collaboration with Yan Ai Min) Colloquium of
EUROMECH 385 (8-11, Sept., Aachen, Germany), pp 83-85, 1998.
48. Limit state of cracked structures by direct mathematical programming technique. (in
collaboration with Yan Ai Min). Advanced Computational Methods in Engineering (Proceeding of
ACOMEN98), Shaker Publishing B. V. The Netherlands, pp187-194, 1998.
49. Shakedown of Structure under thermal and mechanical loading. (in collaboration with Yan Ai
Min). Constitutive and Damage Modelling of inelastic deformation and phase transformation
Proceeding of Plasticity99, Cancun, Mexico, (ed. Akhtar S. Khan), Neat Press, Fulton, Maryland,
pp773-776, 1999.
50. Delamination of multi-layered composites by interface finite elements. (in collaboration with
Yan Ai Min and E. Marchal). Proceed. of APCOM99 fourth Asia-Pacific Conf. on
Computational Mechanics, Vol. 1, Singapore, pp 249-254, Computational Mechanics for next
millennium, Edited by C. M .Wang, K.H. Lee and K. K. Ang, September 1999, Elsevier.
51. Modelling R-curve effects in delamination damage mechanism. (in collaboration with Bui-Quoc
Viet and E. Marchal). CCM12 Conference, Paris, July, paper 124, 2000.
31
52. Elastic-plastic analysis of structures using a displacement incremental two-field national

principle. (in collaboration with Nguyen-Dinh Giang). IMPLAST 2000, 7th symposium on
structural failure and plasticity in Melbourne Australia, 4-6 October 2000.
53. A new truly meshless Galerkin method. (in collaboration with Marc Duflot). Proceedings of Nha
Trang'2000 International Colloquium, August 2000, pp 433-442.
54. Invariant isoparametric metis displacement element. (in collaboration with Dang-Dinh Thi).
Proceedings of Nha Trang'2000 International Colloquium, August 2000, pp 468-489.
55. Mixed-mode delamination in laminated composites with R-curved effect... (in collaboration with
Yan Ai Min). Proceedings of Nha Trang'2000 International Colloquium, August 2000, pp 443-452.
56. A general BEM for analysis of slope stability. (in collaboration with Jiang Yansheng and Wilson
Venturini) Proceedings of Nha Trang'2000 International Colloquium, August 2000, pp 512-519.
57. Three-dimensional slope stability analysis by direct boundary element method. (in collaboration
with Do-Quang Khanh and Jiang Yansheng). Proceedings of Nha Trang'2000 International
Colloquium, August 2000, pp819-828.
58. A programme of boundary element method for 3D elastic solids on a personal computer. (in
collaboration with Do-Quang Khanh and Jiang Yansheng). Proceedings of Nha Trang'2000
International Colloquium, August 2000, pp 539-443.
59. Limit load and deformation estimates by equilibrium back stress fields analysis. (in
collaboration with Yan Ai Min and Vu-Duc Khoi). Proceedings of Nha Trang'2000 International
Colloquium, August 2000, pp 520-530.
60. Prvision de forme des pices cramiques aprs schage et caisson (in French, in collaboration
with Nguyen-Dai Quy). Proceedings of Nha Trang'2000 International Colloquium, August 2000, pp
636-644.
61. Simulation, calculation of mattress of placed blocks for pitched slope protection by SAMCEFMECANO. (in collaboration with Nguyen-Chien Thang and Doan-Duc Bao). Proceedings of Nha
Trang'2000 International Colloquium, August 2000, pp 848-857.
62. Initiation and propagation of mixed-mode interlaminar cracks in composite samples. (in
collaboration with Yan Ai Min and E. Marchal). Proceedings of the 2nd European Conference on
Computational Mechanics, June 26-29, 2001, Krakow, Poland, 15 pages, section 27.
63. A new dual shakedown analysis of structures. (in collaboration with Vu-Duc Khoi and Yan Ai
Min). Proceedings of the 2nd European Conference on Computational Mechanics, June 26-29, 2001,
Krakow, Poland, 16 pages, section 21.
64. Incremental Plasticity Limit of Structures Subjected to variable loading. (in collaboration with
Yan Ai Min and Vu-Duc Khoi). Proceedings of the 8th International Conference on Enhancement
and Promotion of Computational Methods in Engineering Science, July 2001, Shanghai, 10 pages.
65. Finite Element Modelling of Delamination of Laminated Composites, (in collaboration with Yan
Ai Min). Proceedings of the 8th International Conference on Enhancement and Promotion of
Computational Methods in Engineering Science, July 2001, Shanghai, 10 pages.
32
66. Limit analysis of 2D Curved beams by linear optimisation programming. (in collaboration with
Vo-Vinh Bao and Luu-Nguyen Nam-Hai). Proceed. Nat. Conf. Eng. Mechanics, Hanoi, October 1213 (2001), pp 10-20.
67. Modeling of the interactions between interlocking blocks, soil and water of pitched dike
revetment using SAMCEF software. (in collaboration with Nguyen-Van Hieu and Doan-Duc Bao).
Presented at the Phan Rang Conference organised by Vietnamese Society of Mechanics of Fluids
(July 22-23, 2002).
68. Stresses singularity order in the delamination of composite laminate. (in collaboration with
Nguyen-Tien Duong, Hanoi University of Technology). Proc. 7th Nat. Congr. on Mech., Hanoi,
December 18-20, 2002, pp 151-159.
69. A numerical tool useful for the calculation of three-dimensional crack propagation by finite
element method: the fissbox. (in collaboration with Tran-Duc Han and Chau-Dinh Thanh, EMMC
Centre, Ho Chi Minh University of Technology). Proc. 7th Nat. Congr. on Mech., Hanoi, December
18-20, 2002, pp 204-211.
70. Comparative analysis for block mattresses devoted to seashore protection. (in collaboration
with Nguyen-Van Hieu). Proc. 7th Nat. Congr. On Mech., Hanoi, December 18-20, 2002, pp 254261.
71. Analysis of uncracked and cracked plate bending problem using metis finite element model... (in
collaboration with Tran-Thanh Ngoc, Nguyen-Tai Trung). Proc. 7th Nat. Congr. On Mech., Hanoi,
December 18-20, 2002, pp 262-270.
72. Development of a metis finite element model of shell problem with or without crack. (in
collaboration with Tran-Thanh Ngoc, Nguyen-Tai Trung). Proc. 7th Nat. Congr. On Mech., Hanoi,
December 18-20, 2002, pp 271-279.
73. Determination of the toughness of the quasi-3D structure by using the metis finite element
model. (in collaboration with Do-Van Truong). Proc. 7th Nat. Congr. on Mech., Hanoi, December
18-20, 2002, pp 622-633.
74. Investigating three kind of plane contact problem by penalty method... (in collaboration with
Trinh-Bach Tuyet, Ho Chi Minh University of Architecture). Proc. 7th Nat. Congr. on Mech.,
Hanoi, December 18-20, 2002, pp 690-696.
75. Application of the SAMCEF-MECANO to the modelling of the interaction between concrete
block mattresses, elastic foundation and waves. (in Vietnamese, collaboration with Nguyen-Van
Hieu and Doan-Duc Bao), Proceeding of the seminar on the Coastal Protection of sea and river, pp
77-94, Union of Vietnamese Sciences and Technology associations, Hanoi, November 14/2003.
76. Strength of the steel net assemblage of concrete block mattresses used as coastal cover in
Mekong River, Long Xuyen, Vietnam. (in Vietnamese, collaboration with Nguyen-Van Hieu and
Chau-Dinh Thnh). Proceeding of the seminar on the "Coastal Protection of sea and river", pp.
123-138, organised by Union of Vietnamese Sciences and Technology associations and University
of Water Resource of Hanoi, November 14/2003.
77. Global error estimation with displacement and equilibrium meshless methods. (in collaboration
with Marc Duflot). Proceedings of the Sixth World Congress on Computational Mechanics in
conjunction with Second Asian-Pacific Congress on Computational Mechanics Beijing, China,
33
September 5-10, 2004, published by Z.H. Yao, M.W. Yuan and W.X. Zhong, Tsinhua University
Press and Springer, page 316-321, 2004.
78. Duality in kinematical approaches of Limit and Shakedown Analysis of Structures. (in
collaboration with Yan Ai Min and Vu-Duc Khoi). Published in the book "Complementary, duality
and symmetry in nonlinear mechanics", proceed. of Shanghai IUTAM Symposium, Edited by
David Gao, Kluwer Ac. Publ., pp 128-148, 2004.
79. Using metis model to solve torsion problem of prismatic bar. (in collaboration with NguyenTien Duong). The 7th National Congress on Deformation Solid Mechanics Do Son, Vietnam 27-28
August 2004.
80. Calculation of free edge stress for composite laminates by metis element method. (in
collaboration with Nguyen-Tien Duong). International Conference: Mcanique pour lIngnieur
2004 Ho Chi Minh City Vietnam August 16-20, 2004
81. Direct and indirect Mtis element methods for edge delamination in laminates. (in collaboration
with Nguyen-Tien Duong). COMPUTATIONAL MECHANICS WCCM VI in conjunction with
APCOM04, Sept. 5-10, 2004, Beijing, China 2004 Tsinghua University Press & SpringerVerlag.
82. Duality in meshless methods. (in collaboration with Marc Duflot). Proceeding of the 7th National
Congress on Deformation Solid Mechanics Do Son, Vietnam 27-28 August 2004.
83. Dual limit analysis of plate bending, (in collaboration with Le-Van Canh and Nguyen-Xuan
Hung), ACOMEN-080, 3rd International Conference on advanced Computational Method in
Engineering, 30 May to 2 June, 2005, Gent, Belgium.
84. Metis displacement method to calculate free-edge stresses under extension, bending and
twisting, (in collaboration with Nguyen-Tien Duong), ACOMEN-015, 3rd International Conference
on advanced Computational Method in Engineering, 30 May to 2 June, 2005, Gent, Belgium
85. Some new results Scientific in dual analysis applied to meshless method. (Keynote lecture) (in
collaboration with Bui-Quoc Tinh, Marc Duflot, and Stephane Bordas). The 7th World Congress on
Computational Mechanics - WCCM 2006, Los Angeles, California, USA.
86. Modification of a quarter point element in calculating fracture mechanics, (in collaboration
with Le-Hoai Long), Proceedings of the 8th National Congress on Deformation Solid Mechanics,
August 25-26, 2006, Thai Nguyen, Vietnam.
87. Equilibrium model in the element free Galerkin method, (in collaboration with Bui-Quoc Tinh),
Proceedings of the Eighth International Conference on Computational Technology, Edited by
B.H.V. Topping, G. Montero and R. Montenegro, Las Palmas de Gran Canaria, 12-15 September
2006, paper N 177, pages 405-406.
88. Metis Element Model for Interlaminar Stresses in Composite Laminate,
(in collaboration with Nguyen-Tien Duong), Proceedings of the Eighth International Conference on
Computational Technology, Edited by B.H.V. Topping, G. Montero and R. Montenegro, Las
Palmas de Gran Canaria, 12-15 September 2006, paper N 266, pages 619-620.
89. An united algorithm for calculate the limit states of plan frameworks with semi-rigid
connections CEPAO, (in collaboration with Le-Hoang Long), to be presented in The Third
34
International Conference on Structural Engineering, Mechanics and Computation, 10-12 September

2007, Cape Town.
90. Equilibrium model in the element free Galerkin method, (in collaboration with Bui-Quoc Tinh),
Proceedings of the Eighth International Conference on Computational Technology, Edited by
B.H.V. Topping, G. Montero and R. Montenegro, Las Palmas de Gran Canaria, 12-15 September
2006, paper N 177, pages 405-406.
91. Metis Element Model for Interlaminar Stresses in Composite Laminate,
(in collaboration with Nguyen-Tien Duong), Proceedings of the Eighth International Conference on
Computational Technology, Edited by B.H.V. Topping, G. Montero and R. Montenegro, Las
Palmas de Gran Canaria, 12-15 September 2006, paper N 266, pages 619-620.
92. An united algorithm for calculate the limit states of plan frameworks with semi-rigid
connections CEPAO, (in collaboration with Le-Hoang Long), to be presented in The Third
International Conference on Structural Engineering, Mechanics and Computation, 10-12
September 2007, Cape Town.
10.4 Reports (chronological order)

1. On the Statistic Model of Atom (Quantum Mechanic, Master Thesis in French). Faculty of
Sciences, Laboratory of Theoretical Physics (Prof. Pirenne), University of Lige, (1966).
2. Digital computation of stresses and deflections in a box beam. A performance comparison
between finite element models and idealisation patterns (in collaboration with G. Sander and P.
Beckers). Technical Report AFFDL-TR-69-4, January 1969. Collection des Publications, Nr 4, page
87, 1967.
3. Natural strains and stresses for trapezoidal structures analysis (in collaboration with Professor
B. Fraeijs De Veubeke). 37 p., 6 figures, Technical Report AFFDL-TR-69-18, February 1969.
4. Foundation of the finite element method. Report Nr 42 (in French). Laboratory of Strength of
Materials and Stability of Constructions, University of Lige, 1971.
5. A new interpretation of KOLOSOV-MUSKHELISHVILI functions for plane problems in
elasticity. Displacement formulation. Report Nr 12, Laboratory of Strength of Materials and
Stability of Constructions University of Lige, 1972. Presented at French Congress of Mechanics,
17-20 September 1973, Poitiers, France.
6. Mechanics of the behaviour of the non-elastic material. Post-graduate course (in French)
Laboratoire de Mcanique des Matriaux et de stabilit des Constructions, Lige, 140 p. (1980).
7. Finite element solutions of unilateral problems, (in French). September 1980. Laboratoire de
Mcanique des Matriaux et de Stabilit des Constructions, Lige, Belgium.
8. Data preparations and output explanations of the program CEPAO (in French). (in collaboration
with R. Baltus, Gry de Saxc, and E. Lemaire). Report Nr 109. Laboratoire de Mcanique des
Matriaux et de Stabilit des Constructions, 1980.
9. A thermodynamic model for the contact problems of solids (in French and in collaboration with
Gry de Saxc). Report Nr 103, Laboratoire de Mcanique des Matriaux et de Stabilit des
Constructions- 1980.
35
10. A metis finite element with arbitrary stress field included a singular field for fracture. Report Nr
111, September 1981, Laboratoire de Mcanique des Matriaux et Statique des Constructions,
University of Lige.
11. User's Guides of CEPAO-82 - Computer Program for Analysis and Stable Design of ElasticPlastic Structures - Data preparation. (in collaboration with Gry de Saxc, E. Lemaire and P.
Villers), report Nr 127, Laboratoire de Mcanique des Matriaux et Statique des Constructions,
Universit de Lige, 80 p., November (1982).
12. On the Simplex Algorithm in CEPAO-82. Report Nr 133. Laboratoire de Mcanique des
Matriaux et Statique des Constructions, Universit de Lige, March 1983.
13. Cyclic Loading Studies of General Hardening Models, Report Nr 138. Laboratoire de
Mcanique des Matriaux et Statique des Constructions, Universit de Lige. December 1983.
14. On the Plasticity and the Calculation of Limit States by F.E.M. (in French). Thesis of the degree
of "Agrgation" or Special Ph.D. Facult des Sciences Appliques, Universit de Lige,
Collection des Publications Nr 98, 1985.
15. Evaluation des facteurs d'intensit de contrainte des structures 3 dimensions (in collaboration
with Gry de Saxc). Internal report, Nr 175, 45 pages (1988), Depart. M.S.M., University of Lige.
16. Automatisation de maillage en front de fissure. (in collaboration with Gry de Saxc), Internal
report Nr 178, 32 pages (1988), Department M.S.M., University of Lige.
17. Evaluation du taux de restitution d'nergie des structures fissures 3 dimensions. (in French,
in collaboration with Gry de Saxc). Internal report Nr 179, 35 pages (1988), Department M.S.M.,
University of Lige.
18. Manuel d'utilisation du logiciel SAMCEF pour la rsolution des problmes tridimensionnels de
mcanique de la rupture. Final report, SEP-SAMTECH-MSM, Nr 190, March 1989, SAMTECHMSM, Universit of Lige.
19. Analyse Limite des Coudes par la Mthodes des Elments Finis et la Programmation
Mathmatique. (in collaboration with R. Jospin). Internal report MR-01/90, LTAS Fracture
mechanics of solids, December 1990.
20. Maillage automatique d'une bote fissure. Internal report LTAS- Fracture Mechanics of Solids,
Nr MR-02/90, January 1991, 30 pages, 1991.
21. Estimation de l'intgrale J lastique et lasto-plastique par la mthode du domaine de
l'intgration quivalente (EDI). (In collaboration with Kang Chi Hang). Internal report LTASMcanique de la Rupture, University of Lige, January 1992.
22. Limit analysis of elbows using Finite Element Method and Mathematical Programming
Techniques. Report "Limit loads of pipe elbows, contact RA1-0134B", (in collaboration with M.
Save and R. Jospin).
23. On the shakedown theorems and the finite element computation of the shakedown limit loads.
Internal report LTAS- Fracture Mechanics of Solids, June, 1994.
36
24. Modlisation du Dlaminage par la Mcanique de l'Endommagement... (in French, in

collaboration with E. Lambert, G. Laschet, Gry de Saxc), section B of the 1st report:
"Dveloppement et validation de modles pour la caractrisation des proprits mcaniques et
Multimatriaux project sponsored by Walloon Region, LTAS- Fracture Mechanics of Solids,
March, 1994.
25. Gnration linaire d'une famille d'lments fins d'interface. (in French, in collaboration with
E. Lambert, G. Laschet, Gry de Saxc), section B of the 2nd report: "Dveloppement et validation
de modles pour la caractrisation des proprits mcaniques et la description des mcanismes
d'endommagement de structures en matriaux composites, Convention N 2397, project sponsored
by Walloon Region, LTAS-Fracture Mechanics of Solids, November, 1994.
26. Limit loads of pipe and elbows (in collaboration with M. Save and R.J. Jospin). European
commission, nuclear science and technology, Final report, EUR 15696EN, 1995.
27. Calcul des charges limites des coudes par un lment fini de poutre et par la mthode de
programmation mathmatique. (in collaboration with Yan Ai Min), report N 1, FRAMATOMELTAS, LTAS- Fracture Mechanics of Solids, February 1995.
28. Modlisation et simulation de la croissance d'endommagement et de fissuration des structures
mtalliques et composites soumises la fatigue et en prsence des contraintes rsiduelles. (in
collaboration with B. Borle, B. De Bel, Ph. Putz). First report, COST 512, Convention 2670
sponsored by Walloon Region, LTAS- Fracture Mechanics of Solids, March, 1995.
29. Limit analysis and shakedown computation of shells and elbows (in collaboration with Yan Ai
Min and Bui-Cong Thanh). Internal report N R96-2-1. February 1996.
collaboration with Borle, B., De Bel B. and Putz Ph). Internal report, COST Project 512, Period
from September 15.94 to March 15.95. 1995.
collaboration with Borle, B., De Bel B. and Putz Ph). Internal report, COST Project 512, Period
from September 15.95 to March 15.95. 1996.
32. Modlisation du dlaminage par la mcanique d'endommagement. (in collaboration with BuiQuoc Viet). Internal report, LTAS- Fracture Mechanics of Solids, October 1996.
33. On a general method for yield criterion deduction in generalised variables. (in collaboration
with Bui-Cong Thanh). Internal report, LTAS- Fracture Mechanics of Solids, January 1997.
34. Application of shakedown methods to piping. (in collaboration with Bucktthorpe, P. Rathjen, H.
Hoell, Y. Renault and Yan Ai Min). Final report, CEC/WGCS-AG2, FBR study contract RAI-0235
UK, NNC, May 1997.
35. Practical estimation of plastic collapsed limit of curved pipe subjected to complex loading. (in
collaboration with Yan Ai Min) Internal report, LTAS- Fracture Mechanics of Solids, February
1997.
37
36. Limit analysis of defects. (in collaboration with M. Staat, M. Heitzer, Yan Ai Min, Vu-Duc
Khoi, F. Voldoire and A. Lahousse). Report N 3746, Forschungszentrum Jlich, Institut fr
Sicherheitsorschung und Reaktortechnik, Brite-Euram III contract N BRPR-CT97-059, Project
BE97-4547, March, 2000.
10.5 General papers (chronological order)

1. Report concerning the training programme and the Vietnamese universities (1989).
2. Report concerning the activities of the university co-operation sponsored by agency FICU,
AUPELF&UREF (1991).
3. The overseas Vietnamese's and the scientific co-operation with Vietnam (in Vietnamese appeared
in DOAN KET). N 418, France, November 1989.
4. Comments on the scientific co-operation with Vietnam (in Vietnamese), "TUOI TRE CHU
NHAT NEWS PAPER" Ho Chi Minh City, February 25, 1990.
5. On the experiences of the university co-operation with Vietnam. Workshop on Engineering
Education in Developing Countries, Proceedings of the Conference of European Schools for
Advanced Engineering Education and Research (CESAER), Liege 29-30 March, 1993.
6. About some experiences in the education cooperation with the Vietnam, (in Vietnamese), DOI
THOAI N 5, April, 2005, pp50-61.
7. On the programme EMMC and the Belgian grants to Vietnamese Universities. Paper (in
Vietnamese) published in 'TUOI TRE (Youth)", Ho Chi Minh City, Saturday, February, 24, 1996.
8. Final report of the activities of the programme EMMC (1995-1999). (in French), LTAS-Fracture
Mechanics, November 1999.
9. EMMC and MCMC, Master training programmes located in Vietnam. Seminar in Europe
organised by Agence Universitaire de la Francophonie, University of Evry Val dEssonne, 25-27
March 1999.
10. Final activity report of the programme PIP-EMMC, 95-99. LTAS-Mcanique de la Rupture,
November 1999.
11. Report on the feasibility of a creation of an enterprise on the domain of CAO-FEM in Vietnam.
LTAS-Mcanique de la Rupture, May, 2000.
12. The man who tries to find the Ph. D. for the Vietnam Interview realised by Tran-Truc Thuy
(Truc Mai), Vietnamnet-NguoiVienXu electronic magazine 16/2/2004:
http://www.nguoivienxu.vietnamnet.vn/vinhdanhnuocviet/2004/02/51745/
13. I just follow the feelings of my hearth, an appointment with Prof. Nguyen Dang Hung,
Vietnamnet-NguoiVienXu electronic magazine, 2004:
http://www.nguoivienxu.vietnamnet.vn/vandecuachungta/2004/04/59335/
14. The Bologna decree had upset profoundly the European university system, Vietnam could not
ignore! Thanh Nien daily 6/1/2004.
http://web.thanhnien.com.vn/Giaoduc/2005/4/4/78732.tno
38
15. The EMMC and MCMC programmes had helped the Vietnamese Universities to enhance the
level of their postgraduate formation to be near the best ones in the world, Thanh Nien daily
30/5/2004.
http://web.thanhnien.com.vn/Giaoduc/2005/4/4/79189.tno
16. What about the trade mark of a university in the world, (in Vietnamese), Tuoi Tre daily
4/2/2005.
http://www.tuoitre.com.vn/Tianyon/Index.aspx?ArticleID=65780&ChannelID=13
17. Conversation at the beginning of the New Year: about the Belgian delocalised postgraduate
training programmes in Vietnam, (in Vietnamese), Hanoi-Moi Sunday, 09/02/2005.
http://www.hanoimoi.com.vn/vn/print/37152/
18. New measurement to attract overseas Vietnamese: some hearths would be glad again?, in
Vietnamese, Vietnamnet electronic magazine, 23/9/2005.
http://www.vnn.vn/chinhtri/doinoi/2004/09/262004/
19. It is necessary to change the mentality to realise the reform of the Vietnamese system of
education, (in Vietnamese), Vietnamnet-NguoiVienXu electronic magazine, 18/10/2004.
http://nguoivienxu.vietnamnet.vn/diendannguoivienxu/2004/10/336312/
20. Education isnt a good, (in Vietnamese), Vietnamnet-NguoiVienXu electronic magazine,
18/10/2004.
http://nguoivienxu.vietnamnet.vn/diendannguoivienxu/2004/10/336324/
21. One couldnt attract the overseas Vietnamese brain by slogans, (in Vietnamese),
Vietnamnet-NguoiVienXu electronic magazine, 31/01/2005.
http://www.nguoivienxu.vietnamnet.vn/ykienbandoc/2005/01/371531/
22. Artist Trinh Cong Son, just a little breeze from him, (in Vietnamese), VietnamnetNguoiVienXu electronic magazine, 31/3/2005. And in the website (in Vietnamese: Mt thong
Trnh Cng Sn): (in Vietnamese), Published in Hon Viet, HCM City (2005) and in Thoi Van,
N 2, HCM City, 2005. This paper may be found in Internet:
http://www.trinh-cong-son.com/ndhung.html
http://nguoivienxu.vietnamnet.vn/quenhanoinho/2005/03/402519/
23. Travel notes of April and some other months and years, (in Vietnamese), VietnamnetNguoiVienXu electronic magazine, 29/4/2005.
www.nguoivienxu.vietnamnet.vn/nguoivietbonphuong/2005/04/416965/
24. Let us get toward the creation of an overseas Vietnamese scientists organisation, (in
Vietnamese), Vietnamnet-NguoiVienXu electronic magazine, 16/8//2005.
http://www.nguoivienxu.vietnamnet.vn/nguoivietbonphuong/2005/08/479463/
25. Overseas Vietnamese brains are waiting over the world, (in Vietnamese), Vietnamnet
electronic magazine, 24/9/2005.
http://www.vnn.vn/chinhtri/skbl/2004/09/263251/
26. For our country the disparities must disappeared, (in Vietnamese), Vietnamnet electronic
magazine, 29/4/2005.
http://www.vnn.vn/chinhtri/doinoi/2005/04/417091/
39
27. Overview about the postgraduate programme EMMC, Edited by the electronic magazine
Vietsciences, in Vietnamese, Nhn li chng trnh o to cao hc EMMC:
http://vietsciences.free.fr/vongtaylon/kinhnghiemthaytro/nguyendanghung_emmc.htm
28. Prof. Nguyen Dang Hungs interview, edited by the electronic magazine Vietsciences, in
Vietnamese: Phng vn Gio s Nguyn ng Hng:
http://vietsciences.free.fr/vongtaylon/kinhnghiemthaytro/phongvan_ndhung.htm
29. A solution of the actual problem in Vietnam: the education in a free market society . Lecture
presented in the Summer times Conference in Da Nang, July 2005:
http://hoithao.viet-studies.org/Hoithao2005.htm
30. If we miss this present opportunity we are guilty toward our ancestors, (in Vietnamese),
Declaration on the tribune of the Fatherland Front's meeting, Ba Dinh, Hanoi, September 2004:
www.nguoivienxu.vietnamnet.vn/vandecuachungta/2004/10/314488/
http://www.nguoivienxu.vietnamnet.vn/vandecuachungta/2004/09/263094/
31. La matrise europenne en Mcanique des Constructions en voie de vietnamisation, (in
French), ECHOSUD, Bulletin trimestriel de la CUD, N 10, March 2006, pp 14-15.
32. I regret , (in Vietnamese), Vietnamnet-NguoiVienXu electronic magazine, 30/7/2006.
http://nguoivienxu.vietnamnet.vn/doisongnvx/hdvktrongnuoc/2006/08/600511/
www.nguoivienxu.vietnamnet.vn/nguoivietbonphuong/2006/08/600807/
33. If I am Minister? I will fire at least 50% government officials, (in Vietnamese), Vietnamnet
http://nguoivienxu.vietnamnet.vn/doisongnvx/hdvktrongnuoc/2006/08/600511/
http://www2.vietnamnet.vn/giaoduc/vande/2006/08/599139/
http://www2.vietnamnet.vn/giaoduc/vande/2006/08/599739/
34. Condition for Vietnam to be a powerful country, (in Vietnamese), Vietnamnet-NguoiVienXu
http://nguoivienxu.vietnamnet.vn/ykienbandoc/2006/08/607538/
And more than 100 other articles or interviews about education, university organisation,
overseas Vietnamese and the Vietnam, published in Vietnamese press (1990-2006).
10.6 Poetry publications (chronological order)

1. Time is young forever (in Vietnamese), Collection of poems in Vietnamese, edited by Hien-Dien,
Liege, 1973.
2. The shadow of times (in Vietnamese), Collection of Poems in Vietnamese, edited by ThanhLong, Brussels 1973.
3. The beaten track (in Vietnamese), Collection of poems in Vietnamese edited by "TRE
publishing", Ho Chi Minh City, 1997.
4. Collection of poems (in Vietnamese with several authors) edited by Trang An Club, Lao Dong
publication, Hanoi, pp 120, 1998.
40
5. Flowers of love V, Collection of poems (in Vietnamese with several authors) published by
Association of Vietnamese amateur poets, Sacramento, USA, pp 250-253, 1999.
6. Flowers of love VII, Collection of poems (in Vietnamese with several authors) published by
Association of Vietnamese amateur poets, Sacramento, USA, pp 183, 2001.
7. Flowers of love VIII, Collection of poems (in Vietnamese with several authors) published by
8. Flowers of love International, vol. 3, Collection of poems (in French and English with several
authors) published by Association of Vietnamese amateur poets, Sacramento, USA, pp 212, 2002.
9. Hoa Vng (Yellow flowers, in Vietnamese), selected poems of 50 authors, published by Ngoc
An, San Jose, CA, USA, pp 135-140, 2003.
10. Flowers of love IX, Collection of poems (in Vietnamese with several authors) published by
11. Flowers of love International, vol. 4, Collection of poems (in French and English with several
authors) published by Association of Vietnamese amateur poets, Texas, USA, pp 103-106, 2004.
12. Flowers of love X, Collection of poems (in Vietnamese with several authors) published by
Association of Vietnamese amateur poets, Texas, USA, pp 26-29, 2004.
13. Nguyen Dang Hung and his poem about the happiness, (in Vietnamese), VietnamnetNguoiVienXu electronic magazine:
www.nguoivienxu.vietnamnet.vn/butviet/2004/12/352442/
14. Two poems in Vietnamnet-NguoiVienXu electronic magazine: ng di & Bc th ma h:
www.nguoivienxu.vietnamnet.vn/trangthonvx/2004/03/54208/
15. The poem: I choose the spring for my home, (in Vietnamese) (Toy chin moa Xian lame co
noggin), Vietnamnet-NguoiVienXu electronic magazine:
www.nguoivienxu.vietnamnet.vn/butviet/2005/01/361913/
16. Flowers of love XI, Collection of poems (in Vietnamese with several authors) published by
Association of Vietnamese amateur poets, Texas, USA, to be appeared, 2006
...
41
International Editorial Advisory Members Papers
A Mtis Finite Element for Singular Stress Analysis of Junctions

and Delamination Cracks and Composite Structures
Gry de Saxc*, D. Crpin+ and M. Pyrz*
* Laboratoire de Mcanique de Lille, UMR CNRS 8107,
Villeneuve dAscq, France
+Laboratoire de Gnie et Matriaux Textiles
Roubaix, France
Abstract
If the design of laminates can be considered as globally mastered, the one of junctions still holds
unsolved difficulties. In this work, we developed a hybrid finite element to compute accurately the
Stress Intensity Factors associated to the singularities around lines at the junction between different
anisotropic or/and isotropic materials. After numerically determining its complex exponents, the
singularity is used to enhance the stress field of the finite elements near these lines. Using BabukaBrezzi condition enables to select the appropriate stress mode functions. The approach is illustrated
by applications to a delamination crack in cylindrical shells and a junction with aluminium insert in
a glass/epoxy sandwich plate.
Keywords: fracture mechanics, composite materials, junctions, delamination cracks.
Introduction
If the design of composite structures as laminates or sandwich plates can be considered as globally
mastered, the one of junctions still holds unsolved difficulties. The junctions may be of various
kinds, for instance inserts, rivets or sticking. For brittle materials as composites, cracks can be
initiated from regions where the stress field is singular, around junction lines at the junction
between different anisotropic or/and isotropic materials. Generally speaking, such junction lines and
crack fronts are qualified as singular lines in the sequel. As example, a junction with aluminium
insert in a glass/epoxy sandwich plates is considered. Under service loads, even weak, cracks can be
initiated and leads to the pull-out collapse as in Figure 1.
Figure 1: Pull-out collapse of an insert junction in a sandwich plate

43
The 3D geometric nature of these junctions makes the problem of computing the associated Stress
Intensity Factor very awkward. Most of the usual methods (dual singular functions, Leguillon and
Sanchez-Palencia [9], hybrid finite elements, Kim and Im [7]) are unsuited. Thus we developed a
numerical method to compute the singular stress field. In a second step, to obtain accurate values of
the associated Stress Intensity Factors, we implemented a 3D mtis hybrid finite element, based on
a unisolvent displacement field and an equilibrated stress field enhanced by the previously
determined singular field. Babuka-Brezzi inf-sup condition enables to select the appropriate stress
mode functions, as highlighted by a benchmark. The code allows computing stress singularities
around cracks. The approach is illustrated by two numerical applications.
Numerical determination of the singularity exponents
2.1 Singular stress fields around singular lines

Let be a parameter along a singular line. Following the problem, is the arc length or the angle.
Let r and be the polar coordinates in the current plane orthogonal to the singular line. Following
the works of Lekhnitskii [10], Wang and Yuan [18-19], we consider singular fields depending
analytically on the complex variables:
z k = r k , k = cos + k sin , k = 1,2,3 ,
where the complex number k are solutions of a eigenvalue problem depending on the considered
materials. On this ground, Lekhnitskii [10] proposed a stress field for anisotropic elastic beams.
Latter on, S.S. Wang uses Lekhnitskiis solution to build, in the curvilinear coordinate system
(r,,), the asymptotic singular part of the stress field near the singular line in the following form
[18]:
= 2 e K ( ) r ( ) ( k ) ( ) A + ( k ) ( ) B .
))
The complex number is the singularity exponent. Unlike for cracks in homogeneous materials,
it is not generally equal to -1/2. The determination of its value is a first numerical difficulty to face.
This computation already gives qualitative information, helpful to the design. The K are the
associated Stress Intensity Factors and, for the moment, their values are indeterminate. The (k)
are matrices associated to the solutions k of Lekhnitskiis problem. The () are diagonal matrices
with (k) generic value when the k are distinct. The case of multiple values is detailed in [3]. The
A and B are 33 complex unknown matrices associated to each exponent . More details on this
singular fields are given by D. Nguyen Tien and H. Nguyen Dang in [14] and in a twin paper
published in the same book.
2.2 The eigenvalue problem

Next, the continuity condition on the stress and displacement fields are imposed at each interface
between materials and, if necessary, boundary conditions are enforced on free stress surfaces
starting from the singular line. The set of all these condition gives an eigenvalue problem with the
following standard form:
C ( ) q = 0 ,
where the complex square matrix C () analytically depends on , and q is the corresponding
eigenvector containing the elements of A and B. The properties of such systems were detailed by
44
Leguillon and Sanchez-Palencia [8]. The matrix C () is automatically obtained by assembling

elementary matrices corresponding to each material, similarly to the assembly operation in finite
element method.
The exponent corresponds to a singular stress field if its real part is negative. For the elastic
energy being finite, the real part of must be great than to -1. Relevant values of are expected to
belong to the strip -1/2 R () < 0. The eigenvalues are the ones for which the determinant of C ()
vanishes. They are first roughly localised by a 2D bisection method, next accurately determined
using regula falsi method. One of the difficulties is the event of multiple eigenvalues. Multiplicity
is detected and computed by using Rouchs lemma. Next, the corresponding eigenvector is
computed by incomplete LU decomposition method, that allows to build automatically the
corresponding matrices A and B and to determine the eigenstress fields. An example for
visualisation of singular fields computed by the code is presented in Figure 2. The code can treat as
particular cases the stress singularity around a crack front, especially for delamination cracks.
(a) Displacements
(b) Stresses
Figure 2: Visualisation of the singular fields
Computation of the Stress Intensity Factors
3.1 Mtis finite elements

The mtis elements (or mongrel elements) form a special class of hybrid finite elements proposed
by Nguyen Dang [12-13] for their very good convergence properties. In the next, we present and
use only the mtis displacement elements, based on a displacement field u on the element boundary
and a stress field in equilibrium inside the element. According to the principle of the hybrid
elements, the continuity of the displacement field is required on the element interface, but not the
one of the stress vector T (). The displacement field acts as Lagrange multiplier to impose the
stress vector continuity in a weak or relaxed form. The exact solution corresponds to the saddlepoint of a two-field functional or Lagrangean L (,u) [20] :
L ( , u ) = ( W ( ) dV
e
Ve
Ve
(T ( ) ) T u dS +
ST Ve
TdT u dS )
In this formula, W is the complementary energy potential of the material. As usual in FEM, the
stresses and strains are stored in column vectors. The sum is performed over all the elements Ve of
the mesh, and ST is the part of the boundary where the prescribed traction vector Td is imposed.
Hybrid displacement elements are known as exhibiting a better convergence than the corresponding
pure displacement elements, when refining the mesh.
45
The mongrel displacement elements are characterised by the unisolvence property: the
displacement field on the boundary can be extended in a unique manner inside the finite element
within a linear space of complete polynomials of given order. Let S be the compliance matrix of the
material, such that:
W ( ) = 12 T S .
Because of the convexity of this potential and the unisolvence property, the Lagrangean can be
transformed through Gauss divergence theorem to be written as follows:
L ( , u ) = ( (W ( ) T (u )) dV +
e
Ve
ST Ve
TdT u dS ) ,
where (u) is the symmetric gradient of u. Thus interface integrals are replaced by volume integrals
of which the computation is more accurate and simple to implement. Moreover, for uniform mesh
refinement, it is proved than the energy convergence is monotonic decreasing and faster than the
one of the corresponding pure displacement element [12].
The displacement field and the associated strain field (u) linearly depend on the unknown
nodal displacements stored in the column vector q:
u = N q,
(u ) = B q .
The elements of the shape function matrix N are polynomial fields. Far from the singular line, the
stress field linearly depends on unknown stress parameters stored in a column vector h:
r = rh ,
where the elements of the matrix r are polynomial fields and generate stress fields in internal
equilibrium inside each element.
3.2 Singular finite elements

The mtis finite elements were used to compute the Stress Intensity Factors for various types of
problems in Linear Elastic Fracture Mechanics: cracks in homogeneous isotropic and anisotropic
materials ([14], [16]), cracks in bending plates [17] and edge delamination cracks ([2], [15]). The
use of hybrid finite elements enhanced by the singular stress field determined at the first step allows
averting very refined mesh of standard finite elements around the singular line, what leads to huge
computation times. Around the singular line, the stress field is enriched by singular terms
determined in Section 2:
= r +s .
The singular part s of the field is linearly depending on the values of the Stress Intensity Factor at
given points of the singular line, stored in a column vector k:
s = sk .
In order to improve the accuracy of the solution, the singularity is also introduced by modifying the
displacement shape function matrix N as done by Akin [1].
46
Let us introduce the flexibility matrices:

Fr =
e
Ve
Fs =
rT S r dV ,
Ve
Frs =
sT S s dV ,
Ve
rT S s dV ,
the connexion matrices:

Gr =
e
Ve
Gs =
B T r dV ,
Ve
B T s dV ,
and the generalized force vector:

P=
e
ST Ve
N T Td dS .
The discretized form of the Lagrangean L is:

Lh (h, k , q) = 12 h T Fr h + h T Frs k + 12 k T Fs k (Gr h + G s k P) T q .
From which one we deduce the discretized saddle-point problem:
Fr h + Frs k = GrT q , Frs h + Fs k = G sT q , Gr h + G s k = P .

The vector h of the regular stress parameters is eliminated by condensation in each element, so
introducing the reduced flexibility and connexion matrices:
~
Fs = Fs FrsT Fr1 Frs ,
~
G s = G s G r Fr1 Frs ,
and the stiffness matrix:
K r = Gr Fr1GrT ,
leads to the linear system:
Kr
~T
G s
~
G s q P
~ = ,
Fs k 0
that can be solved by standard methods and, in addition to the values of the nodal displacements,
directly provides ones of the Stress Intensity Factors.
3.3 An automatic procedure to select the appropriate stress mode functions

Generating polynomial stress fields in equilibrium inside 2D finite elements is an easy task thanks
to Airys function. For 3D bodies, the general solution of the homogeneous internal equilibrium
equations is due to Beltrami in the form:
= curl (curl A) ,
47
where A is a symmetric second-order tensor field (see for instance [5]). Thus there are six
independent stress functions generating a very huge set of stress fields in internal equilibrium.
Improving the accuracy of the solution by using larger order polynomial stress functions gives rise
to difficulties because it increases the computation times for the inversion of the flexibility matrix
Fr but, above all, because of the existence of spurious stress modes that strongly deteriorate the
accuracy of the element.
To skirt this pitfall, we use the inf-sup condition of Babuka-Brezzi theory. Let and u
belonging to appropriate normed vector spaces, respectively denoted V and U. Let us introduce the
bilinear forms:
a ( , ) =
e
Ve
T S dV ,
b ( , u ) =
e
Ve
T (u ) dV ,
and the linear form:

l (u ) =
e
ST Ve
TdT u dS .
The Lagrangian becomes:

L ( , u ) = 12 a ( , ) b ( , u ) + l (u ) ,
and the saddle-point is reformulated as follows: find V and u U such that for all admissible
variations and u :
b ( , u ) = l (u ) ,
b ( , u ) = a ( , ) .
The existence of the solutions is ensured by the coerciveness of a:

a ( , )
2
V
on the kernel of b, and Babuka-Brezzi in-sup condition,

sup
b ( , u )
,
V uU
to be satisfy for 0 and u 0. The solution of the continuous problem has to be found in infinite
dimensional functional spaces. The inf-sup condition is satisfied because of the coerciveness. For
the discretized problem, the solution is found in finite dimensional spaces of approximation and the
above condition is not satisfied in general. When generating the stress field in equilibrium, the
appropriate stress modes are automatically selected by testing the inf-sup condition in order to
eliminate the spurious modes that do not satisfy it.
This method is highlighted by a benchmark of a trapezoidal laminated beam with a built-in
support at the first extremity and uniform normal traction load at the other one. The discretization is
represented in Figure 3.
48
l
L
t
t = 0.25
L=8
built-in
Figure 3: Trapezoidal laminated beam

The comparison between a classical hybrid element and the present one is made in Figure 4.
With respect to the shape ratio n between the length L and the thickness l of the beam, the
displacement error strongly increases up to more than 40 % for the classical element, while it is
quasi null for the present mtis element.
Figure 4: Displacement error with respect to the shape ratio n = L / l
Numerical examples
4.1 Delamination cracks in cylindrical shells

The problem proposed by Mnch and Ousset in [11] is considered here. A delamination crack is
propagating in a cylindrical shell of axial length L = 100 mm, mean radius r = 100 mm and
thickness e = 5 mm (Figure 5).
49
built
Figure 5: Delamination crack in a cylindrical shell

The delaminated length is Ld = 25 mm while the uncracked length s Lu = r / 2. The material is
elastic isotropic with E = 126 GPa and = 0.3. The mesh is composed of 288 cubic 8-nodes finite
elements. Opposite loads are applied to the crack edges, what leads to the opening mode singularity
(Figure 6). In case (a), the load is uniform along the edges, while in case (b), it is concentrated at the
middle of the edge.
(a) Uniform loads
(b) Concentrated loads
Figure 6: Loads applied to the crack edges

In Figure 7, we plotted the distribution of the Stress Intensity Factor for the opening, sliding and
tearing modes. They are normalized according to:
lim r 0 r 22 = K I ,
lim r 0 r 21 = K II ,
lim r 0 r 23 = K III .
For every load, the distribution of the Stress Intensity Factors (S.I.F.) along the crack front is given
in Figure 7. The unit for the S.I.F. is MPa.mm-1 / 2. The opening mode is clearly the dominant one.
For the event of concentrated loads (Figure 7, (b)), the distribution of KI is bell-shaped around the
force location. Also, the values for the S.I.F. of the two other modes are more significant than in
case of uniform loads, but remain small with respect to the one of the opening mode.
(a) Uniform loads
(b) Concentrated loads
Figure 7: Stress Intensity Factor distribution along the crack front

50
Most of the existing methods of calculating the Energy Release Rate (E.R.R.) are based on the
virtual crack extension method (see for instance [4], [6]). One of the advantages of mtis hybrid
element method is to compute directly the E.R.R. from the S.I.F. (for details of the computation, see
[3]). For the cracked cylindrical shell, the corresponding distribution of the Energy Release Rate is
given in Figure 8. For the event of concentrated loads (Figure 8, (b)), the curve of E.R.R.
distribution exhibits significant values only near the force location, but it is sharper than the
corresponding one for the KI. This results from the fact that the E.R.R. is proportional to the square
of the Stress Intensity Factors.
Figure 8: Energy Release Rate distribution along the crack front

In Figure 9, the values of the E.R.R. are plotted with respect to a geometric parameter without
dimension, (Ld + Lu) / r, growing with the crack length. It can be observed that the E.R.R. is
monotonic increasing and quickly tends to a saturation value, what shows the good influence of the
curvature. Although the quantitative comparison with the results of Mnch and Ousset [11] is not
possible because they did not give the values of the imposed loads, the qualitative agreement is very
satisfactory.
( Ld + Lu ) / r
Figure 9: Influence of the crack length on the Energy Release Rate
4.2 Junction with insert in a glass/epoxy sandwich plate

The sandwich plate is fabricated by linking two laminates to a thick foam core. For the second
computation step, we used a mesh composed of 2872 cubic 8-nodes finite elements (Figure10). The
176 singular elements occupy the regions S+ and S-, respectively around the two junction lines +
and -. The laminates have orthotropic elastic properties given by: EL = ET = 17.61 GPa, ET = 3
GPa, GLT = 1.905 GPa, GTT = GLT = 1.25 GPa, LT = 0.08, , TT =LT = 0.2. The core is made up
of polyvinylchloride foam with isotropic properties: E = 70.10-3 GPa and = 0.4. The junction is a
cylinder of aluminium with E = 71 GPa and = 0.34. The superior laminates cover the insert.
In Figure 11, the values of the singularity exponents computed at the first computation step are
given with respect to the angle parameter along the junction line, through the real and imaginary
parts. The computation provides one real value of and two complex conjugated ones. The
51
imaginary part reveals the oscillating nature of the singularity in a very small region closed to the
junction line. The exponent distribution () has an expected periodicity of / 2, accounting to the
symmetries of the material. The important fact to emphasise is the strong character of the
singularity, with a real part of about -0.45, closed to the value -0.5 for a delamination crack.
S+
S_
+
_
Figure 10: Discretization of the insert/sandwich plate junction
(a) Real part
(b) Imaginary part
Figure 11: Singularity exponent versus angle parameter along the junction line -.
In Figure 12, the distribution of the Stress Intensity Factors is given with respect to the angular
parameter along the inferior junction line -. The S.I.F. K1 is associated to the real exponent while
K2 + i K3 corresponds to the complex exponents. The normalization of the S.I.F. is performed by the
numerical code, according to:
max 1i , j 3 sup r ij0 (r , )
r =1
)= 1.
52
Figure 12: Distribution of the Stress Intensity Factors along the junction line.
Conclusions
The presence of the singularity around the junction lines embrittles the composite structure because
it increases the probability of crack initiation from a small defect in these localized zones. We think
the accurate knowledge of the singular field provided by the values of the singularity exponents and
the S.I.F. should give a better understanding of the collapse of junctions in composite structures.
Our aim now is using this numerical code for a better evaluation of toughness and critical E.R.R. in
delamination tests.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
J.E. Akin, The generation of element with singularities, Int. Journal for Numerical Methods in Engineering,
10, 1249-1259, 1976.
D. Crpin, M. Pyrz and Gry de Saxc, Elments finis mixtes pour le calcul des surcontraintes dans les
assemblages de composites, Revue Europenne des lments finis,11 (2-4), 319-333, 2002.
D. Crpin, Elments finis mixtes singuliers tridimensionnels pour ltude des surcontraintes dans les structures
en matriaux composites, Thse de doctorat, Universit de Lille 1, 2001.
P. Destuynder, Modlisation mcanique des milieux continus, Math. et Applications, Ellipse, Paris, 1991.
M.E. Gurtin, Encyclopedia of Physics, VIa/2 Mechanics of Solids II, S. Flgge Ed., Springer-verlag, Berlin,
1972.
C.G. Hwang, P.A. Wawrzynek, A.K. Tayebi, A.R. Ingraffea, On the virtual crack extension method for the
calculation of the rate of energy release rate, Int. J. Engineering Fracture Mechanics, 59(4), 521-542, 1998.
T.W. Kim and S. Im, Boundary layers in wedges of laminated composite strips under generalized plane
deformation - part1: asymptotic solutions, part2: numerical solutions, International Journal Solids and
Structures, 32, 609-645, 1995.
D. Leguillon and E. Sanchez-Palencia, Computation of singular solutions in elliptic problems and elasticity,
Collection Recherche en Mathmatiques Appliques, Masson, Paris, 1987.
D. Leguillon and E. Sanchez Palencia, Solutions locales en lasticit et effets de concentration de contraintes,
Calcul des Structures et Intelligence Artificielle, J.M. Fouet, P. Ladevze and Ohayon R. Eds., Pluralis, 121138, 1990.
S.G. Lekhnitskii, Theory of elasticity of an anisotropic elastic body, Holden Day, 1963.
A. Mnch and Y. Ousset, Propagation du dlaminage dans les coques composites stratifies, Actes du
cinquime colloque national en calcul des structures, Giens (France), 2, 717-724, 2001.
Nguyen-Dang Hung, On the monotony and the convergence of a special class of hybrid finite elements: the
mongrel elements, Variational Methods in the Mechanics of Solids, S. Nemat-Nasser Ed., Pergamon, 1978.
Nguyen-Dang Hung, P. Detroux, P. Falla and G. Fonder, Implementation of the duality in the finite element
analysis of shells: a mixed-mongrel planar shell element, proceedings of the World Congress on Shell and
Spatial Structures (Madrid), 2, 1979.
53

[14]
[15]
[16]
[17]
[18]
[19]
[20]
Nguyen-Dang Hung, Gry de Saxc, K. Chi-Hang, The computation of 2-D stress-intensity factors using
hybrid mongrel displacement finite elements, Int. J. Computers and Structures, 38(2-3), 197-205, 1991.
Nguyen-Tien Duong, Nguyen-Dang Hung, regular and singular metis finite element models for delamination
in composite laminates, Finite Elements in Analysis and Design, 42, 650-659, 2006.
Gry de Saxc, K. Chi-Hang, Application of the hybrid mongrel displacement finite element method to the
computation of stress intensity factors in anisotropic materials , Int. J. Engineering Fracture Mechanics,, 41(1),
71-83, 1992.
Gry de Saxc, K. Chi-Hang, Computation of stress intensity factors for plate bending problem in fracture
mechanics by hybrid mongrel finite element, Int. J. Computers and Structures, 42(4), 581-589, 1992.
S.S. Wang, Edge delamination in angle-ply composite laminates, AIAA Journal, 22(2), 256-265, 1981.
S.S. Wang, F.G. Yuan, A hybrid finite element approach to composite laminate elasticity problems with
singularities, Journal of Applied Mechanics, 50, 835-844, 1983.
K. Washizu, Variational methods in elasticity and plasticity, Pergamon, 1982.
54
Image-based Micromechanics Analysis using Level Sets

and the Extended Finite Elements Method
Ionescu, Nicolas Mos, P. Cartraud and M. Bringhier
GM Institut de Recherches en Gnie Civil et Mchanique UMR CNRS 6183
Ecole Centrale de Nantes, 1 Rue de la Noe, 44321 Nantes, France
Abstract
The advances in material characterization by means of imaging techniques require powerful
computational methods for numerical analyses. The present paper focuses on the advantages of
coupling the X-FEM and level sets applied to solve microstructures with complex geometries,
extending previous work. The process of obtaining the level set information corresponding to a
digital image of a material structure and using it in setting up and solving an X-FEM problem is
presented and the method is validated using a 2D reference example: finding the effective properties
of a metallic-ceramic composite using the homogenization technique.
Keywords: level sets, X-FEM, homogenization.
Introduction
With the emergence of novel ways of characterizing complex material microstructures,

computational methods to correspond in versatility are developed and evolve. Digital
representations of morphologies of heterogeneous media are used by various numerical methods for
calculating their mechanical or thermal properties, their fracture behavior.
Depending on the nature of the structure of interest, material data is obtained using laser
confocal microscopy [1], (micro)computed tomography (CT) [2-4], scanning electron microscopy
(SEM) [5], (micro) magnetic resonance imaging (MRI) [6]. Visual digital representations of the
data of interest are processed to obtain geometrical information that can be used by a discretization
algorithm, such that a boundary value problem can be solved. There are numerous algorithms that
serve to solving the task of segmenting an image: edge detection, region growing, segmentation
based on watersheds, level set segmentation [7,8], to name a few.
Once the domain of the problem is set and the boundary conditions specified, various numerical
methods can be employed to solve the specific problem. The task that rises is how to handle the
geometrical information in a numerical discretization scheme, with minimal loss of information
through approximations. If the Finite Elements (FE) approach is the numerical tool chosen, the
mesh associated to the geometry is essential in obtaining valid results, be that a conform mesh or
not. Ideally, a mesh does not depend on the way geometry is represented and on the geometry
itself.
The paper presents the advantages of using a numerical method that couples the level set method
[9] with the extended finite element method (X-FEM) [10,11] applied to solving problems of
micromechanics. The use of level sets techniques in the context of image segmentation leads to a
precise representation of the geometry, the advantage of this technique being that changes in
topology are handled implicitly. Designed to solve fracture mechanics problems [11,12] and
applied to analyze complex heterogeneous structures [13,14], the X-FEM formulation does not
require a conforming mesh. Rather, a regular mesh is constructed independent of the geometry and
the physical boundaries read into the X-FEM algorithm in the level set form such that the elements
55
intersected by the physical interface(s) are enriched. The capabilities of the coupling between the
X-FEM and the level set method have been shown in [14]; it has been proven that the method gives
the same convergence rate as conform FE without any meshing problems. However, in [13] the
applications were restricted to microstructures with material interfaces that can be described by
analytical level sets. This work advances the method one step further, presenting the process of
obtaining the level set information corresponding to a digital image of a material structure and using
it in setting up and solving an X-FEM problem, as illustrated in Fig. 1. The applications on hand
are two-dimensional, but the method may be applied as well to three- dimensional problems. One
of the advantages of the approach is that, given a complex image of a high resolution, the level set
segmentation can be resolved at pixel level. In turn, the X-FEM element size is user prescribed
enabling computations on sensible sized meshes that makes use of the pixel based geometrical
information provided by the level set.
The road is set for analyzing problems that arise in working with microstructures: determining
their mechanical response or employing homogenization methods [1,2,15,16].
Data
collection
Segmentation (ITK)
X-FEM meshing
X-FEM
analysis
200 m
a) digital image
b) level set pixel based

representation
c) X-FEM mesh with level set

data embedded (detail)
Fig. 1: Numerical algorithm: data (reproduction of SEM sample of metallic composite - 24%
TiC/stellite) is digitized and segmented using a level set threshold based segmentation filter and
the X-FEM mesh is constructed incorporating the level set data in the enriched elements.
Materials and methods
Level set segmentation

The level set method is a numerical technique formulated to address the evolution of interfaces:
the curve representing the interface is built into a surface called the level set function that tracks
the moving interface [8]. The main advantage of this technique is that changes in the interface
topology are handled implicitly, making it suitable for a large range of applications.
Level sets are used with success in image segmentation tasks [7,8]: a user initialized contour is
evolved based on image features such as grayscale intensity, gradient magnitude or image edges.
The equation governing the motion of a level set (x,t) is:
d
= a + b v ,
dt
(1)
where the scalar a adjusts the self-generated outward motion (propagation term), b scales the
56
motion by mean curvature of the front, and the last term represents a rigid body motion given by
the vector v. If the segmentation is based on the intensity of the image, the range of intensity values
that classify the structure of interest must be declared and the propagation term in the level set
equation defined based on this threshold. If we consider L as the lower threshold value and U the
upper threshold value for the desired range of intensity values, the propagation term in the level set
can be defined as:
g(x) L g(x) < 1 (U L) + L
2
a(x) =
,
U g(x) otherwise
(2)
where g(x) represents the input image and weights the influence of the propagation term.
The present application uses the Insight Toolkit (ITK) [7] for building the image segmentation
application tailored to the specific needs of treating images of material microstructures. For each
image the user has to specify the threshold of intensity values and an initial level set model
(x,t=0). The level set solution is calculated to subpixel precision [7], and interpolated at pixel
level. The processing time is influenced by the surface area of the evolving front and the distance
the front has to travel, so the initial conditions and convergence parameters must be chosen
carefully. The output level set image constitutes the input geometry for the X-FEM problem. Fig.
1 presents schematically the path from an acquired image to the input for an X-FEM analysis.
The X-FEM
The X-FEM is an extension of the finite element (FE) method that was developed from the need
to improve the FE approach of problems with complex geometries [10-14]. As mentioned, in
contrast to the classical formulation, X-FEM does not require the mesh to conform the geometry,
instead a regular mesh is constructed for the domain of interest and the presence of internal
boundaries (cracks, inclusions) is weighted in the formulation of the finite elements at the
corresponding locations [13].
In X-FEM, as for any problem to be solved by a numerical discretization scheme, an X-FEM
mesh needs to be set, the boundary conditions applied and the solution computed. Once the level
set information is obtained from a digital image of interest, the information is fed to the X-FEM
algorithm.
Any digital image is an implicit grid, each pixel representing the unit in terms of geometry and
information content. The level set information corresponding to an image constructed by image
segmentation is as accurate as the original image data is provided at pixel level. But the pixel
scale may not be relevant for modeling the problem. Naturally, a mesh based on the pixel size can
be constructed but neither its dimension nor its content is an advantage. On one hand, very large
meshes are cumbersome and computationally expensive. On the other hand, this approach does not
efficiently use the benefit of a non-conforming mesh, brought about by X-FEM. So the task to be
solved is the construction of a nonconforming mesh whose scale is not dictated by the pixel size in
the original image, but which makes use of the pixel based level set information. The answer to
such a problem is given by the adaptive meshes. Our on-going effort is aimed at implementing
adaptive gridding techniques such as the octree algorithm [17] into the X-FEM. On that path, we
consider de-refined regular meshes, starting from the reference pixel based mesh. A uniform grid
(Fig. 1c) is superimposed on the geometry, the pixel based level set information is interpolated to
the grid nodes of the mesh, and enrichment is applied to the elements that intersect the zero contour
of the level set. The changes in geometry with the degree of refinement of the mesh are illustrated
in Fig. 2 (for clarity the mesh outline was omitted). As expected, smooth boundaries become
jagged for low mesh resolution and geometrical detail is lost (Fig. 2e, f), but for a sensible mesh
resolution the geometry is well characterized (Fig. 2c, d).
57

a)
c)
b)
e)
d)
f)
Fig. 2: a) Two phase representation using the level set sign in a pixel based X-fem mesh.
Delimited area in (a) enlarged to illustrate the changes in the morphology of the level set
with the decrease of the mesh refinement: b) 22562 elements, c) 21282 elements, d) 2642
elements, e) 2322 elements, f) 2162 elements.
Application
To validate the method we consider a reference example, documented in the literature: the
effective composite properties for a sample of a ceramic-metallic composite 24 % TiC/stellite
obtained using the homogenization method and FE have been reported in [18]. The material
properties of the two isotropic phases of are given in
Table 1.
Table 1: Material properties for TiC and stellite.
Properties of phases:
Stellite (nickel-based alloy)
TiC (carbide)
Effective properties[18]
Youngs Modulus (GPa)

183
447
~226
Poissons Ratio
0.3
0.19
-
The sample is reproduced in Fig. 1a, the reference resolution of the grayscale image considered
in this work being 256x256 pixels. First, the level set data is obtained for this resolution, using a
threshold level set filter based on the image intensity (lower threshold 100, upper threshold 255)
and can be seen in Fig. 1b. Then, the X-FEM mesh is constructed, embedding the level set
information. We consider the following mesh resolutions: 22562, 21282, 2642, 2322, 2162
triangular elements.
Since the material studied can not be considered as a periodic medium, its effective properties
are computed using an homogenization scheme with homogeneous boundary conditions [19].
Kinematic boundary conditions are applied here, so the micromechanics problem to solve is: find
the displacements u, strains , and stresses on the domain such that:
div = 0, = a:, = 1 2 (u + u T )
u = Ex, x
(3)
For a given macroscopic strain E, the effective properties are then obtained from the average values
of the strains which can be computed from the energy values. The calculations are carried out on a
single sample that considered as representative volume element (RVE) [20]; the computational
determination of the appropriate size of the RVE is subject for future work.
58
Results
The reported effective value for the Youngs modulus of the TiC/stellite composite is
approximately 226 GPa [18]. The results obtained in this study, for the different meshes considered
are listed in Table 2.
Table 2. Effective material properties for the TiC/stellite alloy for various mesh resolutions.
Mesh size (# of elements)
Poisson's ratio
Young's Modulus
2162
0.273
234.515
2322
2642
21282 22562
0.275
0.276
0.276
0.277
232.468 233.588 233.518 232.823
The relative error between the results obtained for the coarsest mesh and the pixel based mesh is
0.7 %, so even coarser meshes approximate accurately the global effective properties of the
composite. The Youngs modulus found values are at best 2.7 % away from the previously
reported value, listed above. The stiffness decrease induced by the mesh refinement is counteracted
by the geometrical changes introduced by the embedding of the level set data in the mesh the
coarser the mesh the less geometrical details are captured therefore the overall domain occupied by
the hardphase decreases. The geometrical differences between the present model and the unit-cell
based FE model used in [18] may account for the small difference in the overall stiffness results.
The distributions of the micro-stresses are in agreement in all the simulations considered (Fig.
3). However, if the micromechanical fields of the structure are of interest, an appropriately refined
mesh is needed in order to capture the discontinuities in field values at the interface between
materials. As seen in Fig. 3a, the mesh coarseness gives quite a fragmented map of the Von Mises
stress. In contrast, a satisfactory stress distribution is obtained using a mesh of 21282 elements
(Fig. 3c).
a)
b)
c)
Von Mises (MPa)
270
Fig. 3. Von Mises stresses corresponding to the shearing micro-mode for different mesh sizes: a)
2162, b) 2322, c) 21282 elements.
Discussion and conclusions
The present work describes how, starting from a digital image, a numerical simulation is set-up
using level sets and the X-FEM. The method offers a direct way to treat problems of complex
microstructures taking advantage of the versatility of level sets and the non-conforming meshing
algorithm of X-FEM. The method can be applied equally well to two- or three- dimensional
imaging data. A reference example is used for validation and the results obtained are encouraging.
59
As the method is applied to treating microstructures with complex geometries, the accurate
representation of the geometry is compulsory and it is ensured in the image segmentation
preprocessing task by the use of level sets. However detail may be lost during meshing. As seen
from the example presented in this paper, the mesh density alters the way the geometry is embedded
in the X-FEM, and therefore the obtained results. A refined mesh is needed closer to the interface
between different phases of the structure, but away from the interface the results are accurate even
for coarser mesh. Our ongoing effort is directed towards implementing an adaptive octree mesh
algorithm, such that the element size will be dictated by the distance to the material interface.
In summary, this study demonstrates the feasibility of using level sets and the X-FEM for
treating problems of micromechanics. It is worth mentioning that the present framework may be
extended beyond the area of linear elasticity to solving hyperelasticity and elasto-plasticity
problems.
Acknowledgements
The first author gratefully acknowledges the financial support from the French Ministre
dlgu lEnseignement suprieur et la Recherche.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Kanit, T., N'guyen, F., Forest, S., Jeulin, D., Reed, M., and Singleton, S., 2006, "Apparent and effective
physical properties of heterogeneous materials: Representativity of samples of two materials from food
industry", Computer Methods in Applied Mechanics and Engineering, 195, pp. 3960-3982.
Kawagai, M., Sando, A., and Takano, N., 2006, "Image-based multi-scale modelling strategy for complex and
heterogeneous porous microstructures by mesh superposition method", Modelling and Simulation in Materials
Science and Engineering, 14, pp. 53-69.
Babin, P., Valle, G., Dendievel, R., Lassoued, N., and Salvo, L., 2005, "Mechanical properties of bread crumbs
from tomography based Finite Element simulations", Journal of Materials Science, 40, pp. 5867-5873.
Ghosh, S., Krishnan, G., and Dyce, J., 2006, "Image-based macro-micro finite element models of a canine
femur with implant design implications", International Journal of Computational Methods in Engineering
Science and Mechanics, 7, pp. 155-171.
Michlik, P. and Berndt, C., 2006, "Image-based extended finite element modeling of thermal barrier coatings",
Surface and Coatings Technology, In Press, Corrected Proof.
Genant, H. and Jiang, Y., 2006, "Advanced imaging assessment of bone quality", Annals of the New York
Academy of Sciences, 1068, pp. 410-428.
Ibanez L., Schroeder W., Ng L., and J., C., 2005, The ITK Software Guide, 2005 ed. Kitware, Inc.
Sethian, J. A., 1999, Level Set Methods: Evolving Interfaces in Geometry, Fluid Mechanics, Computer Vision
and Materials Sciences, 2nd ed. Cambridge University Press.
Osher, S. J. and Fedkiw, R. P., 2003, Level set methods and dynamic implicit surfaces, vol. 153. Springer, New
York.
Belytschko, T., Parimi, C., Mos, N., Usui, S., and Sukumar, N., 2003, "Structured extended finite element
methods for solids defined by implicit surfaces", International Journal for Numerical Methods in Engineering,
56, pp. 609-635.
Mos, N., Dolbow, J., and Belytschko, T., 1999, "Finite element method for crack growth without remeshing",
International Journal for Numerical Methods in Engineering, 46, pp. 131-150.
Daux, C., Mos, N., Dolbow, J., Sukumar, N., and Belytschko, T., 2000, "Arbitrary branched and intersecting
cracks with the extended finite element method", International Journal for Numerical Methods in Engineering,
48, pp. 1741-1760.
Mos, N., Cloirec, M., Cartraud, P., and Remacle, J.-F., 2003, "A computational approach to handle complex
microstructure geometries", Computer Methods in Applied Mechanics and Engineering, 192, pp. 3163-3177.
Sukumar, N., Chopp, D. L., Mos, N., and Belytschko, T., 2001, "Modeling holes and inclusions by level sets
in the extended finite-element method", Computer Methods in Applied Mechanics and Engineering, 190, pp.
6183-6200.
Kumar, H., Briant, C. L., and Curtin, W. A., 2006, "Using microstructure reconstruction to model mechanical
behavior in complex microstructures", Mechanics of Materials - Advances in Disordered Materials, 38, pp.
60

[16]
[17]
[18]
[19]
[20]
818-832.
Takano, N., Zako, M., Kubo, F., and Kimura, K., 2003, "Microstructure-based stress analysis and evaluation
for porous ceramics by homogenization method with digital image-based modeling", International Journal of
Solids and Structures, 40, pp. 1225-1242.
Zhang, Y. and Bajaj, C., 2006, "Adaptive and quality quadrilateral/hexahedral meshing from volumetric data",
Computer Methods in Applied Mechanics and Engineering, 195, pp. 942-960.
Golanski, D., Terada, K., and Kikuchi, N., 1997, "Macro and micro scale modeling of thermal residual stresses
in metal matrix composite surface layers by the homogenization method", Computational Mechanics, 19, pp.
188-202.
Suquet, P. M., 1987, "Elements of Homogenization for Inelastic Solid Mechanics", in Lecture Notes in
Physics, vol. Homogenization techniques for composite media : lectures delivered at the CISM International
Center for Mechanical Sciences, Udine, Italy, July 1-5, 1985, Lecture Notes in Physics, Sanchez-Palencia E.
and A. Zaoui, Eds. Berlin ; New York : Springer-Verlag, pp. 194-275.
Kanit, T., Forest, S., Galliet, I., Mounoury, V., and Jeulin, D., 2003, "Determination of the size of the
representative volume element for random composites: statistical and numerical approach", International
Journal of Solids and Structures, 40, pp. 3647-3679.
61
Smooth Strain Finite Elements: Selective Integration

Nguyen-Xuan Hung , Stephane Bordas+, and Nguyen-Dang Hung
Division of Computational Mechanics, Department of Mathematics and Informatics,
University of Natural Sciences-VNU-HCM, Vietnam
+
University of Glasgow, department of Civil Engineering, Rankine Building, G12 8LT, UK
LTAS, Division of Fracture Mechanics, University of Liège, Belgium
corresponding author: stephane.bordas@alumni.northwestern.edu
Abstract
This paper furthers the developments of a novel family of finite elements with smooth strains,
offering remarkable properties. In the smooth strain finite element method, elements are divided into subcells. The strain at a point is written as a weighted average of the standard strain
field over the element. This yields superconvergent stresses, both in regular and singular settings, as well as increased accuracy, with slightly lower computational cost than standard FEM.
This superconvergence is attributed to the mathematical equivalence of the method with a quasiequilibrium element. The one-subcell version does not exhibit volumetric locking yields more
accurate stresses but less accurate displacements and is equivalent to a quasi-equilibrium FEM.
In the limit where the number of subcells goes to infinity, the standard FEM is recovered which
yields more accurate displacements and less accurate stresses. We show that writing the volumetric part of the strain field as dictated by the one-subcell version is sufficient to get rid of
volumetric locking and increase the displacement accuracy compared to the single subcell version. Numerical examples on various compressible and incompressible linear elastic test cases
show that high accuracy is retained compared with the standard FEM without increasing computational cost.
Keywords: finite element method, non-local strain, smoothed strains, volumetric locking, stabilized conforming nodal integration.
1 Introduction
The strain smoothing method was first proposed by Chen et al. [4] to overcome some of the
drawbacks of mesh-free methods [5] such as the vanishing derivatives of shape functions at the
nodes and the violation of an integration constraint (IC) in the Galerkin methods. Difficulties
were first realized in the context of mesh-free methods employing Gauss integration since these
methods have to rely on background cells for integration of the weak form. To avoid this, nodal
integration in mesh-free methods was proposed by Beissel and Belytschko [9] and Bonet and
Kulasegaram [10] with the aim to eliminate background cells. However, direct nodal integration
leads to numerical instability and reduced convergence rates. In 2001, Chen et al. [4] showed
that the vanishing derivatives of shape functions at the nodes are the cause of the instability and
the violation of an integration constraint (IC) in the Galerkin method is responsible for the reduced convergence rates. Linear consistent shape functions that can be obtained from moving
62
least squares approximations does not guarantee linear exactness of a mesh-free method based
on a Galerkin weak form, such as the element-free Galerkin method [5]. To satisfy this linear
exactness, Chen and his co-workers proposed a stabilized conforming nodal integration using a
strain smoothing method (SSM) for conferring Galerkin mesh-free methods with stability, consistency, higher efficiency, accuracy and convergence properties. The stabilized nodal integration
was then extended by Yoo et al. [6] to another Galerkin method with a different discretization
technique: the natural element method, where the method was shown to yield impressive results
for highly incompressible materials, without any modification of the integration scheme.
In mesh-free methods using stabilized nodal integration, the entire domain is discretized into
cells for integration purposes. These cells may be the cells of a Vorono diagram [4, 6], and the
integration performed along the edges of each cell. Although mesh-free methods have obtained
good accuracy and high convergence rates, the usually complex and often non-polynomial approximations increase the computational cost. These remarks can lead to thinking that stabilized
conforming nodal integration could be efficiently extended to the finite element world, especially
since NEM and FEM are quite close.
The idea of smooth strain finite elements is to divide standard finite elements into smoothing
cells. The strain field at any point in an element is written as a weighted spatial average over
the element. For piecewise-constant weight functions1 , contour integration along the boundaries
of the smoothing cells is recovered. In the case of one single cell, the elemental stiffness matrix is obtained through integration over the boundary of the finite element. The authors build
in Hung et al. [12], as a complement to the work of G. R. Liu [8], the theoretical foundations
for stabilized conforming nodal integration in FEM2 and carry out detailed convergence studies, showing the H1-superconvergence of the one-cell version of the method. Additionally, it
is shown that the one-subcell element handles incompressibility without volumetric locking. In
Hung et al. [14], we prove the equivalence of the one subcell version of the method with a quasiequilibrium element, thereby justifying the H1-superconvergence properties observed in [12].
In Reference [12], however, it is shown that, if more than one smoothing cell is used, stabilized
conforming nodal integration still yields finite elements that lock in nearly incompressible cases.
This is due to the fact that as the number of subcells increases, the SSFEM discrete space
tends toward the standard displacement FEM discrete space, which is subject to locking. On the
other hand, the single subcell SSFEM is mathematically formally identical to a quasi-equilibrium
FEM, which does not lock. While the latter finite element method yields superconvergent dual
quantities (stresses), it leads to lower displacement accuracy when compared to the displacementbased FEM.
Therefore, we are dealing with a balance between stress versus displacement accuracy and, at
the same time, between non-locking and locking spaces. The displacement-based FEM and the
multi-subcell SSFEM provide reliable displacement (primal) accuracy in compressible settings,
but suffer from Poisson locking. On the contrary, the quasi-equilibrium FEM and the one-subcell
SSFEM do not lock and yield accurate stresses (dual variable), but suffer from a loss in displacement (primal) accuracy.
The intent of this paper is to show that writing the volumetric part of the strains with a single
1
2
constant and equal to 1 on one smoothing cell, zero elsewhere

smooth strain finite elements
63
subcell formulation and expressing the deviatoric part with multiple subcell smoothing is enough
to provide a method which does not lock, but retains both displacement and stress accuracy. The
idea comes from the FEM, where many special treatments exist for solving problems of incompressible elasticity: mixed finite element methods [13, 16, 17], reduced and selective integration
method (SIM) [3, 15], etc.
The outline of the paper is organized as follows. In the next section we present the basic
equations of linear elasticity, the weak form and the finite element discretization. The strain
smoothing stabilization and the finite element discretization using the SSM are introduced in
Section 3. Several numerical examples are given in Section 4. Section 5 closes some conclusions and future works.
2 Governing equations and weak form

Let R2 be a bounded domain with a Lipschitz-continuous boundary . A body force b
is acting within the domain. The governing equilibrium equation for isotropic linear elasticity
writes
+ b = 0 in
(1)
where is the symmetric Cauchy stress tensor. The compatibility equation is
1
= (u + T u)
2
in
(2)
The displacement field satisfies the Dirichlet boundary conditions
u=u
on
(3)
and the stress field satisfies the Neumann boundary conditions

n = t on
(4)
where = , = u t , u t =
The stress field is split into two parts: the deviatoric stress s, and the pressure p.
= s + p1 = Ddev + K u1
(5)
where 1 is the rank two identity tensor, which can presented by 1 = [1 1 0]T , Ddev is the
deviatoric projection of the elastic matrix D, is the shear modulus and K is the bulk modulus
defined by K = E/3(1 2), E is Youngs modulus, and is Poissons ratio.
The principal of virtual work writes
Z
Z
Z
Z
t ud
: Ddev : d + K : ( u1)d =
b ud +
(6)
Assume that the bounded domain is discretized into ne non-overlapping elements, h =

ne
S
e . The standard finite element solution uh of a finite element displacement model is exe=1
pressed as follows
h
u =
np
X
Ni qi = Nq
(7)
i=1
64
where np is total number of nodes in the mesh, Ni is the shape function of degree p of node i, qi
is the associated degree of freedom of that node. The discrete strain field is
h = s uh = Bq
(8)
where B = s N is the strain - nodal displacement matrix (the discretized, symmetric gradient
operator).
By substituting Equation (7)-Equation (8) into Equation (6), we obtain a linear system for q
Kq = g
(9)
where the element stiffness matrix given by

Z
Z
T
B Ddev Bd + K
K=
e
with
Ddev
(10)
4 2 0
1
= 2 4 0 ,
3
0
0 3
Z
and the load vector is
BT D Bd
e
1 1 0
D = 1 1 0
0 0 0
(11)
Z
NT td
g=
N bd +
e
(12)
As the material nears incompressibility, Equation (9) leads to poor accuracy, and this is
known as volumetric (or Poissons) locking. One method to overcome this drawback is selective
integration (SIM) [3]. With the four node quadrilateral element, the idea is to employ a 2 2
Gauss quadrature rule for the deviatoric term and a rule for the volumetric term.
In mesh-free methods based on nodal integration, the convergence of the solution with linear
consistency requires the following integration constraint (IC) to be satisfied (Chen et al., [4])
Z
Z
T
Bi (x)d =
Ni (x)nT d
(13)
h
where Bi is the standard gradient matrix and n is the normal matrix in the form
nx 0
Ni,x 0
Bi = 0 Ni,y , nT = 0 ny
ny nx
Ni,y Ni,x
(14)
The integration constraint emanates from the equilibrium of the internal and external forces of
linear complete Galerkin approximations (Chen et al. [4, 6]). This is similar to the linear consistency that is verified by the constant stress patch test in FEM.
Associating the conventional FEM and strain smoothing (or stabilized conforming nodal integration) developed for mesh-free nodal integration, Liu et al. [8] proposed a smooth strain finite
element, further developed and tested by Hung et al. [12, 13]. In such a technique, elements
exist, similarly to the standard FEM, but may be of arbitrary polygonal shape. The weak form is
developed through a mixed variational principle based on an assumed strain field [7] and the integration is carried out over smoothing cellz s (sub-domains, or subcells) forming a partition of the
65
elements. A strain smoothing (weighted averaging) is performed over the element, to regularize
the local strain entering into the elemental stiffness matrix. If the weight function is constant over
each smoothing cell, and vanishes in the rest of the element, the surface integration over each
smoothing cell becomes a line integration along its boundaries. In the case of a constant weighting function, it is unnecessary to calculate the shape functions gradient on the element interior,
and a 1D Gauss integration along the edges of each cell is sufficient. It is consequently possible
to choose a four noded element with an angle greater than 180 degrees. The strain smoothing
method is explained in more detail in the following section.
3 The strain smoothing method

The strain smoothing method was proposed by Chen et al. [4] as a normalization of the local
strain field. This technique is also known as strain smoothing stabilization, through which the
nodal strain is computed through the divergence of a spatial average of the standard local strain
field. In mesh-free methods, this is sufficient to eliminate defective modes, since through strain
smoothing, the derivatives of the shape functions are not required at the nodes. Applications of
strain smoothing to the FEM can be seen as a stabilized conforming nodal integration method as
defined in Galerkin mesh-free methods. Strain smoothing at an arbitrary point writes
Z
h
ij (xC ) =
hij (x)(x xC )d
(15)
e
where is a smoothing function.There are several choices for this smoothing function. For
simplicity, is assumed to be a step function as follows
1/Ae , x e
xc e (x xC ) =
(16)
0, x
/ e
where Ae is the area of the smoothing cell, eC e , as shown in Figure 1.
Figure 1: Example of finite element meshes and smoothing cells

Substituting Equation (16) into Equation (15), and using the divergence theorem, we obtain
66
hij (xC )
1
=
2AC
Z
eC
uhj
uhi
+
xj
xi
1
d =
2AC
Z
eC
(uhi nj + uhj ni )d
(17)
Now, we consider an arbitrary smoothing cell, C h illustrated in Figure 1 with boundary

nb
S
bC , where bC is the boundary segments of eC , and nb is the total number of edges
eC =
b=1
of each smoothing cell. The relationship between the strain field and nodal displacement is
corrected by replacing B by B:
h = Bq
(18)
The elemental stiffness matrix then writes

Z
Z
T
K=
B Ddev Bd + K
e
T D Bd
(19)
Here, the integrands are constant over each eC and the non-local strain displacement matrix is
in the form
Z
1
(20)
Ni nT d
Bi (xC ) = e
AC eC
From Equation (20), we can use Gaussian line integration along each segment of bC . If the shape
functions are linear along the boundaries of the smoothing cells, one Gauss point is sufficient for
exact integration of the weak form. In this case,
nb
1 X
T C
Ni (xG
Bi (xC ) = e
b )n lb
AC b=1
(21)
C
b
where xG
b and lb are the midpoint (Gau point) and the length of C , respectively.
Considering a mixed variational principle based on an assumed strain field [7], the following
system of linear algebraic equations is obtained
=g
Kq
(22)
Strain smoothing leads to high flexibility such as arbitrary polygonal elements, and a slight
computational cost reduction. The element is subdivided into nc non-overlapping sub-domains
(smoothing cells). Figure 2 is the example of such a division with nc = 1, 2, 3 and 4 corresponding to 1-subcell, 2-subcell, 3-subcell and 4-subcell elements. Then the strain is smoothed
over each subcell. While choosing a single subcell yields an element which is superconvergent
in the H1 norm, and insensitive to volumetric locking, as shown in Reference [12], if nc > 1,
locking reappears. It is also shown in Reference [12] that the finite element method with strain
smoothing is equivalent to a stress (equilibrium) formulated element [1] for nc = 1, and tends
toward the standard displacement solution for nc . Consequently, as approaches 1, the
stress results become more accurate, while the displacement results deteriorate; and as nc increases, displacement results gradually improve, while stress results deteriorate. We present in
this article a method in which an arbitrarily high number of smoothing cells can be used to write
the volumetric part of the strain tensor, while the deviatoric strains are written in terms of a single
subcell smoothing. The method may be coined a stabilized method with selective cell-wise strain
smoothing.
The stiffness matrix is built
67
Using nc > 1 subcells to evaluate the deviatoric term

Using one single subcell to calculate the volumetric term
This leads to the following elemental stiffness matrix
=
K
nc
X
T Ddev Bc Ac + K B
T D BA
e
B
c
(23)
c=1
where Ac is the area of the smoothing cell, C .

The resulting approach with selective cell-wise integration yields stable and optimal convergence for compressible and nearly incompressible cases with isotropic linear elasticity. Additionally, the proposed method can extend well to inelastic materials such as isotropic plasticity and
viscoplasticity. Results in plasticity and incompressible elasticity are shown in Reference [13].
Figure 2: Division of an element into smoothing cells (nc)
4 Numerical results
4.1 Cantilever beam
A 2-D cantilever beam subjected to parabolic loading at the free end is examined in this example
as shown in Figure 3. The geometry is taken as length L, height D and thickness t: L = 8, D = 8
and t = 1. The material properties are: Youngs modulus E = 3 107 and the amplitude of
the parabolic shear force P = 250. The exact solution of this problem is available as given by
Reference [11]. Figure 4 illustrates a uniform mesh with 512 quadrilateral elements.
68
Figure 3: A cantilever beam and boundary conditions

The relative error in displacement norm is defined as
v
undof
uX
X
ndof
t
h
exact 2
ui ui
/
(uexact
)2 100
Red =
i
i=1
(24)
i=1
The error in energy is defined by
u u h =
e
1/2
( ) : D : ( )d
(25)
Under plane stress conditions, Poissons ratio = 0.3, Figure 5 shows the relative error and the
rate of convergence in the displacement norm for a sequence of uniform meshes.
Figure 4: Uniform mesh of 512 quadrilateral elements

From Figures 5 - 6, the presented method gives reliable results compared with 4-node FEM.
Figure 5(b) and Figure 6(b) show that the 2-Subcell, 3-Subcell and 4-Subcell elements exhibit
69
0.5
1.4
FEM
1Subcell
2Subcell
3Subcell
4Subcell
1.2
FEM
1Subcell
2Subcell
3Subcell
4Subcell
2.01
2.005
log10(Red)
0.5
R.Ed
0.8
0.6
1.996
2.002
1.5
0.4
2.001
2
0.2
0
0.5
1.5
2.5
Mesh index
3.5
2.5
1.4
4.5
1.2
0.8
0.6
log10(h)
(a)
0.4
0.2
(b)
Figure 5: The convergence of the displacements (a) relative error; (b) convergence rate.
1.5
0.0404
0.997
Exact energy
FEM
1Subcell
2Subcell
3Subcell
4Subcell
log10(Error in energy norm)
0.0402
Strain energy
0.04
0.0398
0.0396
2.5
0.997
1.5
2.5
Mesh index
(a)
3.5
4.5
FEM
1Subcell
2Subcell
3Subcell
4Subcell
1.01
3.5
1.953
0.0394
0.0392
0.5
0.998
4.5
1.4
1.2
0.8
0.6
log10(h)
0.4
0.2
(b)
Figure 6: The convergence of the energy norm; a) the energy, b) the rate
70
the same convergence rate in both the L2 and H1 (energy) norms as the standard FEM. Moreover,
displacement results for the 3-Subcell and 4-Subcell discretizations are more accurate than the
standard bilinear Q4 FEM solution. The proposed elements also produce a better approximation
of the global energy. In addition, the CPU time required for all elements with the smooth strain
FEM presented here appears asymptotically lower than that of the FEM [12], as the mesh size
tends to zero. It is remarkable that the 1-subcell enjoys a form of superconvergence in the energy
norm, the convergence rate approaching 2. A reason for this promising property is the equivalence of this 1-subcell element with a quasi-equilibrium element as shown in Reference [14].
Figure 6 shows the convergence in energy and the convergence rate for the cantilever beam
problem. Next we estimate the accuracy of the SFEM elements for the same beam problem,
assuming a near incompressible material. Under plane strain condition, Figure 7(a) illustrates
the displacements along the neutral axis for Poissons ratio, = 0.4999. The results show that
FEM, 2-subcell, 3-subcell and 4-subcell solutions yield poor accuracy as Poissons ratio tends
toward 0.5. To remedy this locking phenomenon, selective integration techniques are considered.
Figures 7(b) - 7(c) present the results using selective integration (SIM) and compares to the
displacement-Q4-FEM element.
4.2 Hollow cylinder under internal pressure

Consider a hollow cylinder (Figure 8(a)), which has an analytical solution of elastic problem
shown in [11], has internal radius a = 1, external radius b = 5, Youngs modulus E = 3x107 . A
uniform pressure p = 3x104 is applied to the inner surface (r = a), while the outer surface (r =
b) is traction free. Because of the symmetric characteristic of the problem, only one-quarter of
cylinder is modelled. In the analyses, four different nodal discretizations are considered, namely,
625 nodes, 2401 nodes, 6561 nodes, and 10000 nodes.
Figure 9(a) shows that the strain energy obtained with the proposed method agrees well with
the exact energy. The results with strain smoothing are more accurate than the FEM results. As
shown in Figure 9(b), the same rate of convergence is obtained with the displacement-Q4 element
the 2, 3 and 4-subcell elements. Moreover, the 1-subcell elements gives a superconvergence in
the energy norm. Consider the same problem with nearly incompressible materials, Figure 10(b)
and Figure 11(b) show excellent results and exhibit no volumetric locking.
4.3 Cooks Membrane

This benchmark problem in Reference [2], shown in Figure 12, refers to a clamped tapered
panel is subjected to an in-plane shearing load, resulting in deformation that is dominated by a
bending elastic response. Assuming plane strain conditions, Youngs modulus and Poissons ratio
= 0.4999 or = 0.4999999, Figure 13 plots the vertical displacement at the top right corner.
It shows that the displacement-FEM Q4 element provides poor results while the other elements
based on strain smoothing formulations are reliable, even in very incompressible materials.
5 Conclusions
In this paper, we presented a remarkably simple finite element method avoiding locking when
modelling nearly incompressible materials in elastostatics, and allowing the use of arbitrary
polygonal finite elements. This method relies on strain smoothing over a single or multiple
71
x 10
Analytical solu.
FEM
1Subcell
2Subcell
3Subcell
4Subcell
Vertical displacement v
0.5
1.5
Analytical solu.
SIM
1Subcell
2Subcell
3Subcell
4Subcell
1.5
2.5
2.5
x 10
0.5
0.5
4
x (y=0)
4
x (y=0)
(b)
(a)
x 10
Analytical solu.
SIM
1Subcell
2Subcell
3Subcell
4Subcell
0.5
1.5
2.5
4
x (y=0)
(c)
Figure 7: Vertical displacement for cantilever beam at the nodes along the x-axis (y =0) : (a)
without using the selective technique; (b) applying the selective method = 0.4999; and (c)
applying the selective method = 4999999
72
graph
x 10
4.5
5
4.5
4
3.5
3.5
3
2.5
2.5
2
1.5
1.5
1
1
0.5
0.5
0
(b)
(a)
Figure 8: Sample of discretizations of 576 quadrilateral elements and distribution of von Mises
stresses: (a) Hollow cylinder under internal pressure p; (b) von Mises stresses for 1-subcell
element
31.46
0
1.017
31.44
log10(Error in energy norm)
0.5
Strain energy
31.42
31.4
Exact energy
FEM
1Subcell
2Subcell
3Subcell
4Subcell
31.38
31.36
1.5
FEM
1Subcell
2Subcell
3Subcell
4Subcell
1.017
1.017
1.017
2
2.031
2.5
31.34
31.32
0.5
1.5
2.5
Mesh index
(a)
3.5
4.5
3
1.3
1.4
1.5
1.6
1.7
1.8
1.9
log10(Number of nodes)1/2
(b)
Figure 9: The convergence of the strain energy and convergence rate for the hollow cylinder
problem (Poissons ratio = 0.25): (a) strain energy; and (b) rate of convergence
73
x 10
x 10
0.5
Analytical solu.
SIM
1Subcell
2Subcell
3Subcell
4Subcell
0.5
Analytical solu.
FEM
1Subcell
2Subcell
3Subcell
4Subcell
0.5
1
1.5
Radial stress r(r)
Radial stress r(r)
0
1
1.5
2
2.5
2.5
3
1.5
2.5
3
3.5
Radial distance r
4.5
1.5
2.5
3
3.5
Radial distance r
4.5
(b)
(a)
Figure 10: Radial stress r (r) for the hollow cylinder without and with the selective techniques
( = 0.4999)
x 10
3.5
0.5
x 10
Analytical solu.
SIM
1Subcell
2Subcell
3Subcell
4Subcell
1
1.5
Hoop stress (r)
Hoop stress (r)
2.5
Analytical solu.
FEM
1Subcell
2Subcell
3Subcell
4Subcell
0.5
1.5
0.5
2.5
3
1.5
2.5
3
3.5
Radial distance r
(a)
4.5
1.5
2.5
3
3.5
Radial distance r
4.5
(b)
Figure 11: Hoop stress (r) for the hollow cylinder without and with the selective techniques
( = 0.4999)
74
10
10
9
Top corner vertical displacement v
Top corner vertical displacement v
Figure 12: Cooks membrane and initial mesh
8
7
Q4
SIM
1Subcell
2Subcell
3Subcell
4Subcell
6
5
4
3
2
8
7
Q4
SIM
1Subcell
2Subcell
3Subcell
4Subcell
6
5
4
3
10
20
30
40
50
Number of elements per side
(a)
60
70
10
20
30
40
50
Number of elements per side
60
70
(b)
Figure 13: Vertical displacement at the top right corner of Cooks membrane; (a) = 0.4999,
(b) = 0.4999999
75
subcells. In preceding work, the authors showed that finite elements with such smoothed strains
yield solutions included in a solution space bounded by the standard displacement finite element
solution and a quasi-equilibrium element solution from dual, stress-based formulations. The single subcell version leads to the quasi-equilibrium element, while an infinite number of subcells
would allow the recovery of standard finite elements. A consequence of this property is that the
one subcell element is H1-superconvergent and leads to very accurate stress responses, while the
displacements are of lower quality. Inversely, a displacement-based method yields very accurate
displacements, but slower convergence and less accurate stress fields.
In previous work, we showed that the one subcell version of the smoothed strain finite element
method behaves very robustly in nearly incompressible settings, in which it does not exhibit volumetric locking. However, as noted above, the displacement accuracy of the one subcell element
is lower than that yielded by its multi-cell cousins.
Drawing from the conclusions above, this paper presented a selective integration technique for
smoothed strain finite element methods, in which the volumetric part of the strain field, responsible for locking, is written in terms of a single locking-insensitive subcell smoothing, while the
deviatoric part can be smoothed over an arbitrary number of subcells. As expected, the numerical
results show that the method is locking-free, and more accurate than the one subcell formulation.
Interesting extensions of the proposed method lie along elastic-plastic materials, including
large strain plasticity, where the insensitivity of the strain-smoothing method to element distortion will most likely yield increased flexibility. Also, the possibility to integrate along arbitrary
polygonal elements suggests a very useful application of the method in the context of partition
of unity enriched methods, as it would ease the integration of the stiffness matrix of enriched
elements and permit a straightforward treatment of incompressibility within partition of unity
methods, which has not been reported to date. Forthcoming work shall present applications to
plates and shells.
References
[1] B.M. Fraeijs de Veubeke, Displacement and equilibrium models in the finite element method, In Stress
analysis, edit. Zienkiewicz O.C., Holister G.S., Wiley, London, 1965.
[2] R.D. Cook, Improved two-dimensional finite element, J.Struct. Div. ASCE 100, 1974.
[3] D.S. Malkus and T.J.R Hughes, Mixed finite element methods - Reduced and selective integration technique:
a unification of concepts, Comput. Methods Appl. Mech. Engrg., 15: 63-81, 1978.
[4] J.S. Chen, C.T. Wu, S. Yoon, Y. You, A stabilized conforming nodal integration for Galerkin mesh-free
methods, Int. J. Numer. Methods Eng., 50:435-466, 2001.
[5] T. Belytschko, Y.Y. Lu, and L. Gu, Element-free Galerkin methods, Int. J. Numer. Methods Eng., 37:229 256, 1994.
[6] J.W. Yoo, B. Moran, and J.S. Chen, Stabilized conforming nodal integration in the natural-element method,
Int. J. Numer. Methods Eng., 60:861 - 890, 2004.
[7] J.S. Simo and T.J.R. Hughes, On the variational foundation of assumed strain methods, ASME Journal of
Applied Mechanics, 53:51 - 54, 1986.
[8] G.R. Liu, K.Y. Dai, T.T. Nguyen, A smoothed finite element for mechanics problems, Comput. Mech., in
press, 2006.
[9] S. Beissel, T. Belytschko, Nodal integration of the element-free Galerkin method, Comput. Methods Appl.
Mech. Engrg., 139: 49-74, 1996.
[10] J. Bonet and S. Kulasegaram, Correction and stabilization of smooth particle hydrodynamics methods with
76
applications in metal forming simulation, Int. J. Numer. Methods Eng., 47:1189-1214, 1999.
[11] S.P. Timoshenko and J.N. Goodier, Theory of Elasticity, New York, McGraw Hill 1970.
[12] Nguyen-Xuan Hung, S. Bordas, and Nguyen-Dang Hung, Finite element methods with stabilized conforming nodal integration: convergence, accuracy and properties, Int. J. Numer. Methods Eng., submitted,
2006.
[13] Nguyen-Xuan Hung, S. Bordas, and Nguyen-Dang Hung, On the connection between displacement and
equilibrium finite element method via strain smoothing, Comput. Methods Appl. Mech. Engrg., submitted, 2006.
[14] A. Masud and K. Xia, A Stabilized mixed finite element method for nearly incompressible elasticity, J.App.
Mech. ASME 72, 2005.
[15] T.J.R. Hughes, Generalization of selective integration procedures to antisotropic and nonlinear media, Int.
J. Numer. Methods Eng., 15:1413-1418, 1980.
[16] D.N. Arnold, F. Brezzi and Jr. J. Douglas, A new mixed finite element for plane elasticity, Jpn. J. Ind. Appl.
Math., Part 1 1, 347-367, 1984.
[17] J.C. Simo and F. Armero, Geometrically nonlinear enhanced strain mixed methods and the method of incompatible modes, Int. J. Numer. Methods Eng., 33:1413-1449, 1992.
77
Topology Optimization of Membranes Made of Orthotropic Material

Vincent Rion* and Michal Bruyneel+
*
Goodyear Technical Center,
Colmar-Berg, Luxembourg
+
SAMTECH s.a.,
Lige Science Park, Angleur, Belgium
Abstract
This paper presents a formulation for the topology optimization of continuum structures including
orthotropic materials. 2D membranes are considered. As for isotropic materials, the density is
considered but fibers orientations are also included in the problem in a simple extension of the
SIMP material parameterization. Only weight and stiffness criteria are taken into account. The
optimization problem is solved with sequential convex approximation methods. Optimal topologies
are obtained for several approximations in order to illustrate their ability in solving such a problem.
Keywords: topology optimization, composite material.
Introduction
With their high stiffness to weight ratio and their interesting strength properties, fibers reinforced
composite materials are widely used in automotive and aerospace industries. The composite
structures are made of laminates that include several unidirectional plies. Those plies are stacked
together (Figure 1) in a way that tends to maximize the efficiency of a material that can be
customized to the application.
Given that a huge number of parameters are necessary to define such materials, the design of
composite structures naturally calls for optimization methods. Although the optimal sizing problem
of composite structures has been addressed for some years in an industrial context, more advanced
techniques such as topology optimization are still under research.
Topology optimization aims at finding the optimal distribution of a material inside a prescribed
design domain for a given amount of material. Amongst all the possible solutions, one looks for the
one leading to the stiffest structure [2]. In the frame of FEM, a pseudo-density varying between 0
and 1 is assigned to each finite element, defining void or material, what leads to the definition of
holes in the optimized structure. Usually, isotropic materials are distributed in the structure.
Initially, some specific rank 2 laminated materials (including two densities and one orientation
design variables) were used to solve the optimal topology problem. They were a mathematical tool
allowing to determine theoretically optimal solutions, and the related orientation design variables
were determined with a slow converging optimality criteria only dedicated to 2D membrane
problems [13,10].
For more than 20 years, the Sequential Convex Programming approach, also called
Approximation Concepts approach, proved to be efficient for solving structural optimization
problems [15]. This technique has been successfully applied to optimal sizing problems [12], shape
optimization [18], topology optimization [10], structural optimization with discrete variables [1]
and composite structures optimal design [4]. For considering simultaneously thickness and
78
orientation design variables in composite structures optimization, the initially monotonous

approximation schemes were extended to mixed monotonous-non monotonous ones in the
GBMMA-MMA approximation [5]. This scheme, called GCM in the Boss/Quattro
(www.samcef.com) optimization tool box, allowed to solve industrial problems [8]. The dual solver
of this approximation makes it efficient for treating topology optimization problems including a
very large amount of design variables (of the order of 106) and a small number of design functions
(about 20). Recently GCM was found reliable in solving composite wing structures optimization
problems including more than 300000 constraints and 1000 design variables.
2
3
x p li
k
1
Figure 1: Plies stacked together to form a laminate

This optimization method and some of its variants are tested here for topology optimization
including orthotropic materials. In Section 2, the topology optimization problem for structures made
of an isotropic material is reminded. The topology optimization problem for composites is presented
in Section 3, together with the limitations of the present study. The SIMP law and a proposed
extension to composite materials are explained in Section 4. In Section 5, the optimization method
is described. Finally, Section 6 provides some numerical results on classical benchmarks.
The topology optimization problem
Topology optimization is a very general tool from structural optimization techniques. It allows to
determine automatically the layout of the structure, that is the optimal distribution of the mechanical
properties in a prescribed design domain for a given amount of material, as illustrated in Figure 2.
i = 1 =>
Ei = E0
i = 0
Domain where the

material is
distributed
0
i 0 => E i
i 0
Figure 2: The topology optimization problem (after [4])

Compliance minimization, that results in the stiffest structure that can be exhibited for a given
volume fraction of material at the solution, has been intensively studied in the literature [2]. The
problem writes:
79
min g T q
s.t.
V V
where g and q are the nodal forces and displacements, respectively. V is the available amount of
material to be distributed in the design area.
Local stress constraints were also taken into account [11]. Since the related optimization problem
is huge in terms of design variables and restrictions, those local (that is at the element level)
restrictions were only considered in solving academic applications of limited sizes. Specific
problems including body loads were investigated, as well [7]. This technology, applied to isotropic
materials, has reached maturity and is now used in the aerospace industry (e.g. the TOPOL software
www.samcef.com ).
Topology optimization with orthotropic materials
In the frame of composite structures, two topology optimization problems can be addressed. The
first one is the transverse topology optimization. In this case, one looks for the optimal stacking
sequence at several locations in the structure (Figure 3). In the second problem, the optimal
topology at a given height of the non homogenous laminate is to be found. This is an in-plane
topology optimization (Figure 4). A comprehensive topology optimization problem includes those
two aspects, as well as inter-regional and ply drop-off constraints related to ply continuity and
manufacturability considerations. This is a very difficult problem to be solved, and efficient
formulations should first be derived separately for each of those two problems.
Figure 3: Optimal stacking sequence as a transverse topology optimization problem (after [4])
In this paper, we propose a simple solution for the problem of Figure 4, that is the in-plane
topology optimization formulation. It is based on a direct extension of the SIMP parameterization
used for isotropic materials, and is illustrated here in the frame of the 2D membrane structures.
Even though 2D problems are considered, the method is directly applicable to 3D and shell
problems.
80
Figure 4: In-plane topology optimization problem (after [4])
The SIMP law for topology optimization
4.1 The case of isotropic materials

The design variable is the pseudo-density i that varies between 0 and 1. Such a variable is
defined for each finite element i. The SIMP material law takes the following form:
Ei = ip E 0
i = i 0
(1)
where E0 and 0 are the Young modulus and the density of the base material (e.g. steel), E and are
the effective material properties, and p is the exponent of the SIMP law, chosen by the user
(1<p<4).
In the case of plane stress, the stresses are linked to the deformations via the following relation:
= C
With the SIMP parameterization, it comes that:
Q11
p
C = i Q12
Q16
Q12
Q22
Q26
Q16
Q xx
p
Q26 = i Q yx
0
Q66
Q xy
Q yy
0
0
ip E 0
0 =
2
Qss 1
0
1
1
0
1
0 0
and the material stiffness matrix C depends on the density design variable :
= C( )
(2)
81
4.2 Extension of the SIMP law to orthotropic materials

For orthotropic materials in a plane stress state, the stiffness in the material axes is given by the
following expression:
mE x
C = m xy E y
m yx E x
mE y
0
0 Q xx
0 = Q yx
G xy 0
Q xy
Q yy
0
0
0
Qss
m=
1
1 xy yx
where 4 material properties Ex, Ey, xy and Gxy must be provided. For a material with orthotropic
axes oriented at an angle with respect to the reference axes (Figure 5), the material stiffness is
given by:
c4
Q11
4
Q
22
s
c 2 s 2
Q12
Q (1,2,3) =
= 2 2
c s
Q66
3
Q16
c s
Q26 (1,2,3) cs 3
2c 2 s 2
4c 2 s 2
s4
2c 2 s 2
4c 2 s 2 Q xx
c4
4c 2 s 2 Q yy
c2s2
c4 + s4
2c 2 s 2
(c 2 s 2 ) 2 Q xy
c2s2
cs 3 cs 3 c 3 s 2(cs 3 c 3 s ) Qss ( x, y , z )
c 3 s (c 3 s cs 3 ) 2(c 3 s cs 3 )
with c = cos and s = sin .

As for isotropic materials, the SIMP parameterization can be used here, and the material law for
topology optimization is now written as:
Q11
p
C = i Q12
Q16
Q12
Q22
Q26
Q16
Q26
Q66
= C( , )
(3)
The material stiffness now depends on both kinds of design variables, i.e. the material density and
the fibers orientation.
2
z,3
Material axes
1
Structural axes
Figure 5: Material and structural axes
82
The optimization method
The optimization method used to solve topology optimization of membranes made of an

orthotropic material is described in [8]. In the Sequential Convex Programming, the primary
optimization problem (4) is replaced by the solution of successive convex approximated
optimization problems (5).
min g 0 ( X )
g j ( X ) g max
j
j = 1,..., m
(4)
i = 1,..., n
x i xi x i
min g~0( k ) ( X)
(k )
g~ j ( X ) g max
j
(k )
xi
(k )
xi x i
j = 1,..., m
(5)
i = 1,..., n
The approach used here includes the monotonous MMA approximation [16] and the nonmonotonous GBMMA approximations [3]. The selection of the kind of approximation, either
monotonous or not, is done automatically for each design variable in each design function based on
the derivatives at two successive iterations:
g j ( X ( k ) )
xi
g j ( X ( k ) )
xi
g j ( X ( k 1) )
xi
g j ( X ( k 1) )
xi
>0
=>
monotonous approximation
<0
=>
non monotonous approximation
The mixed monotonous non monotonous approximation is called GBMMA-MMA. The GCMMA
approximation is only non monotonous and doesnt use information from the previous design point
[17]. The GBMMA1 approximation is built by fetching the derivatives at two successive iterations,
while GBMMA2 is a second order approximation using a finite difference of the derivatives to
compute the second order information. Local linear approximations can also be built. More details
can be found in [6].
Numerical applications
The results presented in the sequel were obtained in [14].
6.1 First application

The problem is illustrated in Figure 6. It includes 1800 design variables (900 densities and 900
orientations). The optimal topologies obtained for several starting points are provided in Figures 7
and 8. Depending on the initial values for the orientations, different optimal topologies are obtained.
This is explained by the non convex behavior of compliance with respect to the orientation
variables, as depicted in Figure 9. Since the optimization method builds local approximations of the
structural responses, the global optimum can not be reached.
83
The compliance (Figure 9) is calculated for an element located on the top of the structure. A local
optimum can be reached around 50 whereas the global optimum is about 140 for the considered
element. This conclusion is inverted if a bottom element is considered.
?
Figure 6: Definition of problem 1
Figure 7: Solution for initial fibers oriented at 45
Figure 8: Solution for initial fibers oriented at -45

Dont lose sight that the solution provided by the optimizer corresponds to a local optimum which
could be the global one. This is well highlighted in this first application, where a symmetrical
solution is expected, but the optimizer reaches first a local optimum.
84

Compliance vs orientation in degrees
0.003955
0.0039545
0.003954
0.0039535
0.003953
0
20
40
60
80
100
120
140
160
180
Figure 9: Evolution of the compliance with respect to the orientation in one element
6.2 Second application

The problem is defined in Figure 10. It includes 3750 design variables. The optimal topology and
orientations obtained for an half of the structure are given in Figure 11. A comparison of the
convergence speed for several approximations is provided in Figure 12.
?
Figure 11: Optimal topology for problem 2
Figure 12: Comparison of the convergence speed of several approximations

85
6.3 Third application

The problem is defined in Figure 13. It includes 1200 design variables. Results are provided in
Figures 14 and 15. The size and the boundary conditions are different from problem 2.
?
Figure 14: Optimal topology for problem 3
Figure 15: Comparison of the convergence speed of several approximations

86
7 Conclusions
The optimization procedure proved to be reliable in solving topology optimization problems
including orthotropic materials. The mixed GBMMA-MMA approximation seems to be the most
efficient scheme mainly due to the switch between monotonous and non-monotonous
approximations. Truss-like structures are sometimes observed in the solutions: the use of a lower
value for the exponent of the power law material should prevent this effect and lead to structures
including intermediate densities.
Future work will study the influence of strength constraints (envelop of Tsai-Wu, Tsai-Hill,
criteria) on the optimal topology of composite structures. The related optimization problem will
then include a lot of design function together with a high number of design variables. Finally, being
able to consider fibers orientations as design variables will avoid the singularity of strength
constraints in the optimization problem [9].
Acknowledgments
Professor Claude Fleury (University of Lige) is gratefully acknowledged for making the dual
solver of the CONLIN optimizer available.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
Beckers M. (1997). Optimisation de structures en variables discrtes, Thse de Doctorat, Collection des
Publications de la Facult des Sciences Appliques N 181, Universit de Lige, Belgium.
Bendsoe M.P. and Sigmund O. (2003). Topology Optimization: Theory, Methods and Applications, Springer.
Bruyneel M., Vermaut O. and Fleury C. (1999). Two-point based approximation schemes for optimal
orientation in laminates, Third World Congress of Structural and Multidisciplinary Optimization WCSMO3,
Amherst, NY, May 1999 (CD Proceedings).
Bruyneel M. (2002). Schmas dapproximation pour la conception optimale de structures en matriaux
composites, Collection des Publications de la Facult des Sciences Appliques N 219, Universit de Lige,
Belgium.
Bruyneel M. et Fleury C. (2002). Composite structures optimization using sequential convex programming,
Advances in Engineering Software, 33, pp. 697-711.
Bruyneel M., Duysinx P. and Fleury C. (2002). A family of MMA approximations for structural optimization,
Structural & Multidisciplinary Optimization, 24, pp. 263-276.
Bruyneel M. and Duysinx P. (2005). Note on topology optimization of continuum structures including selfweight, Structural & Multidisciplinary Optimization, 29, pp. 245-256.
Bruyneel M. (2006). A general and effective approach for the optimal design of fiber reinforced composite
structures, Composite Science & Technology, 66, pp. 1303-1314.
Bruyneel M. and Duysinx P. (2006). Note on singular optima in laminate design problems, Structural &
Multidisciplinary Optimization, 31, pp. 156-159.
Duysinx P. (1996). Optimisation topologique : du milieu continu la structure lastique, PhD Thesis, Facult
des Sciences Appliques, Universit de Lige, Belgium.
Duysinx P. and Bendsoe M.P. (1998). Topology optimisation of continuum structures with local stress
constraints, International Journal of Numerical Methods in Engineering, 43, pp. 1453-1478.
Fleury C. (1978). Dimensionnement automatique des structures, PhD Thesis, Facult des Sciences Appliques,
Universit de Lige, Belgium.
Pedersen P. (1989). On optimal orientation of orthotropic materials, Structural Optimization, 1, pp. 101-106.
Rion V. (2002). Optimisation topologique de structures en matriaux orthotropes, Master Thesis, Facult des
Sciences Appliques, Universit de Lige.
Schmit L.A. and Fleury C. (1980). Structural synthesis by combining approximation concepts and dual
methods, AIAA Journal, 18, pp. 1252-1260.
Svanberg K. (1987). The Method of Moving Asymptotes: a new method for structural optimization,
International Journal of Numerical Methods in Engineering, 24, pp. 359-373.
Svanberg K. (1995). A globally convergent vesrion of MMA without linesearch, First World Congress of
Structural and Multidisciplinary Optimization WCSMO1, Goslar, Germany, May 28- June 2, 1995.
Zhang W.H. (1991). Calcul de sensibilit et optimisation de forme par la mthode des lments finis, PhD
Thesis, Facult des Sciences Appliques, University of Lige, Belgium.
87
Limit Analysis Theorems for Materials with a Non-Associated Flow Rule

Mohammed Hjiaj
Laboratoire de Genie Civil & Genie Mecanique

INSA de Rennes
20 avenue des Buttes de Coesmes
35043 Rennes Cedex, France
Abstract
The aim of the paper is to propose an extension of limit analysis theorems for a non-associated
flow rule. The approach is to construct, as a first step, a variational formulation for boundary
value problems involving rigid plastic materials with a non-associated flow rule. The method
used here, proposed by Joachim Telega, is based on an extension of Auchmutys work on variational principles for non-potential operator equations. Having at our disposal the variational
formulation, it is a simple matter to derive the bound theorems of limit analysis. It is found that
for a non-associated flow rule the lower and the upper bound are coupled.
Keywords: Limit analysis, Non-associated plasticity, Variational principles, Non-potential operator.
1 Introduction
The question of extending limit analysis to materials with a non-associated flow rule is an old
but important one. Many frictional soils, such as sands and drained clays, do not display the
large volume change upon shearing that is predicted by an associated flow rule, thus calling into
question the collapse loads that are found from the classical limit theorems. This question is
important in practice, as limit analysis is often used to predict the ultimate (or limit) loads for a
wide range of geotechnical structures such as slopes, tunnels, foundations and retaining walls.
In 1953, Drucker (6) proved that the limit load for any plastic material with a non-associated
flow rule is less than that for the corresponding material with an associated flow rule. After
Drucker, Radenkovic (9) proposed the first extension of the limit theorems to a particular class
of non-associated plastic materials. These materials must satisfy two assumptions: (a) the plastic
potential and the yield function are convex, and (b) the additive constant for the potential can be
chosen such that the surface defined by the plastic potential lies inside the elastic domain. He
proved that the true limit load lies between the values of the limit loads for two associated materials. The first one, M1 , is defined by a yield surface which coincides with the real yield surface
and the second one, M2 , has a yield surface defined by the plastic potential. The limit load for the
first one is an upper bound for the real material, and the limit load for the second one is a lower
bound for the real material. This result follows from the fact that a statically admissible field for
the real material is also a statically admissible stress field for the material M1 , while a kinematically admissible velocity field for the real material is also a kinematically admissible velocity
field for the material M2 . If we consider a material defined by a Mohr-Coulomb criterion, the
88
validity of such an approach will depend on the degree of non-associativity of the material. Unfortunately, it is of little use for the practically important case of a frictional material deforming
at constant volume. Following Radenkovic, Palmer (8) suggested another approach involving the
assembly of a surface inside the elastic domain. For each point on the yield surface f () = 0,
a hyperplane orthogonal to the direction of flow is assembled. If these hyperplanes form an envelope that lies completely within the yield surface, then the envelope is necessarily convex. This
new surface serves as a yield surface for the material under consideration and lower and upper
bounds can be computed in the usual way. For a cone-like yield criterion and potential satisfying
Readenkovics assumptions, the result gives simply the potential. Again, the incompressible case
cannot be treated by this approach.
From a mathematical point of view, the lack of bound theorems is explained by the nonexistence of a pseudo-potential such that its subgradients yield back the flow rule. Therefore, the
non-linear operator governing the problem is not symmetric with respect to a suitable bilinear
form. These problems are called non-potential initial-boundary value problems. To recover a
potential structure of the flow rule (essential for extremum principles), de Saxce (5) proposed
an implicit normality form of the flow rule. He showed that it is possible to derive the flow
rule along with its inverse from a single scalar-valued function called a bi-potential that enjoys
convexity properties. Using this formulation, he was able to propose an extension of the limit
analysis theorems. Unfortunately, their approach doesnt exclude the possibility of the lower
bound load multiplier exceeding the upper bound load multiplier.
It is known that in the absence of symmetry of the operator it is very difficult to construct a
relevant variational formulation. Several methods have been proposed in the literature. A significant contribution has been made by Brezis and Ekeland ((3) & (4)), whom proposed a variational
principle for some parabolic equations. Their approach exploits fundamental concepts of Convex Analysis. A few years later, Tonti (12) proposed a general technique to obtain variational
formulations for a general nonlinear problem. His method is based on the choice of a suitable
linear, invertible, symmetric kernel function. The later allows the construction, in a systematic
way, of an extended functional associated with any nonlinear problem. Tontis ideas has been
further developed by Auchmuty ((1) & (2)) who proposed to use an operator split technique to
construct an extended functional for the linear case. In a recent work, Telega (11) proposed an
extension of Auchmutys approach to more general differential inclusions. He derived extremum
and saddlepoint principles for nonlinear, nonconservative and nonpotential elasticity and the stationary and nonstationary quasi-linear heat equation.
The aim of the paper is to exploit the formulation proposed in (11) to derive extremum principles for initial/boundary value problems involving non-associated rigid plastic materials. Based
on these principles, we provide an extension of the classical limit analysis theorems. The paper
is organized as follows. The next section defines the problem under consideration, describes
the governing equations, and introduces the notation used. In section 3, the incompressible
non-associated material is presented and an alternative formulation of the flow rule is discussed.
Exploiting the partly associated character of the flow rule, the dissipation function is calculated.
As expected, this function depends on both the plastic strain rate and the stress. The method for
deriving extremum principles for non-potential operator equations is recalled briefly, with reference to solid mechanics, in section 4. In section 5, we apply the technique to boundary value
problems involving plastic material with a non-associated flow rule. Having at hand these new
89
principles, a formulation of the bound theorems of limit analysis is proposed.
2 Boundary value problem for rigid plastic solids

In this section, we recall the boundary value problem associated with a rigid plastic material subjected to a quasi-static loading. To ensure the existence of a solution, we must assume that the
initial loads do not cause the collapse of the structure, i.e. the initial loads are below the ultimate
loads.
Consider a perfectly plastic body occupying a bounded domain R3 of boundary ,
subjected to imposed body force b, imposed surface traction t0 on the part t of , and imposed
velocity u on the remaining part u = t of the boundary. We assume that the body is fixed
on u , i.e. u = 0. Velocities and strain rates are assumed to be small, so that the geometry
change can be neglected and the analysis is performed on the reference configuration. We also
suppose that, in the absence of surface tractions, the body can support its own weight. Let V
be the general admissible velocity space defined on . The element u V is a vector-valued
function u : R3 satisfying certain differentiability and integrability conditions. The dual
space F = V of V is the admissible force space on (the bar over a set denotes its closure)
whose elements are specified by the body force b on and the surface traction t0 on t . Let the
geometrical operator (deformation operator) D be a continuous linear transformation from V to
the space of plastic strain rates fields E:
D : V E : u V p = Du = S u
(2.1)
where S is the symmetric part of the gradient operator associated with a vector. The dual
space S = E of E is the admissible stress space on . The element (x) S is a tensorvalued function : E3s satisfying certain differentiability and integrability conditions, say
(x) L2 () and E3s stands for the space of symmetric 3 3 matrices. The behavior of the
material is specified by its flow rule along with the complementary relations:
g
p =
f 0 , 0 , f = 0
(2.2)
(2.3)
where f is the yield surface and g the plastic potential. If the yield surface coincides with the
potential (f g) the flow is associated and the Drucker inequality holds:
0 K :
(2.4)
p ( 0 ) 0 , K
where 0 is any arbitrary stress compelled to belong to K which is the convex set of plastically
admissible stresses:
K = { S | f () 0}
Using Convex Analysis tools, Moreau (7) further transformed Druckers inequality and proposed
an alternative formulation of the flow rule. The key-idea is to replace the element of (), which
constrains the stresses in Druckers inequality, by the indicator function of the set K, defined by
0 if K,
IK () =
+ otherwise
90
which is convex because of the convexity of the elastic domain but non-differentiable with respect
to . Accordingly, the inequality becomes:
IK ( 0 ) + p ( 0 ) IK ()
which means the the plastic strain rate belongs to the subdifferential of the indicator function
IK ():
p IK ()
(2.5)
The subdifferential IK () is the set of all subgradients of IK () at . It defines a multivalued
mapping from S into E:
IK () : S 7 E : 7 IK ()
In particular, for a differentiable function, the subdifferential is reduced to a singleton which
corresponds to the classical gradient. The previous formulation is particularly helpful in deriving
extremum principle. However, this formulation exists only for an associated flow rule since
Druckers inequality does not hold for a non-associated one. By applying the Fenchel transform:
sup [ p IK ()] = ( p )
(2.6)
the inverse of the flow rule, i.e. ( p ), can be obtained:

( p ) ,
(2.7)
where ( p ) is the so-called pseudo-potential of dissipation, which is convex by construction and

is a positively homogenous function of order one. Further, the pseudo-potential of dissipation
gives the rate of dissipation. In terms of Convex Analysis, the pseudo-potential is said to be the
conjugate function of the indicator function and conversely:
( p ) = (IK ())
and IK () = (( p )) .
The definition (2.6) of ( p ) implies that

( p 0 ) + IK ( 0 ) 0 p 0
(2.8)
for each p 0 E and each S. This inequality will play an essential role in the derivation
of minimum principles in plasticity. In (2.8) the equality holds if and only if p IK (), or
equivalently if and only if ( p ). Now, we need to define some sets used in the next
sections.
2.1 Admissible fields

Static admissibility. In the context of limit analysis, a stress field is said to be statically admissible if it satisfies the local equilibrium equations within the domain and on its boundary and
does not violate the plasticity criterion:
in
div + b = 0
n = l t0 on t
(x) K
x
91
where l is the maximum admissible load multiplier. The space Sad defined by
Sad = S | div + b = 0 in , n = s t0 on t
(2.9)
encompasses all stress fields that satisfy the strong equilibrium equations. The space Ssa of
statically admissible stresses can be defined as follows:
Ssa = { S | (x) Sad K}
(2.10)
Kinematic admissibility. A velocity field is said to be kinematically admissible if it satisfies

the local compatibility equations within the domain and on its boundary and complies with the
flow rule:
S 6= 0 in
p = (u)
u = 0
on u
p
(u(x))
IK ()
in
S 6= 0 means that we are looking for a velocity field in which the external
The condition (u)
loads produce non-negative work:
Z
t u d > 0
t
The space Vad defined by

S in , u = 0 on u }
Vad = {u V | p = (u)
(2.11)
contains all velocity fields that satisfy the local compatibility equations. The kinematically admissible velocity space Vka is defined by
Vka = u Vad | (u(x))

IK () ,
t u d > 0
(2.12)
Here, the velocity may eventually be discontinuous at the crossing of certain surfaces, but this
discontinuity is not arbitrary, it must be compatible with the flow rule. Finally, we denote the
space of admissible plastic strain rates by
Ead = { p E | u Vad
such that
p = S u}
and the space of kinematically admissible plastic strain rates by
g
p
p
Eka = Ead | =
(2.13)
(2.14)
The boundary value problem The boundary value problem of limit analysis involves finding
such that the stress field is statically admissible, the velocity field u is
a triple (l , , u)
kinematically admissible, and l is the associated limit load multiplier. We can write this problem
formally by:
F ind (l , , u)
(2.15)
such that Ssa , u Vka
92
3 The non-associated plastic model

It is well known that soil materials have a very complicated behavior. Idealizations are, therefore, often necessary in order to develop simple mathematical constitutive laws for practical
applications. Several models can be found in the literature, most of them are complex and require many parameters, some of which are physically meaningless. The relative simplicity of
the Drucker-Prager model, which simulates several key characteristics of soil behavior with only
three parameters, explains why it is widely used. Here, the governing constitutive relations are
developed with reference to the classical elastic perfectly plastic Drucker-Prager model. Hence,
p
) where
the model is described by the dual variables = ( d , sm ) and p = (e pd , e m
sm
d
p
e m
e pd
= 31 Tr()
= sm I
= Tr( p )
p
I
= p 31 e m
where I is the 3 3 unit matrix. The convex set of plastically admissible stresses (Figure 3.1) is
given by
1
(3.1)
K = S | k d k + sm tan c 0
kd
where c is the cohesion, the friction angle, kd a constant and kk denotes the classical Euclidean
norm. Its boundary defines the yield function:
e&
s
k
K
q
s
m
Figure 1: The Drucker-Prager cone

93
f=
1
k d k + sm tan c = 0 .
kd
(3.2)
The subscript attached to K in (3.1) is used to distinguish the admissible domain of a frictional
material with the one, denoted by K , of a frictional material with = . The flow rule is given
by (2.2) where g is defined by
g=
1
k d k + sm tan
kd
(3.3)
in which is the dilatancy angle and a positive constant with arbitrary magnitude. By adopting
the plastic potential (3.3) the plastic flow satisfies
p
pd
kd kek
e m
(3.4)
The equality in (3.4) is reached everywhere on the yield surface except at the apex. This inequality suggests to introduce the cone dual to K :
p
K = { p E|e m
kd ke pd k}
(3.5)
Alternative formulation of the flow rule The rigid plastic model presented above is nonassociated and we can not derive a convex pseudo-potential that describes it. This fact can be
easily proved using the cyclic monotonicity condition. In fact, the law being non-monotonic, it
is not cyclically monotonic and no convex pseudo-potential can be derived. Simple geometric
consideration allow us to write the flow in the following compact form:
p ) IK
(e pd , E
with
(3.6)
p = e p + kd (tan tan )kek

pd
E
m
Then, the flow rule can be expressed under the form of a pseudo-variational inequality:
(
0
( d0 , sm
) K :
p
p (sm s 0 ) 0 , ( d , sm ) K
e ( d 0 ) + E
d
(3.7)
The previous inequality suggests that the flow rule of the transformed plastic strain rate can be
formulated in the form of the principle of maximim dissipation: the actual stress is an optimal
solution for the maximization problem in the variable :
(
p sm ]
maximize [e pd d + E
subject to ( d , sm ) K
To further transform inequality (3.5) in the same way as for the perfect plastic model with an
associated flow rule. We us the indicator function to enforce the constraints on :
p (sm s 0 ) IK ()
IK ( 0 ) + e pd ( d d0 ) + E
m
The dissipation potential is obtained by applying the Fenchel transform to (3.6):

i
h
e p , E
p IK () = kd (c sm tan ) ke p k
p ) = sup d e p + sm E
(
d
d
(3.8)
(3.9)
94
the inversion of the flow rule is achieved

d sm (e p )
(3.10)
The two quasi-pseudo-potentials are related by the Fenchel inequality:

sm (e 0p ) + IK(sm ) ( 0d ) 0d e p 0
(3.11)
The equality is reached in (3.11) if and only if (3.9) is satisfied. In particular, if is equal to zero,
we recover the expression of the pseudo-potential of dissipation for a purely cohesive material
with an associated flow rule
c (e p ) = kd c ke p k
(3.12)
4 Minimum principle for non-potential operator

In this section, we recall briefly some fundamental results introduced in (11). It is found that
the term variational may be understood in different ways. This term can mean: weak formulation (i.e. principle of virtual work), stationary principle or extremum principle. Although a
weak formulation can be associated with most of the physical problems, a few of them admit
a stationary principle and lesser number admit an extremum principle. Extremum principles
are particularly attractive from both a mathematical and a computational point of view. In fact,
mathematicians used them to prove existence and eventually uniqueness of the solution to the
corresponding boundary value problem. The possibility of searching for the solution of a physical problem as a minimum point of a convex functional on a convex set is especially relevant
from the computational point view. Consider an operator equation, not necessarily linear,
N (u) = f
where u belongs to a properly chosen function space V and f to its dual V .
(i) The weak formulation, being a rather general form of the virtual work principle, is given by
fV
< N (u), v >=< f , v >
v V
Here < , > stands for the duality pairing on V V.

(ii) Stronger is the stationary principle
J (u) stationary
over
Then, an arbitrary variation of the functional yields back the strong form of the equation
J =< N (u) f , u >= 0
N (u) = f
(iii) The strongest are extremum principles (minimum or maximum principles and min-max principles or saddle-point principles).
For instance,
J (u) min over V
where
J = J1 + IC
95
Here C is a set of constraints and IC denotes its indicator function [16,32,33]:
0 if u C,
IC (u) =
+ otherwise
In the next subsection, we recall the method, proposed in (11), to derive extremum principles for
a non-potential operator equation. The relations are developed for the general case and specifications are made with respect to our problem.
4.1
Abstract framework
Consider the problem of solving the following differential inclusion

F (N (u)) + f u (u)
(4.1)
where
- : V Y is a continous linear operator,
- Y is another space whose elements are denoted by q,
- Y is the dual space of Y ,
- N : V V1 is a continuous operator, not necessarily linear,
- F : V1 V is a continuous mapping.
For solid mechanics applications, the operator is the deformation operator assumed here to be
linear. Accordingly, we have that Y E and Y S. The adjoint operator of , denoted by ,
is the equilibrium operator:
divq
in
in
S u

and q =
u =
n q
on t
on u
u
Associated with the problem, consider the lagrangian L : V V R defined by :
L (u, v) = (u) (v) hF (N (u)) + f , v ui.
(4.2)
By setting:
q = v
and q = F (N (u)) + f
the lagrangian (4.2) becomes

L (u, q ; q) = (u) (q) + hq , qi hF (N (u)) + f , ui.
(4.3)
Now, we define the functional G (u, q ) as:

G (u, q ) = sup L (u, q ; q)
qY
= sup [hq , qi (q)] hF (N (u)) + f , ui + (u)

qY
= (u) + (q ) hF (N (u)) + f , ui
(4.4)
The minimum principle associated with (4.1) means evaluating
Find
=
inf
uV, q Y
{(u) + (q ) hF (N (u)) + f , ui}
(P)
subject to q = F (N (u)) + f
96
Remark 4.1 We observe that the last principle involves the adjoint operator which usually
can be calculated easily. Furthermore, we have that
G (u, q ) 0
G (
u, q ) = 0
and
(4.5)
if only if (
u, q ) is a solution to (4.1)
The dual problem means evaluating
= sup H (q)
(P ).
(4.6)
q = F (N (u)) + f } .
(4.7)
qY
where
H (q) =
inf
uV, q Y
{L (u, q ; q)
A point (
u, q ; q) V Y Y is a saddle point of L if,
L (
u, q ; q) L (
u, q ; q) L (u, q ; q) u V, q Y, q Y
satisfying the constraint q = F (N (u)) + f
Properties:
1. If (P) and (P ) are nontrivial, then and are finite with .
2. If > 0, then (4.1) does not posses a solution.
3. If there exists a solution q in Y of (P ) with H (q) = 0 and a (
u, q ) in V Y such that
(
u, q ; q) is a saddle point of L , then (
u, q ) will be a solution of (4.1).
4. If (
u, q ; q) is a saddle point of L, then
0 = G (
u, q ) = L (
u, q ; q) = H (q).
4.2 Applications
4.2.1 Example 1
Let us study the problem (P) more closely and consider the particular case where
F (N (u)) 0.
Then, we have
G (u, q ) = (u) + (q ) hf , ui.
(4.8)
and the problem P can be written as follows

inf{(u) hf , ui | u V } + inf{ (q ) | q Y , q = f }
Since G (u, q ) 0, we have
(u) hf , ui (q ) u V , q Y
(4.9)
Thus we recover, for instance, the principle of the total potential energy of linear elasticity:
(u) minimum
and its dual (the principle of the total complementary energy):
(q ) sup
It suffices to set (u) = u and q = .
97
4.2.2 Example 2
Consider the boundary value problem associated with an elastic solid described by a non-symmetric
elastic operator:
= A
with A 6= AT
(4.10)
In this case N is linear. In term of displacement, the strong form of the governing equations are:
div(A(S u)) + b = 0 in
n (A(S u)) = t on t
u = 0 on u
Let us make the following decomposition:
A = A1 + A2
(4.11)
where A1 is such that its elements satisfy a1ijkl = a1klij . Taking into account the decomposition
(4.11), the problem can be written as (4.1) where
F (u) =
div(A2 (S u)
n (A2 (S u)
in
on t
f=
b
t
in
on t
(4.12)
and
(u) =hdiv (A1 (S u)) , u)i + h(n (A1 (S u)) , u)i
Z
= U (1 ) d
where 1 = A1
1 and U is the potential energy functional. Furthermore, it is clear that
q 1 = A1 .
The conjugate functional is given by
Z
W ( 1 ) d
(q ) =
where W is the complementary potential energy functional. Now, we are able to give the expression of the functional H (u, q )
(Z
b u d
(U (1 (u)) + W (q )) d
inf
u, q
Z
div 2 u d +
t u d
(4.13)
(n 2 ) u d
t
where 2 A2 (S u).
98
The last two terms in (4.13) can be rearranged and we obtain

(Z
Z
Z
inf
(U (1 (u)) + W (q )) d b u d t u d
u, q
Z
+
(4.14)
2 (u) (u) d
In general, this functional is not convex because of the last term. However, the linearity of the
last term ensures that the functional is bi-convex with respect to (u, q ). As a result, we can not
prove that the solution of the problem is unique.
5 The limit analysis problem

5.1 The associated case
In plasticity, the constitutive operator is non-linear and multivalued. In what follows, we will
denote this operator by A and its inverse by A1 such that we can write the flow rule and its
inverse respectively as
p A1 () and A( p )
(5.1)
If the flow rule is associated, the operator A corresponds to the subdifferential of the pseudopotential of dissipation. We consider first an incompressible rigid plastic material with an asso which belongs to the divergence-free space
ciated flow rule. We denote the velocity field by u,
Vad .:
Vad = u [H1 ()]3 | u = 0 on u , div u = 0 in

The linear operator is the deformation operator D consisting of a transformation from Vad to
the space of plastic strain rate E:
= S (u)
e pd (u)
Let be any load multiplier such that l . The boundary value problem can be stated as
div (A(S u))

n (A(S u))
u
p
A( )
=
=
=
b
t0
0
K
in
on t
on u
in
or equivalently
f (u)
(5.2)
where f is given by (4.13b) and (u) by

Z
d
c (e pd (u))
(5.3)
where c (e pd ) is given by (3.13). We note that the integral (5.3) contains also the dissipation terms
on the surfaces of discontinuities. The above expression is simplified by the fact that F 0. It
is quite straightforward to calculate the dual functional (q ) where q :
() = ICc ()
(5.4)
99
where Cc is the set of stress fields that do not violate the yield criterion:
Cc = { = ( d , sm ) , L2 (, E3s ) | (x) Kc (x) x }
) is now given by
The functional G (u,
Z
Z
Z
p
) = c (e d (u))
d + ICc () b u d t0 u d
G (u,
(5.5)
which is separable into a function of u and a function of . Hence, we get the extremum principle:
Z
Z
Z
d b u d t0 u d
(5.6)
inf
c (e pd (u))
ka
uV
and the dual one

sup {ICc ()} .
(5.7)
Ssa
Introducing the kinematic load multiplier

Z
Z
p
d b u d
c (e d (u))
=
k (u)
(5.8)
t0 u d
t
the minimum principle (5.6) yields back the upper bound theorem:
l = inf k (u)
ka
uV
(5.9)
The maximum principle (5.7) gives the lower bound theorem:

l =
sup
Ssa K
s ().
(5.10)
) 0, therefore
Since G (u,
(5.12)
s l k .
5.2 The non-associated case

Now we assume a rigid plastic material with a non-associated flow rule (2.2) and (2.3). Let us
define two fictitious materials with associated flow rules. The first one, M1 , is defined by a yield
surface (3.2) which coincides with the real yield surface and the second one, M2 , has a yield
surface defined by the plastic potential. The dissipation function for M1 is:
c e pm
+ IK ( p ),
(5.13)
( p ) = sup p IK () =
tan
S
where the convex set K (see Figure 3.1) defined by
K ( p ) = { p E | e pm kd tan ke pd k} .
is the dual of K (3.1). The pseudo-potential of dissipation for M2 is denoted by ec (e p ) and its
expression is obtained from (3.13) by replacing c with cte. Referring to (4.1), now we see that
are possible. Let us consider two such choices.

F (N (u)))
6= 0. Different choices of F (N (u))
100
5.2.1 First choice

One can consider the dissipation function of M1 to construct the functional :
Z
d.
(u) = ( p (u))
(5.14)
Let us denote by 1 ( 1 q ) the stress satisfying the inverse flow rule for the solid M1
1 ( p )
(5.15)
2 (u)
is necessarily given by
Then N (u)
= e pd sm (sm (u),
e pd (u))
( p (u))
N (u)
(5.16)
where sm is given by (3.9). The total stress is

= 1 + 2
(5.17)
where 1 is given by (5.15). The functional dual of , denoted by , is given by:

( 1 ) = IC ( 1 )
(5.18)
where
C = { = ( d , sm ) , L2 (, E3s ) | (x) K (x) x }.
Using the decomposition (5.17), we can specify the operator F as
div 2 (u)
in
=
,
F (N (u))
n 2 (u)
on t
and formulate the minimum principle associated with the boundary value problem for nonassociated rigid plastic solids:
(Z
Z
d
(e pd (u))
= inf
1
u,
(5.19)
p (u)
d
2 (u)
t0 u d +
b u d
+ IC ( 1 )
Accordingly, the upper theorem of limit analysis is:

2 (u))
l = inf k (u,
(5.20)
ka
uV
where
Z
d
(e pd (u))
2 (u))
=
k (u,
Z
p
b u d
(u)
d
2 (u)
t0 u d
t
The lower bound theorem is:

=
l ( 1 , 2 (u))
sup
(u)S
sa C
s ( 1 + 2 (u))
(5.21)
101
is taken from the upper bound theorem (5.20).

Remark 5.1 In the lower bound theorem 2 (u)
Consequently, the kinematic and the static approach are still coupled through the term 2 (u).
5.2.2 Second choice
Now, can consider the dissipation function of M2 to construct :
Z
d
(u) = ec (e pd (u))
(5.22)
Then we follow the same steps as for the first choice except that
= 2 (u)
= e pd sm (sm (u),
e pd (u))
ec (e pd (u)),
N (u)
( 1 ) = ICec ( 1 ),
where
(5.23)
(5.24)
e c (x) x }.
Cec = { = ( d , sm ) , L2 (, E3s ) | (x) K
in (5.19)
Accordingly, the lower and the upper bound are obtained by replacing , and 2 (u)
by their expressions (5.22), (5.23) and (5.24).
Remark 5.2 In the last formulation, the dissipation function depends on the constant which
is arbitrary. However, this does not affect the functional since this constant can be eliminated by
rearranging the first and the last terms in (5.19).
6 Summary
In this paper, a new formulation of the theorems of limit analysis applicable to non-associated
plastic material has been given. The study deals with the non-associated incompressible plastic
material but can be extended to more general cases where the dilatancy angle is non-zero. The
derivation is based on a method, proposed by Telega, for constructing extremum principles for a
non-potential operator equation. The functional enjoys convexity properties that can be exploited
in the future for devising appropriate numerical algorithms.
References
[1] Auchmuty, Giles, Variational principles for operator equations and initial value problems, Nonlinear Anal.,
Theory, Methods & Applications, 12 (1988), 531-564.
[2] Auchmuty, Giles, Min-max problems for non-potential operator equations, Contemporary Mathematics, 209
(1997), 19-28.
[3] Brezis, H. and I. Ekeland, Un principle variationnel associe a` certaines e quations paraboliques. Le cas
independant du temps, C. R. Acad. Sci. Paris, Ser. A, 282 (1976), 971-974.
[4] Brezis, H. and I. Ekeland, Un principle variationnel associe a` certaines e quations paraboliques. Le cas
independant du temps, C. R. Acad. Sci. Paris, Ser. A, 282 (1976), 1197-1198.
[5] de Saxce, G. and L. Bousshine, Limit analysis theorems for implicit standard materials : application to the
unilateral contact with dry friction and the non-associated flow rule in soils and rocks, International Journal
of Mechanical Sciences, 40 (1998), 387-398.
102
[6] Drucker, D.C. , Coulomb friction, plasticity and limit loads, Journal of Applied Mechanics Trans. ASME, 21
(1954), 71-74.
[7] Moreau, J.J. , Sur les lois de frottement, de plasticite et de viscosite, C. R. Acad. Sci. Paris, Ser. A, 21 (1970),
608-611.
[8] Palmer, A.C. , A limit theorem for materials with non-associated flow laws, Journal de Mecanique, 5 (1966),
217-222.
[9] Radenkovic, D. , Limit theorems for a Coulomb material with a non standard dilatation. C. R. Acad. Sci.
Paris, 252 (1961), 4103-4104.
[10] Telega, J.J. , Limit analysis theorems in the case of Signiroinis boundary conditions and friction, Arch. Mech.,
37 (2002), 549-562.
[11] Telega, J.J. , Extremum principles for nonpotential and initial value problems, Arch. Mech., 54 (2002), 565592.
[12] Tonti, E. , Variational formulations for every nonlinear problem, Int. J. Engng. Sci., 22 (1984), 13431371.
103
Cold Roll Forming and Metal Cutting Simulation using a 3D Arbitrary

Lagrangian Eulerian Formulation
R. Boman, L. Papeleux and Jean-Philippe Ponthot
University of Liège LTAS-MC&T
Chemin des Chevreuils, 1, B4000 Liège, Belgium
e-mail: jp.ponthot@ulg.ac.be
Abstract
In this paper, the Arbitrary Lagrangian Eulerian formalism is used to compute the steady state of
a 2D metal cutting operation and a 3D U-shaped cold roll forming process. Compared to the Lagrangian case, this method allows the use of a refined mesh near the tools, leading to an accurate
representation of the chip formation (metal cutting) and the bending of the sheet (roll forming)
with a limited computational time. The main problem of this kind of simulation is the rezoning
of the nodes on the free surfaces of the sheet. A modified iterative isoparametric smoother is
used to manage this geometrically complex and CPU expensive task.
Keywords: Cold roll forming, metal cutting, numerical simulation, Arbitrary Lagrangian Eulerian
formulation.
1 Introduction
Being a progressive and continuous process, in which small amounts of forming are applied at
each pass of rolls, cold roll forming is largely employed to bend a long strip of sheet blank into
a desired cross-sectional profile via roller dies.
As the sheet is continuously bent and plastically deformed along a linear axis in order to
progressively shape to the desired contour, a complex three dimensional surface is developed.
The material at different positions is formed by different strain paths.
In order to obtain defect-free products, the amount of change in each pass must be limited so
that the required bends can be formed without the occurrence of plastic longitudinal elongations.
This is essential to the strip edges, where the presence of larger deformation can be detected.
Analytical and experimental approaches have been employed for the study of cold roll forming of simple profiles or pipe sections. In order to analyze more complex profiles and to obtain
in details the distribution of stress and strains in the formed sheet, numerical methods need to be
exploited.
Following this trend, a 3D finite element analysis has been involved in a previous work [3] to
study the deformation of the strip during and after the forming process. A classical Lagrangian
formulation was successfully used to compare the simulation with experimental data. However,
these 3D models become very CPU expensive due to the large number of finite elements that
must be used to get an accurate solution. An alternative approach, considering the Arbitrary
Lagrangian Eulerian (ALE) formulation [13, 1, 2, 5], has been investigated. This formalism
allows to obtain the stationary solution of the process with a minimal number of finite elements.
104
The efficiency of the ALE method is also demonstrated on a 2D metal cutting simulation.
Although the problem is extremely simplified, this application shows that the chip formation can
be easily computed avoiding complex remeshing operations.
This work has been introduced in the software METAFOR [13], the non-linear finite element
code developed at the ASMA/LTAS-MC&T, University of Liège, Belgium.
2 ALE formalism
In order to find the solution of steady state processes by numerical simulation with the classical
Lagrangian formulation, very large and useless meshes have to be considered. For example, when
dealing with the roll forming simulation, the whole sheet has to be finely discretized along its
length even if the results obtained between each set of tools dont need a refined mesh in the final
stationary state. However, these small finite elements cannot be enlarged because a fine mesh is
needed when they will reach a pair of rolls and go through it. They are thus required in order
to reach an accurate steady state solution. Moreover, each surface node is potentially in contact
with the corresponding upper/lower rolls as the simulation progresses in time. Consequently, the
CPU time soon becomes very large as the required accuracy increases.
Another approach is the well-known Eulerian formulation: the medium flows through the
mesh, which is fixed in space. However, boundary conditions are rather difficult to handle particularly frictional contact and free surfaces.
The Arbitrary Lagrangian Eulerian formulation was introduced to overcome these problems.
The mesh can be handled by the software, irrespective of the body motion, so that both previous
formulations can be obtained as particular cases if the mesh sticks to the body or is fixed in space.
In such a formulation, each time step is divided into two phases: the first one is purely Lagrangian
and the second one the ALE phase. The latter is a rezoning phase, where the nodes are moved
according to mesh quality considerations, followed by a convective Eulerian phase, where the
the values stored at the Gauss points are updated. In order to avoid oscillations and instability,
efficient convection algorithms have to be used. In the present paper, a Godunov-type update,
based on a finite volume method, is proposed.
3 Rezoning phase
3.1 Basic methods
The main task concerning the creation of the ALE numerical model is the definition of the movement of the mesh respectively to the body motion. There exists as many methods as different
mechanical problems. The rezoning strategies usually depend on the position of the node to be
moved on the CAD geometry of the model. They can be classified according to the topological
entity (vertex, edge, side, volume) it belongs to.
In our approach, nodes laying on the vertices may move in a Lagrangian or Eulerian way.
They may be fixed along one given direction.
Nodes laying on the edges are remeshed using a spline curve that goes through the Lagrangian
position of the nodes. The new position is computed from the initial curvilinear coordinate of the
node (mesh topology is then kept unchanged during all the simulation irrespective of the edge
deformation).
105
Figure 1: Isoparametric smoothing method.
The nodes inside the volume are moved using a transfinite mapping mesher or traditionnal
volumic smoothing algorithms.
3.2 Plane surfaces

Nodes laying on the free planar sides may be moved with the classical smoothing methods [9]
that are commonly used for improving the mesh quality after a remeshing operation. Among
these method the Laplacian smoothing is the most famous due to its simplicity. Unfortunately, in
its simplest form, this smoother is not able to handle non uniform structured mesh (the algorithm
tends to give each cell of the mesh the same area). In the case of the roll forming process, this
kind of mesh is very useful. The structure of the mesh helps for a given number of elements
to achieve a better accuracy on the results than an unstructured one. Moreover, it is efficient to
define small cells near the contact regions and larger cells far from them. Keeping this in mind,
many other methods, like Giulianis method [6], the angle smoother [14], area pull smoother
[10], have to be discarded.
In the case of a structured mesh of graded elements, the isoparametric smoother [7] seems to
be a good alternative. The basic idea is to reposition the node to the centre of an isoparametric
element associated with the neighboring nodes (see figure 3.2) :
x
n+1
X
1 nel
xn1 + xn3 xn2
=
nel i=1
(1)
where nel is the number of cells containing the node. This iterative method is usually speeded-up
by a Successive Over-Relaxation (SOR) algorithm.
Another technique available for planar surface meshed with a structured grid is the complete
remeshing with a transfinite mapping method [13, 5]. Compared to the previous ones, it has the
big advantage to be non iterative, thus very fast.
3.3 Curved surfaces

One of the biggest problem dealing with the three dimensional ALE formalism is the rezoning of
the nodes laying on curved boundary sides of the mesh. Due to discretization, the faceted surface
is not smooth as soon as the curvature is not zero. Consequently, the nodes cannot be moved on
the facets without changing the total volume of the 3D mesh and destroying the global shape of
the surface.
The extension of the one dimensional spline remesher used for the edges is then proposed
to deal with this problem. At each time step, during the rezoning phase, a spline surface is built
106
Figure 2: Initial surface mesh
Figure 3: Smoothing on the approximate tangent plane
over each side of the mesh. The spline is a set of bi-linear patches constructed on each facets,
considering cubic shaped edges. The resulting surface is not a full cubic patch (twist terms are
missing) but the curvature is well approximated and it is really faster. The shape of the cubic
edges are driven by the approximated normal vector defined at each node.
The surface rezoning is then an iterative process. For each node, an approximation of the
normal vector is computed (see figure 3) :
n=
nbf
X
ni
(2)
i=1
where nbf is the number of surface facets neighboring the node. ni = d1 d2 , the normal vector
at the considered node on the patch i. The vectors d1 and d2 are the edges of the patch i starting
from the node.
Figure 3.3 shows a surface mesh around a node before the rezoning phase. Once the normal
is computed, classical 2D smoothers may be used on the tangent plane (figure 3). The resulting
position is then projected on the spline described above (figure 4). The projection algorithm must
also be able to search the projection outside the neighboring facets if the rezoning method moves
the node far from its original position.
107
Figure 4: Projection on the interpolation spline
3.4 Eulerian boundaries

When dealing with stationary processes, a space region must usually be defined (by boundary
planes that we call Eulerian boundaries boundary lines in 2D) in order to prevent the mesh
from going outside this specified region. Great care must be taken with the nodes that may cross
an Eulerian boundary during the Lagrangian step. A direct projection of the node on the boundary
is not enough to prevent mesh distortions from occuring. That is why a specific rezoning strategy
has been created for them : the node that has crossed the boundary is placed at the intersection
of the most perpendicular edge starting from this node and the boundary.
4 Convection phase
Once the new mesh has been successfully computed, the convection step may begin. When non
linear problems are considered, some important values, like the stress tensor or the equivalent
plastic strain, are stored at the Gauss points and have to be updated from the Lagrangian mesh to
the new one. This convective step consists of solving a classical convection equation, which can
be written :
| + w j
=0
t
xj
(3)
where is a value stored at the Gauss point (e.g. a component of the stress tensor), wj is the
relative velocity between the new mesh and the Lagrangian mesh and is the coordinate system
associated with the new mesh.
Although this scalar equation is well-known in fluid mechanics, it is rather difficult to solve
it because is not a continuous field but is only defined at the Gauss points. Consequently, the
gradient cannot be easily evaluated.
In order to overcome this problem, the values can be extrapolated and averaged to the mesh
nodes and the gradient is computed from the resulting continuous field. But this first simple
method shows a large amount of numerical diffusion. More sophisticated methods must be used.
The Godunov-update technique was firstly introduced by Casadei, Donea and Huerta [4,
8]. This method can be useful on structured meshes of hexahedral hybrid finite elements (8
Gauss points are used to integrate all the values except pressure, for which 1 point is used to
prevent locking). It consists of dividing each finite element into eight (one for the pressure)
108
Figure 5: A 2D finite element and its division in 4 finite volumes
cells surrounding each Gauss point (the figure 5 shows the element split in a 2D case). The
finite volume mesh required for convecting the pressure is equivalent to the finite element mesh.
These auxiliary meshes are automatically built at the beginning of the simulation during the
preprocessing and are updated at each time increment to correctly follow the Lagrangian mesh.
The field to be transferred is assumed to be constant on each cell and thus discontinuous
across them. The finite volume problem is solved by the classical Godunov method and an
explicit Euler scheme is used for the time integration. The resulting update formula [4] is given
below for the cell s :
sn+1 = sn
Ns
X
t
2 Vs
fi (ic s ) (1 sign(fi )),
(4)
i=1
where t is the time step, Vs is the volume of the cell, Ns is the number of boundary facets of the
cell (6 in in the case of an hexahedron), ic is the value of the adjacent cell sharing the boundary
i with the cell s, is an upwind factor and fi is the flux of across the boundary i.
This explicit numerical scheme is conditionally stable. Practically, the largest allowable displacement of the mesh is the smallest length of the mesh. As it can be very restrictive for refined
mesh, a sub-stepping method has been added to manage larger displacements without the need
of creating a highly memory and CPU consuming implicit scheme. However if the number of
sub-steps is too large (this criterion is given by the user), the step is refused and restarted with a
smaller time step (leading to smaller displacements of the mesh).
5 Numerical simulation of metal cutting

The first application is two-dimensional. A stationary metal cutting process [12, 11] is modeled
in order to show the efficiency of the ALE formalism in complex situations. Metal cutting is
one of the most common manufacturing operation. However it is rather difficult to model due to
many physical factors : high strain rates, complex thermomechanical behavior of the material,
damage and wear, friction law, dynamic effects (vibrations), etc. In this paper, the problem is
extremely simplified in order to focus on the mesh motion. In particular, no separation law or
node splitting method is used near the crack tip. A basic elastoplastic material with linear
hardening is used. Thermal, dynamic and frictional effects are neglected.
109
Figure 6: Initial mesh of the ALE simulation
The initial geometry is shown on figure 6 (length=100mm and height=40mm). In this initial
state, a chip is already modeled and meshed in order to keep the same topology of the mesh during
all the computation time and thus to avoid remeshing. The geometry of the chip is not really
important because it is a guess of the final one and is taken circular at this time (radius=30mm).
The angle of the rigid cutting tool is 70 degrees.
The mesh is is kept eulerien almost everywhere except near the chip. The upper boundary
of the bulk metal is continuously remeshed using a cubic spline curve. The mesh of the circular
chip is cut with a boundary line (see section 3.4) in order to let the metal flow outside the mesh
while keeping the boundary straight. Boundary conditions are applied in the upstream boundary.
Thanks to the ALE formalism, the mesh can be refined near the main critical region that is the
crack tip.
The next figures (7, 8, 9 and 10) show the chip formation during the numerical simulation.
The final thickness of the chip and its shape have been automatically computed by the ALE
method.
6 Numerical simulation of cold roll forming process

6.1 Model definition
A symmetrical U-channel is formed by a process divided in 6 pairs of forming rolls (diameter =
210 mm) separated by 500 mm (figure 11). On each roll, the forming angle increases along the
forming direction (i.e. 15, 30, 45, 60, 75 and 90 degrees).
110
Figure 7: Zoom on the initial geometry of Figure 8: Chip formation intermediate rethe chip
sult
Figure 9: Chip formation intermediate result
Figure 10: Final shape of the chip
111
Figure 11: Initial configuration of the ALE simulation
The dimension of the sheet is 3500 x 200 x 1 mm and only one half of the process is modeled
due to symmetry. The material is elastoplastic (E = 210 GPa and = 0.3) with a linear isotropic
hardening rule e = 0e + hpl with 0e = 150 MPa and h = 500 MPa.
Frictionless rigid rolls are considered and the motion of the sheet is imposed on the symmetry
plane (actually, in the real process, the rolls are rotating and friction forces the sheet to advance).
This simplification avoids to consider the convection of friction in the ALE algorithm.
Contact is taken into account using the penalty method. Dynamic effects are neglected (implicit quasi-static algorithm).
A lot of different strategies can be used to reach the steady state of this process. For example,
the simulation could be divided into two steps. The first one could be purely Lagrangian. During
this step, the sheet would be introduced through the rolls until the front reaches the end of the
machine. At this time, a second ALE step, during which the mesh of the sheet would be fixed
in the rolling direction, could begin until the steady state is obtained. The main problem of
this procedure is the large number of contact nodes required during the first phase. Indeed, in
this case, each surface node may be in contact with any roll. Moreover, in order to get a good
convergence, a uniform mesh should be used (that is actually what we want to avoid).
An alternative strategy is then preferred : as far as only the final stationary state of the sheet
is interesting, the first step is modified in such a way that the computation is faster (but far from
the reality). The lower rolls are initially separated from the upper ones and the sheet is initially
placed between them (see figure 11). As the sheet is already in the good position according the
the rolling direction, the first step can be computed using the ALE formalism too. The lower rolls
go up one by one bending the sheet of metal (this actually looks like a deep drawing process).
Using this method, the mesh can be refined near the contact areas (lengths vary from 25
mm far from the rolls to 5 mm close to them) a limited number of contact elements can thus
be defined. In the perpendicular direction, smaller element sizes are also used close to the expected position of the fold (lengths vary from 10 to 2 mm). Consequently, the resulting mesh is
structured but non uniform (see figure 12).
The edges of the sheet are remeshed using spline curves. As the sheet is meshed using only
112
Figure 12: Refined mesh in the neighborhood of the contact regions
one element through the thickness, no volumic node rezoning strategy is needed. One boundary
plane is defined at the front of the mesh (beyond the last roll). This surface is used to cut the
Lagrangian mesh at the end of each Lagrangian step of each time step. At last, both main sides
of the mesh are modified using the spline method described in the previous section combined
with an isoparametric smoother. In order to improve the convergence rate, the nodal positions
are initialized with the previous positions.
6.2 Results
The results are shown in figures 13 and 14. The first step of the simulation is depicted in figure
13. At this moment, the rolls have bent the sheet and are in the correct position compared to the
industrial process. The movement of the sheet is carried on until the solution reaches a steady
state (see figure 14).
At the beginning of the second step, the front of the sheet is almost straight (see figure 15).
Due to the ALE convection, the free surface mesh deforms (figure 16) propagating the bending
from the rolls until the front. Later, the U-shaped sheet is obtained when the stationary solution
is reached (figure 17).
7 Conclusions and future work

A 2D stationary metal cutting simulation has been first presented. In this problem, the final shape
of the chip has been computed using the Arbitrary Lagrangian Eulerian formalism. Many improvements could be added to this simple model. The first one is, of course, a realistic separation
law at the crack tip.
A simulation of a 3D cold roll forming process has been presented. The ability of managing
efficiently the free surfaces of the sheet has been proven. However, this model can also be
113
Figure 13: Configuration when the rolls are in the correct position
Figure 14: Final stationary state
114
Figure 15: Free surface movement snapshot #1
Figure 16: Free surface movement snapshot #2
115
Figure 17: Free surface movement snapshot #3 (stationary solution)
improved : friction could be taken into account and the motion of the sheet could result from the
rotation of the rolls.
This simulation will then be compared to a Lagrangian one and to physical experiments. To
reach this goal, a special procedure must be created in order to compute efficiently the longitudinal elongation of the sheet. This result is easily available in the Lagrangian case but needs an
incremental updating procedure in the case of the ALE mesh.
Another interesting aspect is the computation of the springback when the sheet is formed and
leaves the tools. In this case, the rolls could be removed using a third Lagrangian step.
8 Acknowledgement
The support of the Walloon Region throughout this research under grant PROMETA N 01/1/4710
is gratefully acknowledged.
References
[1] R. Boman and J.-P. Ponthot. ALE methods for determining stationary solutions of metal forming processes.
In Proceedings of ECCOMAS 2000, Barcelona, Spain, 2000. on CDROM.
[2] R. Boman and J.-P. Ponthot. Finite element simulation of lubricated contact in rolling using arbitrary lagrangian eulerian formulation. Computer Methods in Applied Mechanics and Engineering, 193(39-41):pp.
43234353, 2004.
[3] Q.V. Bui, L. Papeleux, R. Boman, J.P. Ponthot, P. Wouters, R. Kergen, and G. Daolio. Numerical simulation
of cold roll forming process. In Proceedings of ESAFORM2005, 2005.
[4] F. Casadei, J. Donea, and A. Huerta. Arbitrary lagrangian eulerian finite elements in non-linear fast transient
continuum mechanics. Technical Report EUR 16327 EN, University of Catalunya, Barcelona, Spain, 1995.
116
[5] J. Donea, A. Huerta, J.-P. Ponthot, and A. Rodriguez-Ferran. Encyclopedia of Computational Mechanics,
Volume 1, chapter 14, pages 413437. Stein, E. and de Borst, R. and Hughes T.J.R., 2004.
[6] S. Giuliani. An algorithm for continuous rezoning of the hydrodynamic grid in Arbitrary Lagrangian-Eulerian
computer codes. Nuclear Engineering and Design, 72(2):205212, 1982.
[7] L.R. Hermann. Laplacian isoparametric grid generation scheme. Journal of the Engineering Division,
102:749756, 1976.
[8] A. Huerta, F. Casadei, and J. Donea. ALE stress update in transient plasticity problems. In E. Onate and
D.R.J. Owen, editors, International Conference on Computational Plasticity, pages 18651876, Barcelona,
Spain, 1995.
[9] S. Hyun and L.-E. Lindgren. Smoothing and adaptive remeshing schemes for graded elements. Communications in Numerical Methods in Engineering, 17:117, 2001.
[10] R.E. Jones. A self-organizing mesh generation program. Journal of Pressure Vessel Technology, Transactions
of the American Society of Mechanical Engineering, 96(3):193199, 1974.
[11] L. Olovsson, L. Nilsson, and K. Simonsson. An ALE formulation for the solution of two-dimensional metal
cutting problems. Computers and Structures, 72:497507, 1999.
[12] O. Pantale, J.-L. Bacaria, O. Dalverny, R. Rakotomalala, and S. Caperaa. 2d and 3d numerical models of
metal cutting with damage effects. Computer Methods in Applied Mechanics and Engineering, 193:4383
4399, 2004.
[13] J-P. Ponthot. Traitement unifie de la Mecanique des Milieux Continus solides en grandes transformations par
la methode des e lements finis (in French). PhD thesis, Universite de Liège, Liège, Belgium, 1995.
[14] T. Zhou and K. Shimada. An angle-based approach to two-dimensional mesh smoothing. In 9th International
Meshing Roundtable, pages 373384, New Orleans, Louisiana USA, October 2000.
117
Fracture Toughness Measurement

using Circumferentially-Cracked Round Bars
Marc Scibetta*, R. Chaouadi* and E. Van. Walle*
*SCKCEN, Boeretang 200, B-2400 Mol, Belgium
Abstract
The accuracy and the domain of applicability of approximate formulae used to evaluate the fracture
toughness of circumferentially-Cracked Round Bars (CRBs) are established using finite element
calculations. In the elastic plastic fracture mechanic (EPFM) domain, new formulae are developed
which increase largely the accuracy and the domain of applicability.
The loss of constraint is evaluated through the Q-factor allowing determining the domain of
validity of one-parameter fracture toughness characterisation. At the same time, an appropriate
"crack length to bar radius" ratio is proposed.
A micromechanical model of cleavage fracture in the transition region is proposed where both
the specimen size and loss of constraint correction are taken into account. This model allows to
measure plane strain fracture toughness using small CRB.
Keywords: fracture toughness, circumferentially-cracked round bars, -factor, loss of constraint,
micromechanical modelling.
Nomenclature
Abbreviations
CT
SENB
CRB
EPFM
LEFM
CMOD
CTOD
SSY
FB
FE
EDI
Compact Tension
Single Edge Notch Bend
Circumferentially-Cracked Round Bar
Elastic Plastic Fracture Mechanics
Linear Elastic Fracture Mechanics
Crack Mouth Opening Displacement
Crack Tip Opening Displacement
Small Scale Yielding
Finite Body
Finite Element
Equivalent Domain Integral
Material properties
E
YS
YS
KIc
Jc
JIc
Young's modulus
Poisson's ratio
Strain at yield strength
Yield strength
Fracture toughness (LEFM)
J at fracture instability prior the onset of stable
crack extension (EPFM)
J near the onset of stable crack extension (EPFM)
Other
Geometry
a
b
G
L
R
B
Crack length
Ligament length
Gauge length
Bar length
Bar radius
Specimen thickness
Eta factor
W
Energy density
U
Energy
C
Compliance
Load point displacement
Gauge extension
P
Load
K
Stress intensity factor
J
J-integral
Xel.
Elastic part of X
Xpl.
Plastic part of X
G(x), H(x), I(x) Dimensionless functions
118
Introduction
The measurement of fracture toughness is based on standard specimens such as Compact Tension
(CT) or Single Edge Notched Bend (SENB) specimen. However, specimen size requirements to
obtain valid measurements generally necessitate large or very large specimens. In this context, there
is an increasing interest [1 to 8] for the use of the small Circumferentially-Cracked Round Bar
(CRB).
As already point out in the literature, the presumed weaker size requirement is not the only
advantage of the CRB. Other advantages are the low cost to machine the specimen, the rotating
bending fatigue allowing easy precracking of specimens, the use of a standard tensile test fixture,
the absence of shear lips formation as the circumferential crack has no end in a plane stress region,
the possibility to load the specimen at very high loading rates [6] and the axisymmetry of the
specimen that avoids time consuming 3D finite element calculations.
Therefore, the CRB has been selected in different experimental programs [1 to 4, 6 to 8, 26],
which confirm the ability of small CRB specimens to measure the fracture toughness. However, the
comparison of these papers reveals significant differences in the method to evaluate experimental
data. Indeed, unlike CT or SENB specimens, a standard test method addressing specific points such
as: the specimen size and geometry, the specimen preparation including precracking, the apparatus
with special attention to the fixture and the extensometers, the procedure, the calculation and
interpretation of results, do not yet exist for the CRB.
Although recent theoretical work on the CRB was done [5], two limitations of the CRB remain.
These limitations are, according to Giovanola and al. [2], the lack of a well developed and validated
analysis method to evaluate experimental data and the absence of guidelines regarding the
minimum specimen dimensions that are required to obtain meaningful fracture toughness data.
The results and analysis presented in this paper address these two points as well in linear elastic
fracture mechanics (LEFM) as in elastic plastic fracture mechanics (EPFM).
In LEFM, the stress intensity factor at crack initiation KIc characterises the fracture toughness. A
review of the available stress intensity factor formulations is performed and compared to our finite
element calculations in order to give guidance on the choice of the formulation and to evaluate its
accuracy. Moreover, we will derive and assess a useful compliance function. The specimen size
requirements as found in the literature are discussed as well.
In EPFM, the equivalent of the stress intensity factor is the J-integral and the fracture toughness
is Jc (unstable fracture) or JIc (stable crack growth). Existing formulation and original proposals,
derived from performed finite element calculations to evaluate the J-integral from the load versus
displacement trace, are reviewed. The accuracy and the range of applicability of these formulations
are then evaluated using finite element simulations of CRB specimens having different crack
lengths and different material properties.
In EPFM, the domain of validity of a single parameter, i.e. the J-integral, governing the fracture
process is limited. As a consequence, the J-integral at crack initiation does not characterise alone
the fracture toughness of the material if the specimen size is too small. However, in the EPFM
domain, no specimen size requirement for the CRB is found in the literature. Therefore, the stress
field ahead of the crack tip, obtained from our finite element calculation, is evaluated in two
situations: a reference solution the so-called small scale yielding SSY, and the CRB application.
119
Furthermore, for cleavage governed fracture, the domain of validity can be further enlarged
when a correction for loss of constraint is taken into account. Based on the micromechanisms of
cleavage fracture and finite element calculations, such corrections are proposed in this paper.
P,
ez
e
R
b
L
er
Figure 1: Geometry of the circumferentially-Cracked Round Bar (CRB) submitted to a tensile load
P. is the load point displacement and G the gauge length.
Linear elastic fracture mechanics (LEFM)
This section is limited to the LEFM domain in which the load is proportional to the displacement
until fracture occurs. The load at rupture is used to determine the critical stress intensity factor. As
several stress intensity factor functions are proposed in the literature, their accuracy is evaluated by
performing finite element simulation. Moreover, we will deduce a useful compliance function and
investigate its accuracy. The specimen size requirements as found in literature, are reviewed as
well.
2.1
K formulation and compliance
Benthem and Koiter [9] developed a formula for the CRB based on asymptotic approximations to
evaluate the stress intensity factor:
P
b
K = G( )
R b2
with
G(x) =
ab
R
1
1
3
[1 + x + x 2 - 0.363x 3 + 0.731x 4 ]
2
2
8
(1)
(2)
where P is the tensile load, a the crack length, 2b the ligament diameter and 2R the bar
diameter (see Figure 1). Two other formulae derived by Bueckner [10] and Haris [11] are compared
see (Figure 2) as a function of the "crack length to bar radius" ratio.
Linear elastic finite element simulations of a CRB with a "crack length to bar radius" ratio
varying between 0.1 to 0.9 were performed. An elastic material with a Poisson's ratio of 0.3 is
chosen. The mesh contained 244 isoparametric 8-node axisymmetric elements with adequate mesh
refinement at the crack tip region. Accounting for symmetry, only a quarter of the 2D section of the
120
CRB was modelled. The stress intensity factor is calculated using the nodal displacements at the
crack lips.
Benthem and Koiter's function is within 2% accuracy of the finite element calculations (Figure
2) and is closer to finite element solution than alternative functions proposed by Bueckner [10] and
Haris [11]. Consequently, we recommend the Benthem and Koiter's function to calculate the stress
intensity factor.
1.2
1.1
FE
Benthem and Koiter [9]
Bueckner [10]
Haris [11]
0.9
0.8
0.7
0.6
0.5
0.4
0
0.2
0.4
0.6
a/R
0.8
1.2
Figure 2: Comparison of eq. (2) and eq. proposed in [10] and [11] with finite element calculations.
The compliance, which is the slope of the load versus displacement curve, is of particular
interest to check the load versus displacement record and to determine the elastic part of the
displacement for EPFM application.
First we can derive analytically a compliance formula [5, 12] from the Benthem and Koiter's
stress intensity function:
C =
with
H(x) =
G
1 - 2
b
=
1
+
4
R
H( )
2
G
R
P
ER
0.25
- 0.3531717 + 0168875
.
x 2 - 01325521
.
x 3 + 0.04880469 x 4
x
- 0.01106345 x 5 + 0.05044688 x 6 - 0.03803811 x 7 + 0.01669878 x 8
(3)
(4)
where C is the compliance, the gauge extension, E Young's modulus and Poisson's ratio.
Using the finite element calculation performed to obtain the stress intensity factor, we can
evaluate the accuracy of eq. (3). For G > 2R, an accuracy better than 4.5% is obtained
independently of the crack length. However, the accuracy of this equation deteriorates when the
gauge length decreases, as is shown in Figure 3.
However, for some applications, it is desirable to use a small gauge length. This can be obtained
through improvement of the accuracy of the compliance formula for small gauge length. Prior to the
development of a formula applicable for the full gauge length range, the crack mouth opening
compliance (CCMOD) is evaluated. This crack mouth opening compliance can be considered as the
limit of the compliance for gauge length tending to zero. By fitting finite element results, an
analytical formula with a 0.5% accuracy is obtained:
121
CCMOD =
with
a
b
I( )
ERb R
(5)
I(x) = 0.635 + 0.14 x + 0.926 x 2
(6)
Now, to obtain an accurate compliance formula for the whole gauge length range, we develop an
empirical formula using eq. (3) and (5):
C=
G
G
2
1
G
b
a b 2 R
2
R
1
=
(
)
(
)
(
)
e
+
+
4
1
H(
)
e
I
P E R 2 ER
R
b R
(7)
This relation has a maximum error of 12.5% independently of the G/R and b/R ratio. However,
when G>2R, eq. (3) is recommended as its accuracy is better than 4.5%.
10
9
8
error %
7
6
5
4
3
a/R=0.3
2
a/R=0.7
1
0
0
G/R
Figure 3: Accuracy of the compliance function eq. (3) by comparison to finite element results for
a/R = 0.3 and 0.7. The meshes contained 244 isoparametric 8-node elements.
2.2
Specimen size requirements
If the specimen is small or the fracture toughness is high, the specimen will probably undergo
plastic deformation. In this condition, LEFM cannot be applied as plasticity is not taken into
account. It means that the evaluation based on eq. (1) will largely underestimate the fracture
toughness. Different authors studied the specimen size requirements supported by experimental
results [4, 12] or finite element calculations [14]. The most restrictive condition for the CRB
resulting from these references in the range of 0.4 < a/R < 0.6 is [15]:
K
a, b 0.375 Ic
YS
(8)
where YS is the yield strength of the material.

This condition is 6.7 times less severe than the validity condition found in the ASTM E399
standard [18] for compact tension and three-point bend specimens. It should be noted that the
fulfilment of eq. (8) means that specimen size does not significantly influence the measured fracture
toughness. However, none of the papers [4, 13, 14] have clearly demonstrated that this value is the
plane strain fracture toughness as evaluated according to ASTM E399 [18].
122
Elastic plastic fracture mechanics (EPFM)
This section, devoted to the EPFM domain, requires careful attention as the fracture toughness
evaluation of the CRB specimen is not sufficiently documented in literature and as the testing of a
small specimen results in plastic deformation.
The J-integral definition is adapted to an axisymmetrical configuration. An -factor, which
allows practical evaluation of the J-integral from experimental records, is introduced. A review of
existing -factor formulations is performed and a description of the finite element methodology is
given. These finite element calculations are used to:
derive new -factor formulations more appropriate for strain hardening materials,
give guidelines for the experimental evaluation of the load point displacement,
evaluate the accuracy and the domain of applicability of the different -factor formulations,
evaluate the optimal "crack length to bar radius" ratio,
establish a specimen size requirement in order to avoid loss of constraint,
establish a loss of constraint correction for cleavage governed fracture.
3.1
J-integral for axisymmetric configuration
In EPFM, prior to the investigation of the different approximate analytical formulae, it is required
to adapt the definition of the J-integral to the axisymmetrical configuration of the CRB specimen.
Indeed, the original definition of the J-integral as proposed by Rice [19] for 2-dimensional plane
strain or plane stress, is path dependent in axisymmetrical configurations [20].
Bergkvist and Huong [20] proposed, for externally cracked round bars, an adapted path
independent integral, which incorporates an additional surface term:
J = -
( W n r - n u r )
r
d +
b
(W - ) d
= r, z
= r, z
(9)
where W is the energy density, any path around the crack, the area delimited by . All
quantities are expressed in cylindrical coordinates (Figure 1).
A convenient and accurate way to evaluate the J-integral from the finite element results is to
transform eq. (9) into an Equivalent Domain Integral (EDI) [15, 21]:
J EDI =
( W r q - q u r ) b + W q -
q d
= r, z
= r, z
(10)
where q is any continuous function with a value of one at the crack tip and zero at the border of .
In this work, the weight function q is a quadratic function of the radial coordinate with a slope of
zero at the crack tip and at the border of [15].
The J-integral can also be interpreted as an energy rate release per unit of crack area:
123

J =
2
1 U pl.
K2 (1 - )
+
2 b b
E
pl.
with Upl. =
P d pl.
(11)
pl.
where pl. is the plastic part of the displacement and Upl. is the plastic energy dissipated during the
test.
For the practical evaluation of the J-integral from the load versus displacement record, an factor is introduced in the J formulation:
J =
2
Upl.
K2 (1 - )
+
E
b2
(12)
In addition, the -factor allows us to introduce an incremental evaluation of the J-integral for a
growing crack based on the deformation theory [22] JD or on the J-modified [23] JM, namely:
2
K2 (1 - )
+ J i+1
pl.
E
J i+1 =
J pl.
( a i+1 - a i ) +
a
i
+
J i+1
D pl. = J D pl.
pl.
i
+
J i+1
M pl. = J M pl.
J pl.
=
pl.
J ipl.
b
(13)
J pl.
i
(i+1
pl. - pl.)
pl.a
(14)
J pl.
i
(i+1
pl. - pl.)
pl.a
(15)
b

2 - 2 +
a
(16)
pl.
i
J pl.

+ Pi+1
Ji
= pl.
+ P
pl.a
pl.a
2 b2
(17)
The adaptation of the J-integral to axisymmetrical configurations is now established. However,

eq. (12) to (17) are not of practical use as long as no accurate value of the -factor is known. This
leads us to review existing -factor functions.
3.2
Review of existing -factor formulations
1) Rice [24] established a J-integral formula for deep crack using dimensional analysis. His
formula combined with the definition of the -factor gives:
= 1.5 - 0.5
P pl.
U pl.
a/R > 0.65
(18)
The crack is considered to be deep if the plastic zone is confined into the ligament. From finite
element simulations using rigid perfectly plastic and strain hardening materials, this condition is
satisfied if a/R > 0.65.
2) Based on experimental results on specimen with 0.35 < a/R < 0.65, Beremin [26] developed a Jintegral formula, which can be easily transformed in terms of the -factor:
= 0.5
P pl.
U pl.
0.35 < a/R < 0.65
(19)
124
3) For shallow cracks, Itho [25] developed an equation based on dimensional analysis:
=
b P pl.
- 1
2a U pl.
a/R < 0.35
(20)
A crack is shallow if the gross section is plastic under a tensile loading. This condition is
satisfied if a/R < 0.35.
4) For a rigid perfectly plastic material, it is easy to establish an -factor as a function of the limit
load PL. The combination of eq. (11) and (12) gives:
= 0.5
b
PL
P L
b pl.
(21)
Using the limit load given by Miller [27], which is exact for a deep crack (a/R>0.69) according
to the Tresca criterion, we obtain:
2.85 for a / R > 0.65
PL
=
R for a / R < 0.65
b2 Y
b
(22)
where Y is the flow stress, defined as the average of yield (YS) and tensile strength (TS). The factor is [30]:
1 for a / R > 0.65
=
0.5 for a / R < 0.65
(23)
To improve the accuracy of the limit load solution, finite element calculations are performed
using a rigid perfectly plastic material in conjunction with the Von Mises criterion [28, 29]. The
finite element software uses an algorithm based on the modified Markov variational principle [29]
in order to compute directly the limit-state without performing step-by-step computation.
Convergence analysis shows that results are known with 1% accuracy.
Figure 4 shows a large difference between the present finite element limit load results and
Miller's results. For deep cracks the 5% difference is due to the fact that Von Mises criterion allows
up to 15% higher limit load than the Tresca criteria. Figure 4 shows that a/R=0.7 is the pivot point
between confined plasticity into the ligament and largely spread plasticity.
The best fitting curve of the data presented in Figure 4, is:
3
for a / R > 0.7
PL
R 3
R 2
R
=
2
- 0.6539 ( ) + 0.03738 ( ) for a / R < 0.7
-1.497 + 3.11352
b Y
b
b
b
(24)
thus:
1
R
R
- 0.01869 ( ) 3
-1.497
+
1.55676
=
b
b
R 3
R 2
R
-1.497 + 3.11352 b - 0.6539 ( b ) + 0.03738 ( b )
for a / R > 0.7

for a / R < 0.7
(25)
125
Lei and Neale [5] derive a similar formulation using a quadratic polynomial for the limit load.
The coefficients of the polynomial were determined explicitly by enforcing continuity and
derivability with eq. (22) at a/R=0.65 and noting that PL/(b2Y)=1 for b=R.
for a / R > 0.65
2.85
R
PL
a
a 2
=
2
( ) 1 0.00771( ) 1522869
.
( ) for a / R < 0.65
b2 Y
b
R
R
(26)
thus:
1
0.00771 + 3.05739
b
=
R
a
a
2R
.
1 - 0.00771
- 152869
( )2
R
R
for a / R > 0.65
(27)
for a / R < 0.65
The relative difference between eq. (25) and eq. (27) is less than 10% in the range a/R from 0.15 to
1 except for a/R near 0.65 where differences reach 15%. This difference can be explained by the
choice of the pivot point between confined plasticity and largely spread plasticity. As Miller's
solution is only exact from a/R = 0.69 and not from a/R=0.65, the pivot point should be a/R=0.69.
Using a/R=0.69 as pivot, the analysis with a quadratic polynomial gives a relative difference with
eq. (25) of less than 3% in the range a/R form 0.36 to 1. Consequently, eq. (25) is recommended
and will be used in this paper.
FE
Fit eq. (24)
Miller [23]
Lei [5]
PL/( Y b2)
3
2.5
2
1.5
1
0
0.2
0.4
0.6 a/R
0.8
1.2
Figure 4: Dimensionless limit load. Finite element results are obtained from the modified Markov
theorem (Von Mises criteria). Convergence analysis shows that results are known with 1% accuracy
[28, 29].
3.3
Finite element methodology
Finite element simulations of a CRB with "crack length to bar radius" ratio varying from 0.1 to 0.9
are performed.
To model actual material behaviour, the incremental theory of plasticity is used in combination
with an isotropic strain-hardening material based on Von Mises criterion with a uniaxial true stressstrain curve described by a power law:
126

if < YS

YS
n
=
YS
if YS
YS
with
E YS = YS
(28)
(29)
To cover a large variety of materials, 3 hardening exponents (n = 0.05, 0.1, 0.2) and 3 different
yield stresses (E /YS = 250, 500, 1000) are selected. A Poisson's ratio of 0.3 is chosen for all
analyses. This choice results in 9 different materials per crack length and a total of 81 finite element
analyses.
Each meshing contains more than 250, 8-node axisymmetric elements with reduced Gauss
integration. Accounting for symmetry, only a quarter of the 2D section of the CRB was modelled.
Because of large geometry changes, the updated Lagrangian procedure is used to account for large
strains and displacements (this hypothesis was not done in the previous analysis [15]); to avoid
large mesh deformation and overlapping at the crack tip, an initial blunted mesh is used. In
accordance with [16, 17, 21], the initial crack tip radius is chosen such that the Crack Tip Opening
Displacement (CTOD) is at least 5 times larger than the initial one. The smallest element located at
the crack tip is typically one fifth of the initial crack tip radius.
The load is simulated by a vertical displacement of the nodes situated at the outer face (L/R=8).
The maximum displacement level is such that:
YS
5 J EDI
= 1
min(a, b)
(30)
where JEDI is calculated using the EDI method (eq. (10)). The domain used in this method is
chosen to be a circle surrounding the crack tip with radius =min(a,b). Although J is theoretically
path independent under specific assumptions [15], it is observed that J is slightly sensitive to the
domain . To better estimate the importance of this sensitivity, the difference between J evaluated
with a radius r= and J evaluated with a radius r=/2 is reported in Figure 5. This Figure shows that
the path independence is less verified for higher loads, suggesting that J is losing its significance as
a single parameter governing fracture. Similar results are found with different material properties
and different "crack length to bar radius" ratio. The dependence of the J-integral on the contour was
found less pronounced when a small strain model was used [15, 5]. The reason is that the less
realistic small strain model is in better agreement with the needed assumptions to obtain a J-integral
independent of the chosen path [15].
3.4
Proposed -factor supported by finite element calculation
1) To take into account the strain-hardening of engineering materials, a new equation is developed
based on a simple fit of our finite element results.
= - 0.155 + 2.06
a
a
- 1.13 ( ) 2
R
R
(31)
The previous fit can be improved by noticing that the -factor is load dependent and that =1 for
a very small plastic region [15]:
127

a
a
10875
.
.
- 2.06
+ 113
. ( )2
1155
R
R
= 1b
1 + 0.007
pl.
(32)
where pl. is the plastic part of the load point displacement based on a specimen with L/R=8.
16
14
finite strain
small strain
100 (J -J/2)/J
12
10
8
6
4
2
0
0
0.1
J/(aYS)
0.2
0.3
Figure 5: J-integral path dependence as a function of the load level. The elastic plastic finite
element calculations are made using a/R=0.5, E/YS=500, n=0.1.
2) Finite element results show, as will be seen later, that none of the previous equations give
satisfying results for shallow cracks. For this configuration, the applied load leads to a stress in the
ligament and the bar greater than the yield strength. The whole bar undergoes plastic deformation.
We therefore suggest to split the plastic energy into two parts: the one induced by the crack ( U crack
pl . )
bar
and the one due to the fact that the bar is plastified ( U pl. ).
U pl. = U crack
+ U bar
pl .
pl .
(33)
Of course, if the yield stress is not reached in the bar, no plastic energy is dissipated in the bar:
if
P
< YS
R 2
then
U bar
pl . = 0
(34)
To link the plastic part of J to the energy dissipated due to the crack, an '-factor is introduced:
2
U crack
pl .
K2 (1- )
J =
+ '
b2
E
(35)
The plastic energy dissipated in the bar can be evaluated by integrating the constitutive equation
multiplied by the plastified volume:
P
2 R
U bar
d pl.
pl . = X R
2
YS
with
P
> YS
R 2
(36)
where X is the plastified length of the bar that is not plastified due to the crack.
Integration of eq. (36) using a power law hardening (see eq. (28)) gives:
128
1
1
1
+1
1
2
n
U bar
=
X
<
>
1 < > 2 +
pl .
YS YS
2
2
n +1
where < >=
P
YS R 2
(37)
and n is the strain hardening exponent.
The analysis of the plastic zone size before the yielding of the whole bar gives the length of the
bar that is plastified due to the presence of the crack. This leads to the good approximation:
b
X = G 4 R
R
(38)
From our finite element calculations an empirical equation for ' is derived:
' = 0.03 + 13
.
a
a
a
+ 0.333 ( )2 103
. ( )3
R
R
R
(39)
3) In cases where the crack mouth opening displacement (CMOD) is measured, it is interesting to
provide a formula based on the plastic energy dissipated under the load versus CMOD record.
Moreover, this type of formulation will eliminate the problem of removing the plastic energy
dissipated into the bar. The J-integral can therefore be written as:
2
U CMOD
pl .
K2 (1- )
CMOD
J =
+
b2
E
CMOD pl .
with
CMOD
pl .
P dCMOD pl.
(40)
where CMODpl. is the plastic part of the crack mouth opening displacement.
Using the finite element results, an empirical formula can be derived:
CMOD
a
a
a
1.0875 0.4 + 0.91

318
. ( ) 2 + 2.2 ( ) 3
R
R
R
= 1 b
1 + 0.0035
CMOD pl.
(41)
3.5
Experimental determination of the plastic part of the load point
displacement
Most of the previous formulae make use of the load point displacement. However, from the
experimental view point, the load point displacement needed to calculate the plastic energy is
evaluated by measuring the gauge extension using an axial extensometer. Of course, the use of
small gauge lengths can lead to an underestimation of the load point displacement. However, as
long as the plasticity remains into the gauge length, the plastic load line displacement is
approximately equal to the plastic gauge displacement (see Figure 5). We thus propose to evaluate
the plastic part of the load point displacement using:
pl. pl.
(42)
To investigate the accuracy of this approximation, we use the previous finite element calculation.
The analysis shows that eq. (42) has a 5% accuracy as long as G2R and a/R0.4. The degradation
129
of the accuracy with decreasing gauge length is illustrated in Figure 5 for a particular configuration
a/R=0.5, E/YS=500, n=0.1.
15
/R=0.047
/R=0.023
/R=0.012
/R=0.0059
(pl.-pl.)/ (%)
10
5
0
-5
-10
0
G/R
Figure 6: Comparison of the plastic load point displacement with the plastic gauge length
displacement for four imposed displacements. The elastic plastic finite element calculations are
made using a/R=0.5, E/YS=500, n=0.1 as described in the previous paragraph.
3.6
Results of -factor comparison and discussion
The -factor formulae are not equivalent, and each of them has a specific domain of applicability.
To select the more adequate formulation, the accuracy is evaluated through comparison to finite
element calculations for different crack lengths (a/R = 0.1 to 0.9).
The finite elements simulations allow comparison to be made between the calculated J-integral
using the EDI method, JEDI, and the J-integral, J, using an -formula. The calculated load versus
displacement curve is used to obtain the dissipated energy. A measure of the relative error is:
J - J EDI
100
(43)
J EDI
This error depends on many parameters, such as: the selected -formula, the "crack length to bar
radius" ratio a/R, the material properties (yield strain and strain hardening) and the load level. To
simplify the interpretation of results, a mean and a maximum error are defined, for a given formula and a specific "crack length to bar radius" ratio:
e mean = 100
1
9
n = 0.05, 0.1, 0.2
E
YS
emax = 100
max
n = 0.05, 0.1, 0.2
YS
max
= 1000, 500, 250
max
= 1000, 500, 250
max 0
max
= 0 to max
J - J EDI
J EDI
J - J EDI
J EDI
(44)
(45)
where emean is the average error and emax is the maximum error over different material properties and
different load levels. A positive or negative value of emax indicates that for the maximum error, the
J-integral is over-estimated (+) or under-estimated (-).
130
The results are summarised in Table 1. From this table, one can see that:
the range of applicability of Rice's formula is limited to deep cracks, a/R > 0.65 and overestimates the J-integral.
the range of applicability of Beremin's formula is for mid-crack length: 0.75 > a/R > 0.35.
Itho's formula for shallow cracks has no range of applicability. The assumptions made to derive
his equation should be verified.
eq. (25) has a wide range of applicability when a/R > 0.35, however the J-integral is overestimated.
eq. (31) and (32) have the same range of applicability a/R>0.35 and a very good accuracy.
eq. (39) can be applied to the whole a/R range and is very accurate but requires the knowledge of
the strain hardening exponent.
eq. (41) can be applied to the whole a/R range and is the more accurate formula for deep cracks.
The disadvantage of this formula is that accurate measurement of the CMOD is required.
Comparison of the different formulations allows to recommend of eq. (32) for a/R>0.35, as long
as P/(R2)<YS. For shallow cracks, depending on the availability of the CMOD or strain hardening
exponent, eq. (39) and (41) are selected.
A part of this work is already covered by Lei in [5]. He selected three crack lengths a/R= 0.75, 0.5,
0.25 and three material properties n=0.1, n=0.2 and a bi-linear hardening material. Analyses were
performed using a small strain model. The analysis was limited to three -factor formulation: Rice's
formula, Itho's formula and a similar equation to eq. (25). Results presented in [5] confirm that:
the range of applicability of Rice's formula is limited to deep cracks, and over-estimates the Jintegral.
Itho's formula for shallow cracks is not accurate.
The -factor from the limit load function over-estimated the J-integral for a/R= 0.5 and 0.75. For
a/R=0.25 the J-integral is largely over-estimated.
Table 1: Difference between each formulation and finite element results for different "crack length
to bar radius" ratio (a/R). Bold is used to enlighten regions for which the error is less than 25%.
0.9
0.8
0.7
0.6
a/R
0.5
0.4
0.3
0.2
0.1
emean
emax
11
33
12
37
16
44
24
58
44
85
78
168
144
318
309
591
1126
1867
Beremin [26] emean

eq. (19)
emax
21
-42
20
-42
20
-41
17
-37
11
-29
15
68
51
163
141
335
613
971
Rice [24]
eq. (18)
Itho [25]
eq. (20)
emean
emax
69
-98
68
-99
69
-101
71
-104
74
-110
66
-99
54
-92
46
127
398
946
Limit load
eq. (25)
emean
emax
16
29
18
31
21
37
7
19
7
22
13
48
24
83
47
123
157
310
Fit 1
eq. (31)
emean
emax
2
-6
3
-6
4
-8
5
-11
6
-16
11
44
18
68
22
64
30
48
131
3.7
Fit 2
eq. (32)
emean
emax
1
4
1
4
2
-5
3
-5
4
14
9
40
17
64
24
110
83
143
'
eq. (39)
emean
emax
2
6
2
7
2
-6
4
-10
6
-17
11
-32
15
-39
18
-39
31
-81
CMOD
eq. (41)
emean
emax
5
-9
3
-6
3
-7
5
-9
5
-12
5
-11
6
-12
10
-18
10
25
Specimen size requirements
In EPFM, when the plastic zone size reaches the external surfaces, the constraint can decrease
drastically. To estimate the best a/R range that allows the highest values of valid fracture toughness,
the J level at which the plasticity reaches a specific bound is plotted as a function of a/R in (Figure
7). This shows that the best a/R range, which maximises J before widely spread plasticity, is
between 0.4 to 0.6.
crosses the z axis
reaches the external surface
the ligament is fully plastic
0.08
0.07
0.06
J/(RYS)
0.05
0.04
0.03
0.02
0.01
0
0
0.2
0.4
a/R
0.6
0.8
Figure 7: Plastic zone size evolution for different "crack length to bar radius" ratio (a/R). The
results are based on finite element calculations with n=0.1 and E/YS=500.
0
-0.2
a/R=0.7
-0.4
Q
a/R=0.5
a/R=0.3
-0.6
-0.8
-1
0.001
0.01
J/(aYS)
0.1
Figure 8: Q-factor as a function of load level for different "crack length to bar radius" ratios. The
material properties are n=0.1 and E/YS=500.
132
From the previous analysis it is useful to recommend a "crack length to bar radius" ratio.
However, this approach is unusable to quantify a specimen size requirement. Consequently, an
approach based on the constraint factor will now be investigated.
The one-parameter J characterisation of the fracture toughness is valid as long as the stress field
at the crack tip is entirely known through J:
ij = f ij (
r
) = f ij ( r*)
J YS
(46)
A constraint factor is introduced to evaluate the deviation from eq. (46):

Q =
*
= 2)
zz ( r * = 2) - SSY
zz ( r
YS
(47)
where SSY refers to the small scale yielding plane strain solution. This Q-factor is sometimes used
as a second parameter [31] characterising rupture. To do so, some complementary verification has
to be done: independence of Q as a function of r* and self-similarity of stress contour. In the current
approach, the Q-factor is only used as a parameter characterising the deviation from the SSY
condition. Figure 9 illustrates that the stress field is controlled by the J - Q parameters.
As the fracture toughness of brittle material is highly sensitive to the Q factor [22], a Q-factor lower
than -0.1 is considered to be significant. For the studied material (n=0.1, E/YS=500) the SSY
solution is established using a large strain model to take into account large strains located in the
blunting zone. The reference value used in eq. (47) is: SSY/YS=3.46.
Three geometries from deep to shallow crack (a/R=0.7, 0.5, 0.3) are chosen. The calculated Qfactor is considered to be valid when the CTOD is at least 5 times larger than the initial one. This
requirement leads to different meshing with various initial crack tip radii.
Figure 8 shows the Q-factor for the three geometries. For low load levels, independently of the
"crack length to bar radius" ratio, a loss of constraint is observed. For shallow crack the constraint
decreases with increasing load, while it increases for deep crack, which is a confirmation of
Giovanola [2] results.
SSY
J/(a sys)=0.000586
J/(a sys)=0.00375
J/(a sys)=0.0336
J/(a sys)=0.241
YY/YS
2
0
rYS/J
Figure 9: Opening stress versus dimensionless distance from the crack tip. The material properties
are n=0.1 and E/YS=500.
133
Negligible loss of constraint (Q>-0.1) is observed in the range 0.3<a/R<0.7 when:

Jc
YS
(48)
(1 2 ) K 2Ic
K2
5 2Ic
E YS
YS
(49)
a, b 2500
which leads to:

a , b 2500
This condition is two times more severe than the ASTM E399 requirement and thirteen times
more severe than eq. (8). The analysis demonstrates that even if remote load versus displacement is
linear (fulfilment of eq. (8)), the stress at the crack tip can be subject to loss of out of plane and in
plane constraint. This can be explained by the fact that at a very low load level, the plastic zone
crosses the z axis resulting in a loss of plane strain condition and, consequently, in a loss of
constraint. The drawback of eq. (47) is that this leads to very large specimens. As an example, a
material with a low toughness 40MPa/m and a yield stress =500MPa leads to a specimen diameter
of 64mm and the tensile machine should provide a load as high as 1000kN (100 tons).
3.8
Correction for loss of constraint (stress controlled fracture mechanics)
The size requirement can be much relaxed in the lower shelf and the transition region if a correction
factor that takes the loss of constraint into account can be developed. This type of correction has
already been proposed by Anderson and Dodds [34] for three-point bend specimens. The loss of
constraint correction is based on a micromechanical model of fracture assuming the hypothesis that
the failure probability F is a function of the cumulative volume V over which the principal stress is
equal to or greater than a certain stress level 1.
F = F[ V( 1 )]
(50)
This hypothesis is supported by the fact that initiation of transgranular cleavage fracture is
associated to the probability of finding a triggering microfeature near the crack tip. We now apply
this micromechanical model to the CRB geometry.
Assuming that negligible crack growth is observed in a loaded cracked specimen, the volume
over which the maximum principal stress is equal to or greater than a certain value 1 is:
V( 1 ) = B
J2
A( 1 )
2
2
YS
YS YS
(51)
where A is a dimensionless function and B is the specimen thickness considered to be equal to 2b

for the CRB case.
Consider a finite body (FB) and an infinite plane strain cracked body from the same material
with a crack front length of one inch. In the infinite cracked body, the Small-Scale Yielding (SSY)
condition is satisfied because the plastic zone is infinitesimally small compared to all other
characteristic lengths and is embedded within a linear-elastic field. The probability of failure is the
same in the two specimens if the volume V(1) is identical in both specimens. Applying eq. (51) for
the two geometries gives:
K 1T SSY
)
B FB A FB (
YS
= K FB
1
)
B1T A SSY (
YS
1/ 4
(52)
134
If the ratio AFB/ASSY is independent of the triggering stress 1, then the correction to obtain the
SSY value is independent of the triggering stress. In this condition, eq. (52) becomes of practical
interest for fracture toughness testing evaluation. If no loss of constraint is observed, then
AFB/ASSY=1 and eq. (52) reduces to the well-known B1/4 size correction. Moreover, eq. (52) can be
simply extended to ferritic steels where a threshold fracture toughness justified by experimental and
theoretical consideration [32, 33] is introduced.
B A
K 1T SSY = ( K FB 20) FB FB
B1T A SSY
1/4
+ 20
(MPa m )
(53)
The A function in SSY (ASSY) is obtained by imposing displacement of the first term of the
linear elastic solution to a circular region containing an edge crack. The analysis of finite element
results allows to calculate the area where the maximal principal stress is greater than a given value
1. The analysis is performed using 8 different flow curves covering a wide variety of material
properties: 3 hardening exponents (n = 0.05, 0.1, 0.2) and 3 different yield stresses (E /YS = 250,
500, 1000). Figure 10 shows 3 different regions. In the first region, the area is independent of the
material properties and is governed by the elastic field. The second region is governed by the strain
hardening coefficient, which is in accordance with the HRR field. In the third region, the area is a
function of both the strain hardening coefficient and the yield stress. In this region, it is observed
that above a given stress ratio, the area for which the stress is larger than a given value tends
towards zero. It means that no area ahead of the crack tip is submitted to a stress higher than a given
value, which depends on the strain hardening and the yield stress. This maximum stress is obtained
owing to the fact that the finite element simulations are done with a finite strain model and do not
allow an infinite stress ahead of the crack tip as is found in the more simplified HRR field.
Prior to the development of a loss of constraint correction, it is required to assess the effect of the
trigger stress 1. To study this effect, (ASSY/AFB)1/2 is evaluated in Figure 11. This Figure shows
that the correction factor is independent of the trigger stress in the range 2.4<1/YS<3.4 as long as
the J/(aYS) < 0.1. In fact the two-parameters fracture mechanics model break down for extensive
deformation for which the self similarity of stress field is lost [22].
n=0.2 E/s=250
n=0.2 E/s=500
n=0.2 E/s=1000
n=0.1 E/s=250
n=0.1 E/s=500
n=0.1 E/s=1000
n=0.05 E/s=250
n=0.05 E/s=500
1E-1
1E-2
ASSY
1E-3
1E-4
1E-5
1E-6
1E-7
1E-8
1
4 1/YS 5
Figure 10: Dimensionless area function in SSY condition. Finite element calculations are performed
with a mesh containing 423 8-node element with reduced Gauss integration. The updated
Lagrangian procedure is used in conjunction with an initial blunted mesh as described in previous
paragraph.
135

5
0.43
0.21
(ASSY/ACRB)0.5
4
J/(aYS)
3
2
0.076
0.001
0.00025
0
2.4
2.6
2.8 1/YS
3.2
3.4
Figure 11: Loss of constraint correcting as a function of the critical stress. The finite body is a CRB
with a/R=0.5, n=0.1 and E/YS=500.
To study the effect of the crack length, three CRB geometries are studied. Finite element results
are fitted in a log-log plot using the linear regression method (see Figure 12). A study of the
accuracy of the proposed solution was performed and shows that results are sensitive to meshing, to
initial CTOD and 1/YS. The accuracy of the determined ACRB/ASSY is about 40%, resulting in an
acceptable 10% accuracy on the correction factor (ACRB/ASSY)1/4. However, this sensitivity and
especially the sensitivity of 1/YS increases with the load level. As a consequence, we suggest to
restrict the use of this approach to the requirements:
a
1 J FB
0.2 YS
(a / R 0.5) (0.6 a / R ) J FB
b 10
+
012
.
0.2
YS
1 J FB
b
for
012
. YS
for
a/R < 0.5
(54)
for
0.5 < a/R < 0.6
(55)
0.6 < a/R
(56)
As shown in Figure 10 the strain hardening coefficient has an important effect on the ASSY
function and might consequently be an important parameter to consider in the analysis of the
correction factor. Therefore, the effect of the strain hardening exponent is evaluated and presented
in Figure 13. Results show that materials with high strain hardening exponent are less sensitive to
loss of constraint, which is also confirmed in the case of three-point bend specimens [34].
Discussion
In the LEFM section, the stress intensity functions were reviewed and compared to finite element
calculation. The stress intensity function of Benthem and Koiter eq. (1) was found in very good
agreement with the present finite element calculation. A maximum error of 2 % was found in the
whole a/R range. A compliance function eq. (3) was derived from the of Benthem and Koiter's
function. This compliance function is also in very good agreement with the finite element
calculation as long as the gauge length is sufficiently large, G>2R. A maximum error of 2 % was
found in the whole a/R range for G>2R. Based on the finite element calculations a compliance
function for short gauge length eq. (7) was developed giving a maximum error of 12% for the
whole a/R range and G/R range. This new compliance function for small gauge length is of
practical interest in testing the CRB for which a direct measurement of CMOD is performed. The
specimen size requirements to remain in LEFM were also reviewed. It reveals that the more
136
restrictive specimen size requirement is still 6.7 less severe than current ASTM E399 standard eq.
(8). This should allow the use of smaller specimens than standard bend specimens. However, the
experimental equivalence of small CRBs versus large bend specimens is not yet clearly
demonstrated.
1
0.6
0.6
ACRB/ASSY
0.4
0.35
0.25
0.25
0.17
0.17
0.11
0.1
0.06
0.06
a/R=0.7 n=0.1
0.03
0.03
a/R=0.5 n=0.1
a/R=0.3 n=0.1
0.015
0.01
0.001
J/(aYS)
0.01
0.1
Figure 12: Effect of "crack length to bar radius" ratio on SSY correction using 1/YS=3. The
material properties are n=0.1 and E/YS=500.
a/R=0.5 n=0.2
a/R=0.5 n=0.1
a/R=0.5 n=0.05
1
0.6
ACRB/ASSY
0.4
0.3
0.3
0.17
0.17
0.15
0.15
0.1
0.11
0.04
0.03
0.015
0.01
0.001
0.01
J/(aYS)
0.1
Figure 13: Effect of strain hardening exponent on SSY correction using 1/YS=3. The CRB has
a/R=0.5 and E/YS=500.
In the EPFM section, the J-integral was adapted to axisymmetric configuration and an -factor
was introduced for the practical evaluation of the J-integral in experimental testing. A review of
existing -factors was performed. One of them, derived from the limit load eq. (23), showed a
discontinuity at a/R=0.65. This discontinuity was identified to be a result of a poor estimation of the
limit load function. Consequently, finite element calculations were performed in order to better
estimate the limit load function that led to eq. (25). The comparison of eq. (25) to a similar
expression found in the literature eq. (27) showed an abnormal difference round a/R=0.65. This was
137
identified to be an erroneous choice of the pivot point between confined and largely spread
plasticity in eq. (27). An extended number of finite element analyses were performed which lead to
the development of three new -factor formulations. The first one eq. (32) is based on a simple fit
of the finite element results. The second, eq. (39) is based on splitting the dissipated energy into two
components and is a tentative to better describe a shallow crack for which the plasticity is not
confined in the ligament. The last one, eq. (41), developed for experiments for which the CMOD
can be measured during a test. All -factors except eq. (41), are using the load point displacement.
However, experimental displacements are generally gauge length extensions measured with
extensometers. The plastic part of the load point displacement and the plastic part of the gauge
length extension eq. (42) are found to be very similar (less than 5% difference). This equivalence is,
however, restricted to large gauge length G>2R and deep crack a/R>0.4. For this reason it is
recommended to perform experiments with extensometer having a gauge length larger than 2R. A
comparison of the different -factor formula to finite element result was performed. This
comparison showed that the -factor formulae found in the literature are not very accurate (20%
difference with finite element calculation was generally found in the range of applicability). The
developed -factor from finite element fitting gives better result. However, -factor for shallow
crack cannot be capture by the fit. A better result for shallow cracks is obtained with eq. (39) and
eq. (41) which require, respectively, the flow curve and the CMOD.
Widely spread plasticity leads to a loss of constraint. For this reason the plastic zone formation
was studied and leads to recommend a "crack length to bar radius" ratio of about a/R0.5. To better
quantify the loss of constraint the Q-factor is used. It reveals that to obtain negligible loss of
constraint (Q>-0.1), the specimen size requirement is 2 times more severe than the ASTM E399
recommendation eq. (8). The loss of constraint is identified to be the consequence of plastic zone
interaction, which occur at low load level and induce a state between the plane strain and plane
stress condition. This result seems to be in contradiction with the specimen size requirement found
in the literature for the LEFM domain. However, the specimen size requirement established on the
Q-factor is based on a small decrease of stress ahead of the crack tip. The importance of such a
small stress decrease depends on the fracture mechanism and the process zone.
For ferritic steel in the transition region, a micromecanical model is developed to account
simultaneously for size effect and loss of constraint eq. (53). To establish the loss of constraint
correction a reference condition, called small scale yielding (SSY), is established and compared to
the CRB. These corrections are found to be independent of the critical triggering stress as long as
specimen size requirements are satisfied, eq. (54) to (56). The limitation of this two-parameters
fracture mechanics model is due to the extensive deformation for which the self similarity of stress
field is lost [22]. An other limitation of this approach is related to the ductile fracture mechanism
occurring in the upper transition region. Ductile crack growth which is not taken into account in this
approach increase the sampled volume of material subject to a given maximal principal stress. As a
consequence the cleavage fracture probability is increased. Moreover, tensile geometries are
particularly sensible to unstable ductile tearing.
The correction coefficients were established in Figure 12 and 13 for different crack length and
stress hardening coefficient. The loss of constraint is higher for shallow crack and low strain
hardening exponent.
Conclusions
The detailed theoretical study of the CRB in LEFM and EPFM showed that:
138
(1) Benthem and Koiter's stress intensity function, eq. (1), is very accurate and is recommended for
the evaluation of the fracture toughness of CRB in LEFM.
(2) The compliance of a CRB can be evaluated by eq. (3) as long as G>2R and by eq. (7) for
shorter gauge length.
(3) The specimen size requirement reviewed in the literature eq. (8) is 7 times less severe as
compared to ASTM E399.
(4) The J-integral in axisymmetric configuration is different from the one in 2D plane stress or
plane strain. An additional surface term should be taken into account eq. (9).
(5) To evaluate the load point displacement for specimens with a/R>0.4, an extensometer with
gauge length larger than 2R is recommended.
(6) The eq. (32) for the -factor should be chosen to evaluate the J-integral from experimental
traces of CRB for which a/R>0.4.
(7) For shallow cracks a/R<0.4, eq. (39) and eq. (41) are recommended according to the
availability of flow curve or CMOD measurements.
(8) The optimum CRB geometry has a "crack length to bar radius" ratio of a/R0.5.
(9) The specimen size requirement in order to avoid loss of constraint, is twice more severe than
the ASTM E399.
(10)Eq. (53) is recommended to evaluate the fracture toughness of ferritic steel specimens in the
transition region. This equation takes simultaneously size and loss of constraint into account.
(11)Loss of constraint corrections, which significantly enlarge the domain of applicability of the
CRB in the transition region, were established in Figure 12 and 13. The specimen should
however satisfy the size requirement given by eq. (54) to (56). Moreover, ductile crack growth
invalidates the procedure and can induce unstable ductile tearing.
Acknowledgements
The authors are grateful to Yan Ai Min and Prof. Nguyen-Dang Hung for making available their
work on plastic collapse analysis of structures. The present work is partially financed by Tractebel
Energy Engineering. Their support is kindly appreciated.
References
[1]
[2]
[3]
[4]
[5]
[6]
G. Gage, Measurement of transition fracture behaviour using miniature-scale specimens, Advances in

fracture mechanics, TAGSI Symposium, TWI Cambridge, 9-10 January 1996.
J. H. Giovanola, H. Homma, M. Lichtenberger, J.E. Crocker and R.W. Klopp, Fracture toughness
measurements using small cracked round bars, Constraint Effects in Fracture: Theory and Applications,
ASTM STP 1244, Mark Kirk and Ad. Bakker, Eds., American Society for Testing and Materials, Philadelphia,
1994.
J. H. Giovanola, R.W. Klopp, J. E. Crocker, D. J. Alexander, W. W. Corwin and R. K. Nanstad, Using Small
Cracked Round Bars to Measure the Fracture Toughness of a Pressure Vessel Steel Weldment: A feasibility
study, Small Specimen Test Techniques, ASTM STP 1329, W. R. Corwin, S. M. Rosinski and E. van Walle,
Eds, American Society for Testing Materials, 1997.
E. Lucon, Cylindrical specimens for evaluating a serviced component's fracture toughness properties,
Proceedings of the Ninth European Conference on fracture ECF9, Varna (Bulgaria), 21-25 September 1992.
Y. Lei, and B.K. Neale, Non-linear, axi-symmetric finite element analyses of a circumferentially cracked bar
specimen, Int. J. Press. Ves. and Piping, Vol. 73, 199-210, 1997.
Nakano Motohiro and Kishida Keizo "Measurement of dynamic fracture toughness by longitudinal impact of
precracked round bar", Int. J. of Pressure Vessels and Piping, Vol. 44, pp. 3-15, 1990.
139

[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
M. Scibetta, and R. Chaouadi, Fracture toughness derived from small circumferentially-cracked bars, Small
Specimen Test techniques, ASTM STP 1329, W. R. Corwin, S. M. Rosinski, E. van Walle, Eds., American
Society for Testing and Materials, 1997.
A. H. Sherry, B. K. Neale, G. Gage and D. J. Sanderson, The use of circumferentially-cracked bars for the
measurement of fracture toughness, ECF 11 - Mechanisms and mechanics of damage and failure, 2-6
September, 1996.
J. P. Benthem and W. T. Koiter, Asymptotic approximations to crack problems, in Methods of Analysis and
Solutions of Crack Problems, (Edited by G. C. Shi), chap. 3, Noordhoft International Publishing, Groningen,
131-178, 1973.
H. C. Bueckner, Discussion on stress analysis of cracks, by P. C. Paris and G. C. Sih, ASTM STP 381, 8283, 1965.
D. O. Haris, Stress intensity factors for hollow circumferentially notched round bars, J. Basic Engng., 89, pp.
49-54, 1967.
M. Scibetta, Precracking of round notched bars, report BLG-708, SCKCEN Mol, Belgium, February 1996.
Shen Wei, Zhao Tingshi, Gao Daxing, Li Dunkang, Li Poliang and Qui Xiaoyun, Fracture toughness
measurement by cylindrical specimen whith ring-shaped crack, Engng Fracture Mech., Vol. 16, No. 1, 69-82,
1982.
R. N. Ibrahim, and H.L. Stark, Validity requirements for fracture toughness measurements obtained from
small circumferentially notched cylindrical specimens, Engng Fracture Mech., Vol. 28, No. 4, 455-460, 1988.
M. Scibetta, Fracture toughness evaluation of circumferentially-cracked round bars, report BLG-716,
SCKCEN Mol, Belgium, May 1996.
McMeeking R.M. "Finite deformation analysis of crack-tip opening in elastic-plastic materials and
implications for fracture", J. Mech. Phys. Solids, Vol. 25, pp. 357-381, 1977.
O'Dowd N.P. and Shih C.F. "Family of crack-tip fields characterized by a triaxiality parameter -I structure of
fields", J. Mech. Phys. Solids, Vol. 39, No. 8, pp. 989-1015, 1991.
ASTM E399-83 "Standard test method for plane-strain fracture toughness of metallic material"
J. R. Rice, A Path independent integral and approximate analysis of strain concentration by notches and cracks,
J. of applied Mechanics, 379-386, June 1968.
H. Bergkvist, and G. L. Huong, J-integral Related quantities in axisymmetric cases, Int. Journ. of Fracture,
13, 556-558, 1977.
B. Moran, R. J. Asaro and C. F. Shih, Effects of Material rate sensitivity and void nucleation on fracture
initiation in a cicumferentially cracked bar, Metallurgical Transaction A, Vol. 22A, 161-170, January 1991.
T. L. Anderson, Fracture mechanics fundamentals and applications, CRC Press second edition, London, 1680, 1995.
H. A. Ernst, Material Resistance and Instability Beyond J-Controlled Crack Growth, Elastic Plastic Fracture,
ASTM STP 803, Vol. I, ASTM, Philadelphia, 191-213, 1983.
J. R. Rice, P.C. Paris and J. G. Merkle, Some further results of J-integral analysis and estimates, ASTM STP
536, 231-245, 1973.
Y. Z. Itoh, T. Murakami, and H. Kashiwaya, Approximate formulae for estimating the J-integral of a
circumferentially cracked round bar under tension or torsion, Engng Fracture Mech., Vol. 31, No. 6, 967-975,
1988.
F. M. Beremin, Calculation and experiment on axisymmetrically cracked tensile bars: prediction of initiation,
stable crack growth and instability, 6th SMIRT, Paris, 1981.
A. G. Miller, Review of limit loads of structures containing defects, Int. J. Pres. Ves. and Piping, Vol. 32,
197-327, 1988.
M. Scibetta, and A. M. Yan, Experimental and theoretical determination of the -factor for circumferentiallycracked round bars, report BLG-737, SCKCEN Mol, Belgium, May 1997.
A. M. Yan, Contributions to the direct limit state analysis of plastified and cracked structures, doctoral
thesis, University of Liege, Belgium, December 1997.
B. K. Neale, The fracture behaviour of a circumferentially cracked bar in tension, TAGSI Symposium, TWI
Cambridge, 9-10 January 1996.
C. F. Shih, N. P. O'Dowd, and M. T. Kirk, A Framework for Quantifying Crack Tip Constraint, Constraint
Effects in Fracture, ASTM STP 1171, American Society for Testing and Materials, Philadelphia, 2-20, 1993.
K. Wallin, The scatter in KIc-results, Engng Fracture Mech., Vol. 19, No. 6, 1085-1093, 1984.
K. Wallin, Statistical modelling of fracture in the ductile-to-brittle transition region, Defect Assessment in
Components - Fundamentals and Applications, ESIS/EGF9 (Edited by J. G. Blauel and K. -H. Schwalbe),
Mechanical Engineering Publication, London, pp. 415-445, 1991.
T. L. Anderson and R. H. Jr. Dodds, Simple constraint corrections for subsize fracture toughness specimens,
Small Specimen Test Techniques Applied to Nuclear Reactor Vessel Thermal Annealing and Plant Life
Extension, ASTM STP 1204, W.R. Corwin, F. M. Haggag and W. L. Server, Eds., American Society for
Testing and Materials, Philadelphia, 93-105, 1993.
140
Analytic and Numeric Shakedown Solutions of Cylinder

and Sphere Structures under Thermal Loading
*
Yan Ai Min*
Dept. of Aerospace & Mechanical Engineering
Abstract
This paper describes shakedown analysis of structures subjected to varying thermal and mechanical
loads. Analytic formulation of cylindrical and spherical structures was developed to compare with
numerical calculations. The obtained results were presented in BREE figures that define safe design
loading domains of the structures and provide useful ideas for engineers. Excellent agreement
between analytic and numerical solutions verified both theoretical and numerical approaches.
Keywords: shakedown analysis, thermal loading, cylinder & sphere structures.
Introduction
Thermal loading is known to have great effect on shakedown or inadaptation behaviour of

structures. Large amounts of studies have been carried out since the first investigation initiated by
Prager in 1956. An extensive treatise on this subject was published by Gokhfeld & Cherniavsky [1].
Numerical studies of particular problems were addressed in many papers, for example [2-4].
Recently, an efficient kinematical algorithm was developed to study structures under thermal and
mechanical loading [5-6]. Mises criterion with temperature-dependent yield stress was also adopted
to approach better practical engineering material. On the other hand, few analytic solutions of some
simple structures were published in the literature [1, 7-9], which are very useful to verify numerical
algorithms and to provide directly design formulae of structures.
In this paper, we present an application of both linear and nonlinear kinematical approaches to
shakedown analysis of structures subjected to combined thermal and mechanical loading. Analytic
formulation of some cylindrical and spherical structures was developed to illustrate theoretical
methods and to compare with numerical solutions. The obtained results are presented in BREE
figures that define clearly safe design loading domains of the structures studied.
Theoretical consideration
Thermal stresses appear when the expansion or contraction of structural components due to
temperature variation T is constrained. For instance, when a rise in temperature in a homogeneous
body is not uniform, different elements of the body tend to expand by different amounts, which
conflicts the requirement that the body remain continuous. In this case, the thermal strain must be
accompanied by the thermal stress. In numerical calculation, thermal strain is usually considered as
a kind of initial strains that are proportional to the temperature change T, such as:
0 = T T
(1)
141
where T a linear thermal expansion coefficient vector. In FEM, generalised node thermal load due
to the thermal strain may be defined as:
g eT = B T D 0 dV = B T D T TdV
Ve
Ve
(2)
Total node load vector g can be obtained by supposition of all kind of external loads to find
structural strain . Therefore, fictitious elastic stress field including the thermal stress is obtainable
by the following relation
e = D( T T )
(3)
where the strain matrix B and the Hooke's matrix D should be defined according to the adopted
finite elements.
It should be pointed out that owing to the self-equilibrium property of thermal stresses, thermal
load alone cannot lead to instantaneous collapse failure. This self-equilibrated system of stresses
has no influence on the plastic collapse limit of other loading, provided it does not significantly
alter the geometry or the yield surface. However if the thermal stress field varies with time, it leads
possibly to incremental plasticity (ratchetting) and/or alternating plasticity (plastic fatigue). These
two inadaptation failure modes may be determined by the so-called shakedown analysis. Note that
shakedown represents generally a safe state of structures when plastic deformation ceases
developing and structures recovery to complete elastic state. In this work, a kinematical approach
will be applied, which is based on Koiter theorem [10] as described below:
Shakedown occurs if and only if, for any potentially active loading path, the following inequality is
satisfied for any admissible cycle of plastic strain.
dt
e
ij
& ijp d V d t D (& ijp , T ) d V
(4)
where ije is fictitious elastic response by (3), including thermal stress due to thermal loading. D
represents here plastic dissipation. In the following section, we will take some simple and typical
structures as example showing how to apply this theorem into analytical and numerical calculations.
3
Analytic solutions of cylinder and sphere under thermal loading
and internal pressure
We will discuss in this section analytic solutions of cylinder and sphere structures subjected to
internal pressure, and thermal loads along the thickness. These solutions will be useful in practical
engineering and for comparison with the numerical results. Some previous related work was
reported by Bree [8], Gokhfeld & Cherniavsky [1], Knig [10-11] , Morelle [4] among the others.
3.1
Cylinder subjected to thermal loading and internal pressure
A long cylinder is subjected to varying internal pressure p and to repeated temperature difference T
along wall thickness direction showed in Fig. 1. The loading domain may be represented as:
142
z
D = [ p (0, p0 ),T (0, T )]
T = T0
ln(b / r )
ln k
(5)
(6)
where k=b/a is ratio of external and internal radius.
Fig. 1: Distribution of temperature

through the thickness of cylinder
Elastic stress field due to internal pressure is well known as:
rp =
p
b2
p
b2
p
(
1
)
,
=
(
1
+
),
k 2 1
r2
k 2 1
r2
zp = ( rp + p ) = 2
p
k 1
2
(7)
where superscript p means internal pressure; is a coefficient: =0 for plane stress condition,
= (Poisson ratio) for plane strain condition, =1/2 for an end-closed cylinder. The thermal
elastic stress field due to temperature difference along thickness is (for a long pipe but assuming the
axial displacement is not prevented):
1
1
b2
ln b / r
b2
1
ln b / r
T
Tr = q 2
+
+
q
(
1
)
(1 2 ) +
,
2
2
r
ln k
ln k
r
ln k
k 1
k 1
1
2 ln b / r
2
Tz = q 2
+
= Tr + T
1
ln
ln
k
k
k
(8)
T ET0
; superscript T means thermal load; E Young's module. The total resultant
2(1 )
stresses may be obtained by the supposition of ( 8) and ( 9) such as
where q =
ij = ijp + Tij .
(9)
3.1.1 Solution with Tresca criterion
The principal stresses accompanying the plastic behaviour of the cylinder are assumed to satisfy the
inequalities: > z > r . Tresca's criterion, cf. Fig. 2, may be written in a matrix form:
N h
with
1 0 1 0 1 1
NT =
,
0 1 0 1 1 1
(10)
T = r ,
143
hT = y y y y y y
Fig.2:. Tresca's yield surfaces

If the stress locus over a loading cycle passes over all yielding surfaces, one may write plastic strain
rate by using normal law of plasticity:
& r = & 1 & 3 + & 5 & 6 , & p = & 2 & 4 & 5 + & 6
(11)
where & l is plastic flow rate at lth yield surface. In this linear model, the kinematical theorem (4)
may be expressed as [13]:
al l dV hl l dV
V
(12)
where al = max N lij ij (t ) , l = & l dt and hl is related to lth piece of yield surface. Eq. (12) may
t
be used to find two plastic failure limits as below.

i) Incremental plasticity limit
Incremental plastic deformation represents a uniform radial extension of cylinder. One could
suppose that only yield surface 6 is satisfied such as r = y . According to Knig [11], the
following velocity and strain rate field may be adopted as the mechanism of incremental plasticity,
which satisfies the compatibility and incompressibility conditions of plastic strain rate:
c
c
c
u& = , & = & 6 = 2 , & r = & 6 = 2
r
r
r
(13)
where c is a factor. By (7-9), one has
144
(14)
(r ) r (r ) = f 2 (r ) p + q
f 2 (r )
ln k
2b 2
where f 2 (r ) = 2
. Taking the term in the parentheses of above equation (14) as zero, we
(k 1)r 2
define a separating radius
r02 =
2b 2 ln k
(k 2 1)
(15)
Now one can calculate the envelope of elastic stresses:

For r < ro
a16 (r ) = f 2 (r ) p
(16)
For r > ro
a62 (r ) = f 2 (r ) p0 + q
f 2 (r )
ln k
(17)
Although internal pressure varies as (0,1)p, its maximum value p is always taken for the
computation of incremental plasticity limit. This fact explains that the incremental plasticity limit
will be the same as instantaneous plastic limit when only internal pressure exists. Apply now linear
upper bound formula (12):
Left integration of (12):
r0
b
1 r02 b 2 ln(b / r0 )
a T d V = a61 & 6 d V + a62 & 6 d V = c p0 + q 2
+
(18)
2
a
r0
ln k
k 1 r0
Right integration of (12):
hT d V = c y ln k
(19)
Therefore, the linear upper bound theorem leads directly to the incremental plasticity limit equation:
1
b 2 ln(b / r0 )
pi + qi 2 (1 2 ) +
= y ln k
ln k
r0
k 1
with
pi = y ln k , Ti =
or
p0 T0
+ =1
pi Ti
2(1 ) y ln k
ln(b / r0 )
1
b2
+ 2 (1 2 )
aT E
k 1
r0
ln k
(20a)
(20b)
The solution (20) is exact for Tresca material since at all stages of a loading cycle it is associated
with a statically admissible stress field.
ii) Alternating plastic limit
Assume that the plastic fatigue occurs locally at a location of radius rf (a rf b) and yield surfaces
6 and 5, shown in Fig.2, are satisfied during loading cycle. However according to the kinematical
theorem, the determination of rf must lead to a minimum plastic fatigue limit. The mechanism of
collapse is represented in the following form:
& 5 = & 6 = c(r f )
(21)
145
where (rf) is Dirac's distribution and c is a factor. The envelope of elastic stress in (12) during
loading process is same as that defined by (16-17), while during unloading process, the envelope of
elastic stress is calculated as
For r<r0
f 2 (r )
a51 (r ) = q
ln k
(22)
For r>r0
a52 (r ) = 0
(23)
Left integration of (12):

r0
r0
max aT d V = a51& 5 + a61 & 6 d V + a62 & 6 d V

rf
Right integration of (12):
hT d V = 2C y rf
(24)
(25)
It is easy to verify that minimum solution of plastic fatigue limit requires rf =a, leading to
alternating plasticity (plastic fatigue) limit:
k 2 1
1
p f + q f 1 2
= y (1 2 )
k
2k ln k
with
p f = y (1
1
) , Tf =
k2
or
p0 T0
+
=1
p f Tf
2(1 ) y
1
k2
T E ( 2
)
k 1 2 ln k
(26a)
(26b)
We note that the internal pressure is considered varying in the above solution. If the internal
pressure is constant, only solution Tf remains. That is to say, the constant internal pressure does not
have influence on plastic fatigue limit of thermal loading. The above thermal loading limits may be
organized in following simplified form:
8(1 ) y
(27)
, Ci1 when k1
Incremental plasticity limit: Ti = Ci
T E
4(1 ) y
Tf = C f
(28)
Plastic fatigue limit:
, Cf1 when k1
T E
Ci and Cf are defined in Fig. 3. We see that the shakedown limit always coincides with the plastic
fatigue limit in the case of unique thermal loading and it is about two times smaller than the
incremental plasticity limit for a very thin-walled tube.
3.1.2 Solution with von Mises criterion

i) Approximate ratchetting limit
It is difficult to obtain an exact incremental plastic limit with non-linear Mises's criterion. We
proposed in [13] an approximate solution of incremental plastic limit, which was proved as a
reasonable lower bound.
p0 T0
+ =1
pi Ti
(29a)
146
2.4
2.2
2
1.8
Ci =
1.6
1.4
ln k
ln(b / r0 )
1
b2
+ 2 (1 2 )
4
k 1
r0
ln k
C 1.2
1
0.8
0.6
Cf =
0.4
2(
0.2
1
k
)
k 1 2 ln k
2
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1/k= a/b
Fig. 3: Coefficient of shakedown limit of tube under

through-thickness thermal loading
with
pi =
2
3
y ln k
Ti =
2(1 ) y ln k
ln(b / r0 )
1
b2
+ 2 (1 2 )
aT E
k 1
r0
ln k
(29b)
ii) Exact plastic fatigue limit

Instead of using kinematical formulation, we consider directly the fact that plastic fatigue occurs
when variation of effective stress anywhere passes over double yielding limit of material [6, 10]:
max eq ( p, T , r ) = 2 y
r
(30)
where eq is elastic Mises-equivalent stress variation during a loading cycle. From above analysis
we have known that the plastic fatigue occurs at the inner wall of cylinder. Using (7-9) with r=a, we
can give the following solution by a long but not very difficult mathematical deduction:
qf
y
3 pk 2 p + 2 p + (16 3 p 2 12 p 2 2 + 12 p 2 3 p 2 k 4 6 p 2 k 2 + 12 p 2 k 2 )1/ 2
1
k2
4( 2
)
k 1 2 ln k
(31a)
where q f =
T ET f
2(1 )
and p =
pf
y (k 2 1)
147
3 pk 2 + 16 3 p 2 k 4
=
For =1/2 (end-closed pipe):
k2
1
y
4( 2
)
k 1 2 ln k
qf
For 1 / 3 (plane strain):
qf
y
9 pk 2 p + 3 48 9 p 2 k 4 6 p 2 k 2 p 2
k2
1
12( 2
)
k 1 2 ln k
If p=0 or p constant, we obtain from (31) the plastic fatigue limit of thermal loading as:
2(1 ) y
Tf =
1
k2
T E ( 2
)
k 1 2 ln k
(31b)
(31c)
(32)
It is interesting to notice that it is the same as (26b) with Tresca's criterion.

Without thermal loading, namely qf =0, we have the plastic fatigue limit of pressure loading from
(31):
2 y (k 2 1)
pf =
(33a)
4 2 4 + 3k 4 + 1
For =1/2 (end-closed pipe)
pf =
For 1/3 (plane strain)
pf =
2 y
3
(1
1
)
k2
(33b)
2 y (k 2 1)
(33c)
1/ 9 + 3k 4
The difference between (33b) and (33c) is very small, the maximum of which is 3.7% as k1. The
combination of (29) and (31), or (32) if p is constant, constructs a shakedown loading domain with
Mises' criterion.
3.2
Sphere subjected to thermal loading and internal pressure
Owing to the central symmetry, two principal components of stresses (in the two meridional planes)
are equal so that Mises' and Tresca's yield conditions coincide. The temperature is usually supposed
to be distributed across the thickness according to the function:
T = T0
b / r 1
,
k 1
T (0,1)T ,
k=b/a
(34)
which has a similar form as shown in Fig. 1. The elastic stress distribution generated by the internal
pressure p and the temperature field (34) can be expressed by elastic analysis:
p
b3
p
b3
p
p
p
r = 3
(1 3 ) ,
= = 3
(1 + 3 )
(35)
k 1
r
k 1
2r
Tr =
3b
b
3b b
q
q
k (k + 1) + ( ) 3 , T = T =
k (k + 1) ( ) 3
r
r
2r 2 r
k 1
k 1
(36)
148
where q =
3
T ET0
. Then the total elastic stress field is
and = 2
3(1 )
k + k +1
ij = ijp + Tij
(37)
i) Incremental plastic limit of sphere:

We propose, as the mechanism of incremental plasticity, the following velocity and strain rate
fields:
u& =
c
c
c
1
, & r = & 6 = 3 , & = & = & 6 = 3
2
2
r
2r
2r
(38)
It is easy to verify that this strain rate field satisfies conditions of both compatibility and
incompressibility. The proposed plastic mechanism has also been verified by our numerical
calculations shown in Fig. 4.
10
9
8
7
6
u 5
4
Numerical results
Proposed mechanism
2
1
0
0
10
Fig. 4: Incremental mechanism of sphere under internal pressure

By using this strain field and repeating a similar procedure as in 3.1.1, we could obtain finally
incremental plastic limit condition as below, which is the same as that obtained by Gokhfeld &
Cherniavsky [1] using a so-called "overload method".
with
pi = 2 y ln k ,
p0 T0
+ =1
pi Ti
6(1 ) y
(k 1)(k 2 + k + 1) ln k
Ti =
2 2
T E
(k + k + 1)3/ 2 3(k 2 + k )
3
(39a)
(39b)
ii) Plastic fatigue limit for arbitrarily varying pressure and thermal field
Applying the method that has been used in the cylinder analysis to the present sphere problem, we
obtain plastic fatigue limit equation:
149
p0 T0
+
=1
p f Tf
where
(40a)
4 y
4(1 ) y k 2 + k + 1
1
pf =
(40b)
(1 3 ) , T f =
k
aT E
k + 2k 2
3
The shakedown limit of thermal loading in (39b) and (40b) may be written in the following
forms:
2 .6
2 .4
2 .2
2
Ci =
1 .8
1 .6
(k 1)(k 2 + k + 1) ln k
8
3 3
(k 2 + k + 1)3/ 2 4(k 2 + k )
1 .4
C
1 .2
1
0 .8
0 .6
k 2 + k +1
Cf =
2k 2 + k
0 .4
0 .2
0
0
0 .1
0 .2
0 .3
0 .4
0 .5
0 .6
0 .7
0 .8
0 .9
1 /k= a/b
Fig. 5: Coefficient of shakedown limit of sphere under

through-thickness thermal loading
8(1 ) y
Incremental plastic limit: Ti = Ci
, Ci 1 when k1
t E
4(1 ) y
Plastic fatigue limit:
Tf = C f
, Cf1 when k1
t E
(41)
(42)
Ci and Cf are defined in Fig. 5. We see that under unique thermal loading the shakedown limit
always be dominated by the plastic fatigue limit, and it is about two times smaller than the
incremental limit for a very thin-walled sphere. By comparing (27-28) with (41-42), it is interesting
to note that the results of very thin-walled sphere are the same as that of a thin-walled cylinder. The
combination of (39) and (40), constructs a shakedown domain of a sphere subjected to radial
thermal loading and internal pressure.
Numerical verifications
A mathematical programming algorithm ELSA, developed in [6], was used to compare with the
presented analytic solutions. The examples demonstrate also some typical shakedown properties of
structures under combined mechanical and thermal loading.
150
4.1 A long thin-walled cylinder under radial thermal loading and internal
pressure
Internal radius a = 490 mm
external radius b = 500 mm
T : varying (0,1)T
p : varying (0,1)p or (1,1)p
E = 151000 Mpa, = 0.28
T = 1.5 10-5
y = 116.2 MPa
Fig. 6: A long thin-walled cylinder under pressure and thermal loading

This model was proposed in [14] from a nuclear device. The plane strain condition was considered.
However, the free axial extension due to temperature variation was allowed. Axisymmetric finite
elements were used and Mises's criterion was adopted. Internal pressure was supposed either
arbitrarily varying or remaining constant. Thermal loading repeated as temperature difference
(0,1)T along the wall thickness. The results of shakedown analysis are represented in BREE figure
7, where Tf and pf are alternating plastic limit of, respectively, temperature difference and internal
pressure, defined by (32-33c).
/f
2
1.8
Analytic solution
3
1.6
Numerical results of ELSA

1.4
Result proposed by NNC
plastic
fatigue
1.2
2
F
combined plasticity
C
0.8
Po
1
B
0.6
shakedown
E
0.4
ratchetting
A
0.2
elastic behaviour
R
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
p/pf
Fig. 7: BREE figure for a long cylinder under radial thermal loading and
internal pressure Tf: (32), pf (33c)
151
Line 1 represents the plastic fatigue limit (alternating plasticity) for arbitrarily varying radial
thermal loading and internal pressure (31). Line 2 is the plastic fatigue limit for varying thermal
loading and constant internal pressure (32). Line 3 represents the approximate incremental plasticity
limit (29). Line 4 shows then the elastic limit of the cylinder. Therefore by these lines, Fig. 7 may
be sub-divided into a number of sub-regions, which correspond to different states of the structure;
these include the following:
(a) Completely elastic behaviour regions A is assured by line 4; (b) Shakedown domain is in
region B if internal pressure varies arbitrarily, or (B+C) if the pressure is constant. In this domain,
the structure that plastified in first loading cycle will shake down elasticity after a finite number of
loading cycles; (c) Plastic collapse may occur due to mechanical loading alone. This point is
marked R calculated by (29b); (d) Plastic fatigue occurs due to thermal loading alone, marked S,
obtained by (32); (e) Mode of plastic fatigue D (or D+C) occurs for relatively low pressure and high
thermal loading; (f) Mode of ratchetting E which will lead to radial extension of cylinder due to the
repeated pressure and thermal loads.
Beyond these regions, the structure will fail in a possibly mixed mode. It is seen that the lower
bound solution P0 proposed in [14] has a safety factor of about 1.23 with respect to the present
solution of plastic fatigue limit. The obtained numerical and analytic solutions are in good
agreement.
4.2
A thick-walled sphere under radial thermal loading and internal pressure
Consider now a thick-walled sphere subjected to radial thermal loading and internal pressure. The
geometry, material and the finite element meshes are shown in Fig. 8. Owing to the central
symmetry, one fourth of the sphere was modelled by axisymmetric finite elements.
The radial thermal load varies between (0,1)T with two cases of internal pressure: constant
(1,1)p or arbitrarily varying (0,1)p. The results of shakedown analysis are represented in Fig. 9.
We see that the numerical results are in excellent agreement with analytic solutions. As the last
example of cylinder, Fig. 9 may be subdivided into a number of sub-regions, which correspond to
different modes of deformation:
(a) Completely elastic behaviour regions A. (b) Shakedown domain happens in region B if
internal pressure varies arbitrarily or in region (B+C) if the pressure is constant; (c) Plastic fatigue
occurs in domain D (or D+C if p varies); (d) Plastic collapse occurs due to mechanical loading
alone, marked R, by (40b).
Beyond these regions, the structure will fail in a possibly mixed mode (plastic fatigue and
ratchetting). Comparing the previous thin-walled cylinder with the present thick-walled sphere, we
notice that alternating plasticity (plastic fatigue) is more importance for a thick-walled structure
while ratchetting (instantaneous plasticity) may be dominant for a thin-walled structure.
Conclusions
Theoretical analysis and numerical computation were performed in parallel for some cylinder and
sphere structures subjected to combined thermal and mechanical loading. It was shown that starting
from the classical Koiter's shakedown theory, the upper bound or exact analytic solutions may be
derived for some simple structures. An excellent agreement between analytic and numerical
calculations verified precision and reliability of both analytic and numerical methods. While a full
152
inelastic analysis requires a continuation of calculation over a large number of cycles, a fairly
complete description of the ratchetting and plastic fatigue of the structure may be easily obtained by
the present methods.
a = 0.4m
b = 1.0m
p: varying (0,1)p
or constant
T: varying (0,1)T
according to (34)
E = 210000 MPa
= 0.3
T = 1x10-5 1/C
a
b
Fig. 8: Finite element meshes for a thick walled sphere

It has been demonstrated that the plastic failure mode of structures depends on the loading
condition and geometry. For the first example of a thin-walled shell, the collapse failure occurs
often due to ratchetting (instantaneous plasticity). However, the possible plastic fatigue should be
borne in mind for a thick-walled structure especially when thermal loading exists. The obtained
results in BREE figures indicate clearly failure mode in different loading regions.
/ f
1.5
Analytic solution
1.4
Numerical solution
1.3
Plastic fatigue
1.2
E
Combined plasticity
1.1
1
0.9
0.8
Fatigue if p varies
0.7
Shakedown if p is constant
0.6
0.5
Ratchetting
0.4
Shakedown
0.3
0.2
Elasticity
0.1
p/p
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.1
1.2
1.3
1.4
1.5
Fig. 9: BREE-type shakedown diagram for sphere under radial thermal loading
and internal pressure. Tf , pf: by (40b)
153
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
Gokhfeld DA & Cherniavsky OF., Limit analysis of structures at thermal cycling, Sijthoff & Noordhoff, The
Netherlands,1980.
Weichert D, Gross-Weege J., The numerical assessment of elastic-plastic sheets under variable mechanical and
thermal loads using a simplified two-surface yield condition, Int. J. Mech. Sci.; 30: 757-767, 1988.
Karadeniz S, Ponter ARS., A linear programming upper bound approach to the shakedown limit of thin shells
subjected to variable thermal loading, J. Strain Analysis; 19(4): 221-230, 1984.
Morelle P., Numerical shakedown analysis of axisymmetric sandwich shells: an upper bound formulation, Int.
J. Num. Meth. Engng.; 23: 2071-2088, 1986.
Maier G, Pan LG, Perego U., Geometric effects on shakedown and ratchetting of axisymmetric cylindrical
shells subjected to variable thermal loading, Engng. Struct.; 15: 453-465, 1993.
Yan A.M., Nguyen-Dang Hung, Kinematical shakedown analysis with temperature dependent yield stress, Int.
J. Num. Meth. Engng,; 50: 1145-1168, 2001.
Vu-Duc Khoi., Yan A.M., Nguyen-Dang Hung, A dual form for discretized kinematic formulation in
shakedown analysis, Int. J. Solids and Structure; 41(1): 267-277., 2004
Bree J., Elastic-plastic behaviour of thin tubes subjected to internal pressure and intermittent high-heat fluxes
with application to fast-nuclear-reactor fuel elements, J. Strain, 1967.
Burgreen D., Structural growth induced by thermal cycling, J. Basic Engng, 90: 469-475, 1968.
Knig JA., Shakedown of elastic-plastic structure, Elsevier & PWN-Polish Scientific Publishers, 1987.
Knig JA., On the incremental collapse criterion accounting for temperature dependence of yield point stress,
Archives of Mechanics; 31: 317-325, 1979.
Koiter WT., General theorems for elastic plastic solids, In Progress in Solid Mechanics, Sneddon IN, Hill R
(eds). Nord-Holland, Amsterdam, 165-221, 1960.
Yan AM., Contributions to the direct limit state analysis of plastified and cracked structures, PhD Thesis
(1997), Collection of publication of University of Liege, No. 190, Belgium, 1999.
Sinclair E. NNC contribution to CEC WGCS AG2 contract (ETNU-0110B) on Numerical techniques for
shakedown, Technique Report C9708/TR/001, NNC Company, UK, 1996.
154
EMMC & MCMC Students Papers
Metis Element Method for Delamination in Composite Laminate

Nguyen-Tien Duong* and Nguyen-Dang Hung+
*Hanoi University of Technology, Vietnam
+LTAS, Division of Fracture Mechanics, University of Lige, Belgium
Abstract
The "metis" finite element method is originally developed by Nguyen Dang Hung [1]. This method
overcomes the lack of no-monotonic convergence provided by general hybrid displacement and
stress models. The monotonic convergence of the "metis" element models is investigated in [2]. The
unisolvent condition of the external displacement field permits to transform the integral of contour
into integral of volume. This is not the case for classical hybrid element.
The present paper uses this method to establish the model for delamination in composite
laminate. The anisotropic elasticity theory and solution of Lekhnitskii for generalized plane strain
state are presented. Wangs singular solution is used to determine the stress singularity order and
eigen coefficients. Then the formulae for quasi three-dimensional metis elements are established.
The stress intensity factors and the energy release rates for composite laminates under extension are
determinated and compared with literature [3, 4].
Keywords: metis, delamination, stress intensity factor, energy release rate.
Introduction
Graphite-epoxy composite materials have high strength, high stiffness and low weight. So they have
been reliably used in many applications from sport articles to aerospace structures. A mode of
degradation generally observed in laminated composite structures is delamination between the
composite layers. The delamination is one of the most frequently encountered types of fracture in
composite laminates due to high stress concentration at free edges. Due to its complex nature, not
only the characterization of the interface properties is really difficult but also their modelling. It
involves not only geometric and material discontinuities but also inherently mixed fracture modes I,
II and III.
Classical finite element methods converge very slowly towards the solution because of the bad
representation of the singularity in the vicinity of the crack tip. Nguyen Dang Hung [1] was
proposed a special class of hybrid finite element named metis element. The metis displacement
element is special case of the original hybrid stress element, but the boundary of displacement field
of this is unisolvent. This means that the displacement field is defined completely and continuous
perfectly both within the domain, and on the boundary in such that it is conform in the connection
between elements. The unisolvent condition of the external displacement field of this type of
element permits to transform the contour integral respectively to the volume integral for 3D
problems and the surface integral for 2D problems. By its advantages, this element was used in
many studies [5-12]. In this paper, we use this element to solve the edge delamination problem in
composite laminate structures.
156
Solution for generalized plane strain pr
Consider a multilayer composite laminate (figure 1) subjected to a uniform axial extension along
the z axis z = e(x,y). The composite laminate is assumed to be sufficiently long so that in the region
far from the ends, effects of load and of boundary condition are negligible by virtue of Saint
Venants principle. Consequently, stresses in the laminate are independent of the z axis. This
particular state of stress and strain is named generalized plane strain state:
i = i (x,y); i = i (x,y); where i = 1, 2, ...6.
y
(1)
y
x
x
Fig. 1. Multilayer composite laminate
Lekhnitskii [13] suggests that the normal strain distribution in the z direction is known and
imposed by the formula:
(2)
3=z=e(x,y)
Replace the equation (2) into the law of Hooke: i = Sij j (i,j = 1,2,...6), z has the form:
z = ( e S3 ) / S33 (, = 1, 2, 4, 5, 6).
(3)
Eliminate z from the law of Hooke, it leads to:

d
d
= S
+ % where S
= S S 3 S3 / S33 and % = eS 3 / S33
(4)
d
(, = 1, 2, 4, 5, 6) is the reduced compliance tensor of Sij (i,j = 1,2,...6).
Where: S
The integration of the compatibility equations gives a reduced form of the displacement field
after eliminate the rigid modes:
u 3D (x, y, z) = X 3 z 2 /2 + K 5 y z + u(x,y)
3D
(a)
v (x, y, z) = - K 3 z /2 - K 5 x z + v(x,y)
(b)
w (x, y, z) = (- X 3 x + K 3 y + )z + w(x,y)
(c)
3D
o
3
(5)
Where: 3o is the mean strain; K3 is the bending curvature in the weak inertia plan Oyz; X3 is the
same quantity in the strong inertia plan Oxz; K5 is the warping angle about z-axis; the unknown
functions u (x,y), v(x,y) and w(x,y) only depend on x and y.
The problem is reduced to a 2.5D (D is the dimension) problem:
- The geometry is a 2D domain
157
- The displacement and stress fields are the solutions of 3D problem with external loading 3o ,
K3, X3, K5 and free traction on the boundary .
From the limited conditions, it conduits to solve the problem in limit:
(a)
(b)
Find {u, , } such that:
(c )
(d )
= Sd + o
= def u in
equ = 0
t ( ) = 0
(6)
on
Where: def is the partial derivative operator of strain; equ is the partial derivative operator of
equilibrium; o is the vector of the imposed initial strains; is the section of the structure; is
the boundary of the section .
T = [1 , 2 , 4 , 5 , 6 ] ; T = [ 1 , 2 , 4 , 5 , 6 ] ;
1o = %1 , 2o = %2 , 4o = %4 + K 5 x, 5o = %5 K 5 y, 6o = %6
t() is the tension on the boundary :
t() = in
(7)
The equilibrium plane stress field (x, y, xy) is derived from stress function of Airy F(x,y), the
anti-plane stress field (xz, yz) from function of Prandtl (x,y):
1 =
2F
2F
2F
; 2 =
; 6 =
; 4 =
; 5 =
xy
x
y
y 2
x 2
(8)
The compatibility equations of Beltrami - Michell are reduced to the form [13]:
L4F + L3 = 0 ; L3F + L2 = 0
(9)
Where: L2, L3 and L4 are the partial derivative operators in order 2, 3 and 4, respectively.
The general solution of the equation system (9) was found by Lekhnitskii [13]:
3
F = [Fm ( zm ) + Fm +3 ( zm )]
m =1
(a);
= [m Fm' ( zm ) + m Fm' +3 ( zm )]
(b)
(10)
m =1
The expression of m was presented in [8].

S.S. Wang [3] proposed a particular form of F for the delamination problem:
Cm zm + 2
Fm ( zm ) =
( + 2)( + 1)
(11)
158
Where: is singularity order. The quantities Cm are arbitrary constants to be determined. The
stress field becomes:
3
= [Cmrm zm + Cm +3 rm zm ] where rmT = m2 , 1 , - m , -m , mm
(12)
m =1
Where: m is roots of the characteristic equation of compatibility equations of Beltrami Michell.

From the law of Hooke and the compatible equations, the displacement field is deduced:
3
u = [Cm Tm zm +1 + Cm +3Tm zm +1 ] /( + 1) where TmT = [ pm , qm , tm ]
(13)
m =1
pm = S11d m2 + S12d S14d m + S15d m m S16d m

with qm = S21d m + S 22d / m S 24d m / m + S 25d m S 26d
(a)
(b)
tm = S41d m + S42d / m S 44d m / m + S 45d m S46d
(c)
There exists 6 continuity conditions of stress and displacement vectors along the interface of
two adjacent plies k and k+1 (figure 2) and 6 other free stress conditions on the edge of the crack to
determine 12 unknowns Cm. Solving that equation system leads to eigen value problem. The detail
of determination of the eigen coefficients Cm and the singular stress order are presented in [14].
y
=
Ply k
= - Ply k+1
z
Fig. 2. Edge delamination
between kth and (k+1)th plies
Metis displacement method for generalized plane strain problem
The interface equilibrium leads to the conditions of the continuity of the tractions:
t ( ) = t + ( ) t ( ) = 0 on Si
(14)
In the technique of the hybrid elements of displacement, these conditions as well as the
equilibrium condition of boundary (formula 6.d) are released while introducing on the edge and the
interfaces Si a field of Lagrange multipliers u. The Lagrangian of the problem is as follow:
( , u ) = e ( , u ) = W ( )d uT t ( )dS uT t ( ) dS
e
Si
(15a)
Where the potential is:

1
W ( ) = T S d + T o
2
(15b)
159
Introduce (14) into (15a), it becomes:

e ( , u ) = W ( )d
e
(16)
uT t ( )dS
The unisolvent principle permits to transform the contour integral in 2 dimensions into the
surface integral:
uT t ( )dS =
( def u + u equ )d
T
Taking account of the internal equilibrium in each element (6c), the equation (16) becomes:
1
e ( , u ) = T S d + T o T def u d
e 2
(17)
All elements of the structure (E) are divided into the elements around the crack tip named
singular elements (S) and the other elements named regular elements (R):
E = R U S; R I S = .
In each regular finite element, introducing a discretized displacement field u = N(x) q and the
discretized stress field = R R ( x ) h e1 .
Where: q is a nodal displacement vector; N(x) is a shape function matrix; he1 is the stress
parameter vector unknown of the element e and R(x) is a choice matrix in such a manner that the
internal equilibrium is respected: equ R R (x) = 0 .
Let B(x) being the derivative
that: = def N (x) q = B(x) q .
matrix
of
the
shape
function
matrix
such
Introducing these two discretized fields into (17), it becomes:
1
e ( h e1 , q ) = hTe1Fe1h e1 + hTe1 v oe1 hTe1G Te1q
2
Where: Fe1 = R TR S d R R d ; v oe1 = R TR o d ; G e1 = BT R R d
e
(18)
(19)
The variational equation (18) with respect to he1 gives:
Fe1h e1 + v oe1 G Te1q = 0
(20)
The variational equation (18) with respect to q gives:

e ( h e1 , q )
q
= G e1h e1
(21)
Let: g e = G e1h e1 called the generalized force of each element.
160
Replace he1 from (20) into the formula of ge in (21), it has the form:
g e = K eq g eo e R
(22)
Where: K e = G e1Fe11G Te1 ; g oe = G e1Fe11 v eo1
In each regular finite element, introducing a discretized displacement field u = N(x) q and the
discretized stress field = R + S = RR(x) he1+ RS(x) he2.
Where: he1 represents the regular stresses; he2 the singular stresses; RS(x) is the elastic
asymptotic solution of anisotropic fracture mechanics.
Numerical tests have pointed that the better results are obtained when we identify he2 as common
nodal variables h2 on the common crack tip node of finite elements. Introducing these two fields
into (17), it takes the form:
1
e (h e1 , h 2 , q) = hTe1Fe1h e1 + hTe1 v oe1 hTe1G Te1q + hTe1Fe12h 2 +
2
1
+ hT2 Fe 2h 2 + hT2 v oe 2 hT2 G Te 2q
2
Where: Fe1 = R TR S d R R d ;
e
(23)
v oe1 = R TR o d ; G e1 = BT R R d ;
e
Fe2 = R S R S d ; Fe12 = R S R S d ; v = R TS o d ;
e
T
S
T
R
o
e2
Ge2 = B R S d
T
Take the variation of discrete functional (23) with respect to he1 one obtains:
Fe1h e1 + Fe12h 2 + v oe1 G Te1q = 0
e S
(24)
The variational equation (17) with respect to q restores the equilibrium equation of the structure
and taking account of (18) and (23):
g + g
eR
Where:
eS
g e = G e1h e1
e R
(a)
g e = G e1h e1 +G e 2h 2
e S
(b)
=0
(25)
(26)
Replacing he1 from (24) into (26b), it becomes:

g e = K eq + G *e 2h 2 g oe
e S
(27)
Where: G *e 2 = G e 2 G e1Fe11Fe12
A new equation is created from equations (22) and (27):
Kq + G *2h 2 = g o
(28)
161
Where: K = K e ;
eE
g o = g oe ; G *2 = G *e 2
eE
eS
The variational equation (23) with respect to h2 gives:

(29)
G *2T q F2*h 2 = v o2*

Where: F2* = FeT12 Fe11Fe12 Fe 2 ;
v o2* = FeT12 Fe11 v oe1 v oe 2
eS
eS
The equation systems (28) and (29) can be written in the canonical form:
% % = g%
Kq
% = K
Where: K
*T
G 2
(30)
go
G *2
T
T
T
%
%
=
=
;
q
q
,
h
;
g
o*
2
F2*
v2
Solving of the equation system (30) will provide unknown displacements q at the nodes and the
singular stress parameters h2. Stress intensity factors can be obtained with h2 by the formula [11]:
3
m =1
m =1
K i = 2 {he1[Re( Cm ,1ri ,m ) + Im( Cm +3,1ri , m )] +

3
m =1
m =1
+ he2 Re( Cm,2 ri , m + Cm +3,2 ri ,m ) + he3 Im( Cm,3 ri ,m + Cm +3,3 ri , m )}
(31)
With: i = 2, 4, 6 (K2 = KI, K4 = KIII, K6 = KII) and r6,m in (31) is r3,m in formula (12).
The components of strain energy release rate were calculated by using force and displacements
near the tip with Irwins virtual crack closure technique as [15]:
GI =
1
Fyi t11 ( vm vm' ) + t12 ( vk vk' ) + Fyj t21 ( vm vm' ) + t22 ( vk vk' )
(32)
Where Fyi is the force in the y direction at node i, vm is the displacement at node m, etc. (figure
3). The values of tij are constants given in [15].
x
i j
Fig. 3. Nodes used in G calculations

162
The expressions for GII and GIII were used with Fy replaced by Fx and Fz and v replaced by u and
w, respectively. The total strain energy release rate, G, is calculated by the formula:
(33)
G = GI +GII +GIII
Numerical application
The numerical tests are performed in taking a multilayer symmetrical angle ply composite laminate
made of graphite-epoxy:
E3 = 20x106 psi; E1 = E2 = 2.1x106 psi;
G12 = G23 = G31 = 0.85x106 psi; 12 = 32 = 31 = 0.21;
The dimensions of the plate: b = 8 in, h1 = h2 = h = 1 in, a = 1 in (figure 4).
The axial strain z is assumed to be uniform: z = e = constant.
y
h1
Singular metis elements
h2
Fig. 4. Mesh of the quarter of structure (56 elements)
Due to geometric and lamination symmetry, only a quarter of the cross sectional area is
considered.
Following results of the present method are compared with those of Wang [3] who used the
boundary collocation method and with those of Qian and Sun [4] who used the displacement ratio
method.
4.1 Stress intensity factors

The results of Ki by the present method, by Wang [3] and by Qian and Sun [4] for various fiber
orientations of degrees were presented in figures 5, 6 and 7.
0.7
KI/106z (psi in1/2)
0.6
0.5
-KI
0.4
Metis
[3]
KI
[4]
0.3
0.2
0.1
0
-0.1 0
15
30
45
60
75
90
Fig. 5: Stress intensity factor KI

163
The results in these figures indicate that present method and approaches of Wang [3] and of Qian
[4] bring back the almost identical value in KIII. There are substantial different in KI and KII, in the
magnitude and in the sign.
0.14
KII/106z (psi in1/2)
0.12
Metis
[3]
[4]
0.1
0.08
0.06
0.04
0.02
0
-0.02 0
15
30
45
60
75
90
Fig. 6: Stress intensity factor KII

The figure 6 shows that there is an important difference of KII between the results of Wang and
those of Qian. Qian explained this difference of KI and KII because of the difference of relative
displacement of surface of crack between the different methods.
-KIII/106z (psi in1/2)
5
4
Metis
[3]
[4]
1
0
-1
15
30
45
60
75
90
Fig. 7: Stress intensity factor KIII

The results of stress intensity factors Ki (I = I, II, III) for various fiber orientations of degrees
are presented in the figures 8.
1
Ki/106e (psi in1/2)
0
0
10
20
30
40
50
60
70
80
90
-1
-2
KI
KII
-3
KIII
-4
-5
Fig. 8: Stress intensity factor Ki of laminates under extension
164
It is found that the KIII is always clearly dominant. This tearing mode corresponds to a shearing
out of plan, due to the interlaminar stress yz. The KI is also significant. It is due to the normal stress
y. The KII, due to the interlaminar stress xy, remains always very weak. The maximum value of
the KIII is about six and thirty times higher than that of the KI and the KII, respectively. Therefore
the difference of the KI and KII in present solution with the solution of Wang and Qian may not
significantly influence the structural analysis.
The influence of laminate geometric variables on the delamination behavior is illustrated by
examining the change of KI, KII and KIII with the relative thickness of the upper and lower plies
h1/h2. In the graphite-epoxy composite [45o/-45o]s with:
h1 + h2 = w = 2 in; the crack length a = 1 in;
(34)
Laminate width 2b = 4 in; the loading condition z = e = constant.
The delamination stress intensity factors for various h1/h2 are presented in the figure 9. In a
general way, current curves resemble those of Wang, some of present values are extremely different
from those of Wang only for the KI and KII. The KIII of present result are lower than those of Wang
of 10% in magnitude and they have same sign.
1.2
-KIII
Ki/106z (psi in1/2)
1.0
0.8
Metis
[7]
0.6
-KI
KI
|KII|
0.4
0.2
0.0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 2
h1/h2
10
Fig. 9: Ki with various ratio h1/h2
4.2 Strain energy release rate

The total strain energy release rates G are shown in figure 10. The current result corresponds to the
result of Wang [3] and of Qian [4]. The maximum difference between present method and the
methods of Wang and of Qian is only 1%.
10
G/106z2 (psi in)
8
Metis
Wang [3]
Qian et al. [4]
6
4
2
0
-2
15
30
45
60
75
90
Fig. 10: Total strain energy release rate G
165
The figure 11 presents the influence of laminate geometric variables on the strain energy release
rate with the parameters given in (34). The result is compared with that of Wang [3]. The agreement
between the two methods is very good. The difference of the values of G in present method and in
method of Wang is less than 2%.
G/106z2 (psi in)
0.5
0.4
0.3
Metis
Wang [3]
0.2
0.1
h1/h2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10
Fig. 11: Total strain energy release rate G with various ratio h1/h2
4.3 Influence of interlaminar crack length to edge delamination

The figure 12 shows the total strain energy release rates G for the graphite-epoxy composite
laminates given in (33) with various interlaminar crack length and the effects of laminate width on
delamination crack extension.
The figure 12 indicated that present results are good agreement with the results of Wang. The
difference of the values of G between two methods is only about 1%. We see that, in two methods,
the energy release rate reaches a maximum value at a length ac approximately equal to one or twoply thickness in the considerable composite plate. The length ac depends on the (2b/2h) ratio.
0.7
b=8 in
0.5
0.4
b=4 in
b=2 in
0.3
Metis
Wang [3]
0.2
0.1
7.
8
7.
2
6.
6
6.
0
4.
2
3.
6
3.
0
2.
4
1.
8
1.
2
0.
6
0.
1
5.
4
a/h
4.
8
G/106z2 (psi in)
0.6
Fig. 12: G with various crack lengths and laminate widths
Conclusions
The metis finite element method has been shown to be accurate for calculating stress intensity
factors and strain energy release rate of edge delamination in composite laminates.
The metis finite element gives a new method to calculate and analyze structures in general and
composite structures in particular. This element is very useful for studying the crack or
166
delamination where there is a high stress concentration that demands a fine mesh when it is used by
other finite elements methods.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Nguyen-Dang Hung, On the monotony and the convergence of a special class of hybrid finite element: the
mongrel element, Variational methods in mechanics of solids (Ed. by S. Nemat-Nasser), Pergamon (1978).
Nguyen-Dang Hung, Sur un classe particulire dlments finis hybrides: les lments mtis, Proceeding of
International Congress on Numerical Methods for Engineering GAMNI, Dunod, Paris (1979), 53-63.
Wang S. S., Edge delamination in angle-ply composite laminates, A.I.A.A. Journal, Vol. 22, 1984, pp 256264.
Qian W., Sun C.T., Calculation of stress intensity factors for interlaminar cracks in composite laminates,
Compostes Science and Technology, Vol. 57, pp. 637-650, 1997.
Nguyen-Tien Duong, Nguyen-Dang Hung, Direct and indirect metis element methods for edge delamination
in laminates, COMPUTATIONAL MECHANICS WCCM VI in conjunction with APCOM04, Sept. 5-10,
2004, Beijing, China 2004 Tsinghua University Press & Springer-Verlag, Paper No 317.
Nguyen-Tien Duong and Nguyen-Dang Hung, Metis displacement method to calculate free-edge stresses
under extension, bending and twisting, ACOMEN 2005 3rd International Conference on Advanced
Computational Methods in Engineering, 30 May 2 June 2005, Ghent, Belgium, Paper No 15, 2005.
Nguyen-Tien Duong, Nguyen-Dang Hung, Using metis model to solve torsion problem of prismatic bar,
Vietnam Journal of Mechanics, VAST, Vol. 27, No. 3, pp. 149-157, 2005.
Nguyen-Tien Duong, Nguyen-Dang Hung, Regular and singular metis finite element model for delamination
in composite laminates, Finite Elements in Analysis and Design, pp. 650-659, Vol. 42, 2006.
Nguyen-Dang Hung, Gery de Saxc and Kang C. H., The computation of 2-D stress intensity factors using a
hibrid mongrel displacement finite elements, Engineering Fracture Mechanics, 1991, 38, pp 197-205.
Chi-Hang Kang, Une famille dlments hybrides singuliers pour ltude des plaques fissures mtalliques et
composites, Thse de doctorat, Universit de Technologie de Compigne, 1991.
Fleury C., Nguyen-Dang Hung., Guerlement G., Fryns G., Dveloppement et validation de modles pour la
caractrisation des proprits mcaniques et la description des mcanismes dendommagement de structures en
matriaux composites, Rapport SF-193, Universit de Lige, 1994.
Gry de Saxce, Conception dun lment fini hybride mtis monocouche pour la modlisation du delaminage
dans les matriaux composites, rapport interne du service M.M.S. de la F.P.Ms, Projet Multimatriaux,
octobre 1992.
Lekhnitskii, S.G., Theory of elasticity of an anisotropic body, holden-day, San/Fransco, California, USA,
1963.
Nguyen-Tien Duong and Nguyen-Dang Hung, Stress singularity order in the delamination of a composite
laminate, The 7th National congress on Mechanics, Hanoi - December 18-21, 2002, Page 151-159.
Raju I.S., Crews J.H. and Aminpour M.A., Convergence of strain energy release rate components for edgedelaminated composite laminates, Eng. Fract. Mech., Vol. 30, No3, pp. 383-396, 1988.
167
A United Algorithm for Limit State Determination

of Frames with Semi-Rigid Connections
*
Nguyen-Dang Hung*, Hoang-Van Long+

LTAS, Division of Fracture Mechanics, University of Lige, Belgium
+
Le Qui Don Technical University, Hanoi, Vietnam
Abstract
A computer program is proposed to calculate the limit states of the frameworks with the following
problems: Elastic analysis, Limit analysis, First-order and second-order elastic-plastic analysis,
Shakedown analysis, Optimization-Limit, Optimization-Shakedown have been realised by NguyenDang Hung in 1980, namely CEPAO. All of these calculations are connected with the problem of
automatic search of independent mechanism in the sense of limit analysis. This permits to solve a
lot of problems in a unified computer package with considerable reduction of memory and time
consuming. The elastic and elastic-plastic analysis use equilibrium-stiffness method, while other
above mentioned problems are solved by an internal linear programming algorithm. Concerning the
stable design concept, the package finds the practical discrete profiles existing in the Eurocode and
realises the stability checks. The present paper proposes an extension of this general software to the
case of semi-rigid connections. New implementations are performed such that large dimension
problems may be solved without difficulties. Some new numerical results for simple as well as large
dimension frames are presented showing the performances of the program CEPAO compared to the
recent literature.
Keywords: Limit analysis, Shakedown analysis, Elastic-plastic analysis, Optimization, Semi-rigid
connection, Linear programming.
Introduction
The CEPAO computer program had been early developed in the Department of Structural
Mechanics and Stability of Constructions of the University of Lige by Nguyen-Dang Hung et al. in
the 1980s [1-5]. This software is a unified package for solving automatically the following
problems happened for frame structures: Elastic analysis, limit rigid-plastic analysis with
proportional loadings; step by step elastic-plastic analysis; shakedown analysis with variable
repeated loadings; optimal plastic design with fixed loading; optimal plastic design with choice of
discrete profiles and stability checks; shakedown plastic design with variable repeated loadings;
shakedown plastic design with updating of elastic response in terms of the plastic capacity.
With the CEPAO, efficient choice between statical and kinematical formulations is realised leading
to a minimum number of variables and a considerable reduction the dimension of every procedure.
The basic matrix of linear programming algorithm is implemented under the form of a reduced
sequential vector, which is modified during each iteration. An automatic procedure is proposed for
the construction of the common characteristic matrices of elastic-plastic or rigid-plastic calculation,
particularly the matrix of the independent equilibrium equation. Application of duality aspects in
the linear programming (LP) technique allows direct calculation of dual variables and avoids
expensive re-analysis of every problem.
168
However, the old version of the CEPAO envisages only rigid joint frame, while the semi-rigid
behaviour of beam-to-column connections in steel frames is well recognized in the practices [6-18].
That is the fundamental reason for the present work. In fact, robustness of all procedures inside
CEPAO needs also to be appraised when the introduction of the semi-rigid behaviour of the
connections is implemented. On the other hand, real costs of the connections are newly taken into
account in the design problems.
General assumptions and fundamental equations
The following assumptions have been made

- Loading is quasi-static;
- The behaviour of the materials more precisely the section of the frames and their connections
subjected to bending is elastic - perfectly plastic. The semi-rigid behaviour of the connections is
characterized by two parameters: the initial stiffness R and the plastic capacity Mpc;
- The axial elongations of the frames are negligible;
- The weight of a member is proportional to the plastic capacity.
The equilibrium and compatibility relations
According to the references [1-3], the equilibrium and compatibility relations for all two types of
deflection mechanisms and joint mechanisms at the first- order are:
e = Cm ; r = C T w .
(1a,b)
In which e is the vector of reduced forces; m, r are the vector of bending moments and rotations of
the critical sections respectively; w is the vector of independent displacements; C is the matrix of
independent equilibrium equations, CT is the matrix of independent mechanisms.
The equations (1a,b) are the general equations used in all modulus of calculation in the CEPAO, in
the analysis as well as in the optimization, in the statical method as well as in the kinematical
method.
Elastic-plastic analysis
3.1 Elastic analysis by equilibrium-stiffness method
Fig.1: Element subjected to moment and normal force

Considerer now an element k with the length of lk and the semi-rigid joints, simultaneously
subjected to the bending moments and normal forces at its two ends like shown in Fig.1. Let
rkT = i j and m Tk = M i M j be the net rotation and the moment at ends of this element. We
have the following relation:

rk = Fk m k ,
where Fk is the flexibility matrix of the element taking into account large rotations:
169
Fk =
With:
uk =
lk
2
lk
6Ek I k
ii (u k ) + 6 / ik
ji (u k )
ij (u k )
.
jj (u k ) + 6 / jk
(2)
N k / E k I k ; ik = Rik l k / E k I k ; jk = R jk l k / E k I k ;
ii = jj = (3 / u k )(1 / 2u k 1 / tg 2u k ) ; ij = ji = (3 / u k )(1 / sin 2u k 1 / 2u k ) ,

where Ek is Youngs modulus and I k is the cross-sectional inertia, and Rik, Rjk are the initial
stiffness of the ends i, j respectively. In the case of first-order, we have ii(uk) = jj(uk) =2, and
ij(uk) = ji(uk) =1.
Let L k be localization Boolean matrix of the member k. The complementary strain energy of the
structure may write:
W = m T LTk Fk L k m / 2 = m T Fm / 2 .
k
Applying now the principle of minimum of the total complementary potential energy:
= m T Fm / 2 m Tr , one obtains:
m = F 1r = Hr
(3)
Let rC be the rotation vector corresponding to pinned joints or pinned supports, and rR be the rest
of the vector r. Equation (3) may be detailed as follows:
m R H RR
0 = T
H RC
H RC
H CC
rR
r .
C
(4)
Then the equilibrium relation (1a) and the compatibility relation (1b) become:
[C R
m
CC ] R = e ;
0
T
rR C R
r = T w .
C C C
From (4), one deduces m R = H RR rR ,

where
1
H RR = H RR H TRC H CC
H RC .
(5), (6)
(7)
(8)
Now replacing m R in (5) and taking account of Equation (6), one obtains finally:
w = K 1e ,
with
K = C R H RR C TR .
(9)
(10)
Equation (9) gives directly the independent generalized displacement in terms of reduced applied
forces e and the whole solution of the problem may be deduced in consequence. It appears that this
equilibrium - stiffness method allows getting the elastic solution by appropriate use of the basic
characteristic matrix C. The foregoing elastic calculations are necessary for an understanding of the
step-by-step elastic-plastic method described in the following.
3.2 Elastic-plastic analysis by hinge-by-hinge method in the first-order

We are dealing with the problem of determination of the entire history of deformation of the
structure when the multiplier increases to the limit value corresponding to the collapse state. The
method consists in the elastic solution of an auxiliary structure in which plastified critical sections
170
are replaced by pinned joints or hinges, and the determination of the increment of the loading
multiplier, which provokes the plastification of the next critical section. Local unloading is taken
into account by comparison of the signs of increments of moments to the corresponding former
values of the moment distribution.
Basing on the equations (7), (9), we can obtain the increment of the independent displacements and
the increment of bending moment due to an increment of loading multiplier from following
equations:
H C T K 1e
w = K 1e , m = RR R
,
0
with H RR and K are defined respectively by (8) and (10) corresponding to the auxiliary structure.
We mean an auxiliary structure the rest of the structure where plastic hinges are assimilated to
pinned-joints.
To determine the minimum increment , which provokes the next plastic hinge, one has to
calculate distribution of moment and to determine the section where yield condition is attained. The
next step consists in replacing the new plastic hinge by a pinned joint in a new rearrangement of the
rotation vector and its conjugated moment vector. The limit state is reached when the matrix K of
the considered auxiliary structure is singular. The displacements and the moments of the actual state
are finally obtained by the sum of the above auxiliary state: w = w 0 ,...,+ w n ;
m = m 0 ,...,+ m n . Whenever w and m are known, the displacements of all critical sections and
the distribution of normal forces and shear forces are determined as described earlier [1-5].
3.3 Elastic-plastic analysis by hinge-by-hinge method with P- effect

A step by step analysis based on an incremental variational principle of HEILLINGER-REISSNER
type has been proposed [5]. As the elastoplastic analysis hinge by hinge with the first-order
(paragraph 3.2), each step of this procedure leads to the appearance of a plastic hinge. In each step,
the iterations are realized by taking into account P- effect. The calculations in each iteration are
realised on the auxiliary structure in which plastified critical sections are replaced by pinned joints.
When the P- effect is considered, the relation between reduced forces and independent
displacements becomes [5]:
e = Kw K *u w = (K K *u )w .
The term K *u w is a correction taking into account the deformed geometry of structure. The matrix
K *u depends on the values of normal forces of the members. The detailed procedure is presented in
ref. [5].
Rigid-plastic analysis and design by LP
The theoretical development of rigid-plastic analysis and design by linear programming technique
has been extensively described in the literature in [19-23]. Here we simply restrict to describe some
practical aspects of the CEPAO package.
4.1 General formulation

The canonical formulation of the LP considered in the CEPAO is:
171
Min = c T x Wx = b
where is the objective function; x, c, b are respectively the vector of variables, of costs and of
second member. W is called the matrix of constraint. For commodity of the calculations, the
objective function is considered also as a variable, and the matrix formulation is arranged so that the
basic matrix of the initial solution is appeared clearly as in the following:
c1T
W1
x1
1 c T2 0
= .
0 W2 b
x 2
(11)
1 c T2
The basic matrix of initial solution is: X 0 =
.
0 W2
Equation (11) can be written under a general form:

W* x* = b* .
(12)
We will precise in the following the matrices W*, x*, b* and X0 for each problem.
4.2
Rigid-plastic analysis by kinematical method
This approach is based on the kinematical theorem which states that the collapse factor is the
smallest value among the set of multipliers + corresponding to licit mechanisms.
Change of variables
Let m is the vector of the plastic capacity of the sections. Let the vector s, such that their
components are: s i = M pci / M pi in the sections with the semi-rigid connections, and si = 1 in the
rest of sections (fully-rigid connections). The vector of realistic plastic capacity is: m s = s T m .
In the kinematical method, the unknowns are the rotations (r) and the independent displacements
(w). These quantities may have any sign (negative or positive). In linear programming procedure we
need non-negative variables so that we adopt the change the variables as in following:
r&i = r&i + r&i ; w& k' = w& k + w& 0 ; with r&i + , r&i , w& k' 0 .
The way to fix the value of w& 0 , which depends on the real structure, such that w& k' are always nonnegative is explained with details in the reference [24].
The second member of equalities (12) is not always non-negative, so an initial admissible solution
necessary for the simplex technique is somewhat not guaranteed in general situation. It appears that
the following arrangement leads to good behaviours of the automatic calculation. Let S be a
& 0 . Let E is a unity matrix of dimension nr x nr.
diagonal matrix, such that: S = diag 1 x sign of C T w
And let consider the new rotation rate and plastic capacity distribution:
r&+ = 0.5(E + S)r&+ + 0.5(E S)r& ; r& = 0.5(E S)r&+ + 0.5(E + S)r& ;
m s' + = 0.5(E + S) m s + + 0.5(E S) m s ; m s' = 0.5(E S) m s + + 0.5(E + S) m s .
Then, the vector of variables, matrix of constraint, vector of second member corresponding to the
problem (12) for limit analysis and shakedown analysis have the following form:
With limit analysis problem:
&'
x *T = w
r&+
r&
&0
b *T = 0 SC T w
&0 ;
+ eT w
172
0 T
W * = SC T
e T
m s'T+ 1
E
0T
m s'T 0
1 m s'T 0
0 ;
E
0 ; X 0 = 0 E
0 0 T 1
0 T 1
X 01
1 m s'T 0
E
0
= 0
0
0 T 1
With shakedown analysis problem:
&'
x *T = w
r& +
0 T
W * = SC T
0 T
s+
r&
&0
b *T = 0 SC T w
m s'T+ 1 m s'T 0
1
0 E
0 ; X 0 = 0
E
0
m TE + 0 m TE 1
1 m s'T
ms'T 0
E
0 ; X 01 = 0 E
0 m T
m TE 1
E
0
1
where is an artificial variable which must be taken out of the basis vector in the simplex process;
is a constant chosen in relation with the value of w& 0 [24]; mE is the vector of the envelope of the
elastic responses with the considered loading domain.
4.3
Rigid-plastic design by statical method
For design problems, kinematical approach is not efficient for automatic calculation. We adopt the
statical formulation for all optimal plastic design problems.
Objective function
The traditional costs function in the rigid-plastic design problem is the conventional weight of the
all elements. With the assumption that the weight of each element is proportional to the member
lengths and the member plastic capacity, with a structure having np groups of elements, the
objective function may be put down:
np
Z = M pk l k
(13)
k =1
When the semi-rigid behaviour of the connections is considered, it is necessary to take into account
the costs of the connections in the total costs [8, 12, 13, 18]. Some researches had solved this
problem [14-16]. In present work, we utilise the function of the total costs proposed in the reference
[15]. According to this reference, the total costs of a member i with two ends under semi-rigid
connections may have the following cost function:
Z i = i li ai + 0.2 i l i ai 0.8 i l i ai i + 1.6 i l i ai i2
(14)
where i, li, ai are respectively the material density, the member length and the cross section area.
The coefficient is the scale factor defined as: i = 1 /(1 + 3Ei I i / Ri l i ) . It is evident that: 0 1,
in which =0 at the pin-jointed end-connections and =1 at the fully rigid connections. We
observe in (14) that the cost of a steel member is increased by 20% if it has pin-jointed endconnections, and by 100% if its end-connections are fully rigid.
Coming back to the rigid-plastic design problem, but with the new objective function (14), for the
plastic design problem, it is convenient to define the conventional length calculating
by li = li + 0.2li 0.8l i i + 1.6l i i2 so that (13) has the similar form:
np
Z = M pk l k = m *T l
(15)
Change of variables
[ ]
Let D = d kj be a Boolean matrix which indicates the fact that kth design variable governs the
critical section j: m s = D T m * .
173
In the statical method, the unknowns are the vector of bending moments (m) or the vector of
residual moments (), which may have any sign (negative or positive). To have non-negative
variables with reduced number of variables, we adopt the change the variables in following:
For optimal plastic design:
m' = m + m s , then, 0 M 'j 2M pj .
m' = + m E + + D T m * 0 .
For shakedown plastic design:
Then, the vector of variables, the matrix of constraint, the vector of second member corresponding
to the problem (12) for optimal plastic design and shakedown plastic design have the forms below:
Optimal plastic design problem
x *T = m' T 1 P T
RT
0 T 1 0 T
E 0 E
*
W =
0 0 0
C 0 0
0T
QT
m *T ,
b *T = 0 0 T
1 0 T
0 2D T
0 E
; X0 =
0 0
0
T
0 0
E CD T
0T
0
E
0
lT
0T
eT ,
0T
0
.
0
0T
0
E
0
Where P, R, Q are slack variables; T is a technological matrix.
Shakedown plastic design problem
x *T = m' T PT Z s P+T Y T S T Q T m *T , b *T = 0 0 T 0 T (m TE + + m TE ) Cm E + ,
0 T 0 T 1
0
E 0
*
0
0
W = 0
0
E E
0
C 0
0T
0T
0T
0T
E
0
0
0
E
0
0
0
E
0
0
0
2D T
0
; X 0 = 0
T
0
0
T
0
CD
lT
0T
E
0
0T
0
E
0T
0
0
0
0
0
0
E
0
0T
0
0 .
0
E
Where P+ , P , S, Q and Y are non-negative slack variables.
4.4 Direct calculation of dual variables

In CEPAO, dual variables are obtained without using the dual alternative approach. The dual
variables are the moment distribution in kinematical approach and the independent displacements
(to obtain the real mechanism) in statical approach. Therefore, the dual properties of LP have to be
pointed out and physical significance of the dual variables has to be established. The details of these
theoretical deductions are presented in the references [1, 3].
4.5 Stability checks for steel structure

In the design of steel structures, the designed variables which are the bending plastic capacities
constitute the theoretical value. For practical purposes, an automatic choice of manufacture shapes
is performed by introduction of a list of variable shape of beams and columns (profiles IPE, HEA,
HEB). Whenever this choice is made, a systematic verification of stability conditions is carried out
for all members. We will summarize here the stability checks performed a posteriori in the CEPAO
package. The program has to find automatically an optimum shape if the stability conditions are not
satisfied [1-3]: Local buckling of flange; local buckling of the web; Influence of the normal and
174
shear force; buckling of compressed column, taking into account the effective length of column and
reduced coefficient of buckling. Concerning the influence of the semi-rigid behaviour of the
connections on the effective length of the columns, we modify the beam stiffness of a braced frame
by a factor of 1/(1+2EI/lRk) and that of an embraced frame by a factor of 1/(1+6EI/lRk).
Numerical examples
Four examples are proposed in this section to illustrate the applications of the CEPAO. In the three
first examples the results are compared with those obtained by some other authors (Raffaele
Casciaro et al. [25], F. Tin Loi et al. [9, 10], S. Baset et al. [26]). While, the fourth example
considers a large dimension frame analysed by CEPAO for different sorts of analysis and design.
For the semi-rigid frames, we apply the moment rotation relation for connections is used by F. Tin
Loi et al. [9]. In this reference, the semi-rigid elastic-plastic properties in pure bending condition,
namely initial stiffness and plastic moment capacity, are chosen according to the simple and elegant
classification system proposed by Bjorhovde et al. [11]. The initial stiffness of each bilinear
moment (Mpc) - rotation () relation is defined as:
Mpc =EI/db,
(16)
where is a constant; db is the connecting beam depth. The graphical illustration of this behaviour is
shown in Fig.2, where the ranges for rigid, semi-rigid and flexible behaviour are shown. Like in the
reference [9], intermediate values of moment capacities for given stiffness were interpolated here in
accordance with the dashed line shown Fig. 2. The example 2 and example 4 presented in this study
are calculated in term of the values of s and of reporting in the Table 1.
Table1: Relation between s and
s
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
25.000
10.0000
6.2667
4.4000
3.2800
2.5333
2.0000
1.1667
0.5185
0.0000
Note that the relation between in the (16) and in the (2) is: = l/db.
Fig. 2: Idealized moment rotation relation for connections

Example 1-Limit and shakedown analysis for rigid frame: A series of frames with different
numbers of story (ns) and bay (nb) being already considered by Raffaele Casciaro et al. [25] are
shown on the Fig. 3 (the units not mentioned). A constant story height hs = 300 and a constant bay
length lb = 400 is assumed for simplicity and three loading cases are considered: two distributed
vertical loads p1 and p2 and a seismic action defined as transversal force linearly increasing by P3
from the ground to the top floor (see Fig.3). Some mechanical properties are reported in Table 2,
and the load domain is defined by: 9p110; 0p25; -500P3500.
175
The Table 3 presents the results of limit and shakedown analysis. According to the results given by
CEPAO, with the case (*) (see Table 3, column 6), the incremental plasticity occurred, while the
fatigue (the alternating plastic) is appeared in the case (**). The sections for which the fatigue
occurs are A, B, C (see Fig.3) respectively for 46 frame, 59 frame and 610 frame.
Discussion: The load multipliers obtained by [25] and by CEPAO are coincided in the case of limit
analysis for all series of frames and in the case of shakedown analysis for 34 frame. While the
differences are respectively: -10,5%, -6,4% and -6,5% for 46 frame, 59 frame and 610 frame in
the shakedown analysis.
The Table 4 presents the load multipliers in the case of shakedown analysis for 46, 59, 610
frame, with the following assumption: The alternating plastic occurs in the sections A, B, C with the
envelope of bending moment calculated by the software SAP2000. As the load multipliers in the
Table 4 are the upper bounds, the real load multipliers cannot exceed these values. The differences
between the results obtains by CEPAO and the above-mentioned values is about from 3,5% to
6,4%, and those of ref. [25] are from 9,4% to 15,3%. It is useful to denote that the differences of the
value of the envelope of the bending moment between SAP2000 and CEPAO is due to the lumping
of the uniformly distributed load at the central point and the two ends of each element in the
CEPAO.
Table 2: Example 1 Mechanical properties for the series of frames
Young modulus (E)
Moment of inertia (I)
Plastic capacity (MP)
Column
300000
540000
1800000
Beam
300000
67500
450000
Fig. 3: Example 1-geometry and loads for the series frames
176
Table3: Example1 Load multiplier for limit and shakedown analysis calculated by ref. [25] and
CEPAO
Limit analysis
Type of frame
Shakedown analysis
(ns x nb)
Ref. [25]
CEPAO
Difference
Ref. [25]
CEPAO
Difference
34 frame
2.4612
2.4612
0.0%
2.0134
2.0102(*)
0.0%
(**)
46 frame
1.8610
1.8610
0.0%
1.3993
1.2655
59 frame
1.2000
1.2000
0.0%
0.7533
0.7076(**)
-6.4%
610 frame
1.1532
1.1532
0.0%
0.7209
0.6771 (**)
-6.5%
-10.5%
Table 4: Example 1 Load multipliers for the fatigues occurring in the section A, B, C using the
envelope of bending moment calculated by SAP2000.
Envelope of bending moment
Sections
Plastic capacity
Load Multiplier
M+
M-
(Mp)
=2Mp/(M + + M -)
Section A (46 frame)
282717
477057
450000
1.1846
Section B (59 frame)
563240
757155
450000
0.6816
Section C (610 frame)
591835
785758
450000
0.6533
Example 2-Limit and shakedown analysis for semi-rigid frames: These examples are already
considered by F. Tin Loi et al. [9]. The aim of this study is to find the difference between
shakedown and collapse limits for varying connection strengths with two following mechanical and
geometric data (in tonne and metre units):
Fig. 4: Example 2 Frame geometry and loading
Data a: The frame geometry and loading are shown in the Fig.4a, the properties of all elements
are: E = 2.1107; I = 118.510-6; MP = 20; db = 0.3;
Data b: The frame geometry and loading are shown in the Fig.4b, with the following properties:
for the column: E = 2.1107; I = 85.210-6; MP = 10; for the beams: E = 2.1107; I = 118.510-6; MP
= 20; db = 0.3.
177
These two examples are analysed by CEPAO. The results are compared with those from F. Tin
Loi et al. [9] and illustrated on the Table 5 and the Fig. 5a and 5b. With the data a, the alternating
plasticity occurs in the shakedown analysis with s = 0.1 and 0.2.
Table5: Example 2 Load Multipliers
Connection strengths (s)
Type of analyse
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Example 2, data a
*
Collapse by ref. [9]
Collapse by CEPAO
3.70
4.01 4.36 4.67
5.05 5.22 5.43
5.60 5.81 6.02
3.67
4.00 4.33 4.67
5.00 5.20 5.40
5.60 5.80 6.00
Shakedown by ref. [9]
2.42
3.04 3.77 4.01
4.25 4.53 4.77
5.01 5.29 5.57
Shakedown by CEPAO
2.54
3.28 3.78
4.28 4.54
4.81
5.06 5.30
4.03
5.56
Example 2, data b
*
Collapse by ref. [9]
Collapse by CEPAO
0.53
0.80 1.02 1.14
1.25 1.34
1.42
1.49 1.19 1.49
0.53
0.80 1.02 1.14
1.25 1.33
1.42
1.48 1.48 1.48
Shakedown by ref. [9]
0.50
0.71 0.91 1.00
1.12 1.18
1.25
1.31 1.35 1.35
Shakedown by CEPAO
0.50
0.71 0.91 1.01
1.12 1.19
1.26
1.32 1.35 1.35
(*) We obtain these results by treating the figure in reference [9] (Fig. 6 and Fig. 8) in the case of pure bending hinges.
Fig.5: Example 2-Variation of load multiplier with connection strength (Tab. 5)

a- data a; b- data b
Example 3 Elastic-plastic analysis with 2nd order for rigid frame: The purpose of this example is
the determination the load multiplier under second order effect for the rigid frame shown in Fig. 6a.
a. The data is used by F. Tin Loi et al. [10], in ton and inch units: E = 1.3393104, H = 144; L =
90; 2P1 = P2 = 8.375; 2P3 = P4 = 0.402; members 1,2: I = 115.06; MP = 502.3; members 5, 6: I =
86.69; MP = 393.5; members 9, 10: I = 43.69; MP = 259.3; members 13, 14: I = 34.71; MP = 205.3;
members 15, 16: I = 55.63; MP = 244.0; and other members: I = 122.34; MP = 428.0.
We observe a very good agreement between the value given by CEPAO (2.036) and the one by F.
Tin Loi et al. [10] (2.037)
b. The data is analysed by S. Baset et al. [26], in kip and inch units: E = 3103; H = 144; L =
180; P1 = P2 = 30; 2P3 = P4 = 7.2; members 1,2,5,6,9,10,13,14, I = 344; MP = 2560; members
11,12,15,16: I = 891; MP = 3822; members 3, 4,7,8: I = 984; MP = 4204.
CEPAO provides a value of 2.123, while S. Baset et al. [26] give 2.079. The difference is 2.07%.
178
At Fig.6 b, c, d, e we compare the hinge dispositions in the frame given by CEPAO and other
researchers [10, 26].
Fig. 6: Example 3 Frame geometry and loading and hinge dispositions

a- Frame geometry and loading; b, c, d, e - The hinge dispositions (b- Ref. [10], c CEPAO with
data a, d Ref. [26], e - CEPAO with data b)
Example 4 Analysis and optimization for large dimension semi-rigid frame by CEPAO: A twenty
stories three bays semi-rigid frame with geometry and loading shown on the Fig.7 is analysed by
CEPAO with the following studies: elastic analysis; first order elastic-plastic step by step analysis;
elastic-plastic step by step analysis with second order effect; rigid-plastic analysis; shakedown
analysis; optimization-limit; optimization-shakedown.
Concerning loading domain, for shakedown problems, two cases are considered: a) 011, 02 1
and b) -111, 021. For fixed or proportional loading obviously we must have: 1=2=;
For the optimal problem, forty different groups of elements are chosen as conception variables
(Fig.7) and the adopted fixed load factor is = 0.25. Here only the costs of semi-rigid connections
are considered. In the optimal-shakedown problem, the results of the iterative process consisting of
updating the inertias depending on the plastic capacity: Ik/Imax = (Mpk/Mpmax)1.4.
For the analysis problems, seven different groups of elements are considered (Table 6). The yield
stress p = 23.5104 KN/m2.
Tables 7, 8 and Fig.8-11 present some results, in which KN and m units are used.
Table 6: Example 4 Profile used for analysis problems
Groups
1-5
6-10
11-15
16-20
21-25
26-30
31-35
36-40
Profile
IPE550
IPE500
IPE450
IPE330
HE600A
HE550A
HE450A
HE360
Table 7: Example 4 Load Multipliers of analysis problems

Connection strengths (c)
Type of analyse
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Rigid-Plastic
0.080 0.121 0.162 0.202 0.244 0.284 0.324 0.360 0.396 0.432
Elastic-plastic first order
0.080 0.121 0.162 0.202 0.244 0.284 0.324 0.360 0.396 0.432
Elastic-plastic second order
0.053 0.088 0.127 0.167 0.205 0.241 0.280 0.316 0.354 0.392
Shakedown, load domain (a)
0.065* 0.110 0.145 0.181 0.217 0.253 0.288 0.324 0.359 0.394
Shakedown, load domain (b)*
0.037 0.070 0.096 0.134 0.166 0.198 0.229 0.260 0.290 0.320
(*) alternating plastic occurs.
179
Table 8: Example 4 Results of optimal problems

Connection strengths (s)
Type of optimization
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Theoretical weight (x106)

Optimal Limit (*)
0.280
0.242
0.215
0.196
0.180
0.168
0.158
Optimal Shakedown (*)
0.315
0.260
0.230
0.208
0.192
0.192
0.169
Optimal Limit (**)
0.408
0.385
0.378
0.378
0.383
0.393
0.405
Real weight (tonne) after stability checks

Optimal-Limit (*)
62.13
59.84
56.80
52.18
51.73
49.83
49.41
Optimal-Limit (**)
67.73
66.53
64.53
61.00
64.32
66.93
71.99
(*) member weight considered, (**) member + semi-rigid connections weight considered.
Fig. 7: Example 4- Frame geometry, groups of element and loading

180
Fig. 8: Example 4-Load-deflection result of step by step analysis

a - First order; b - Second order
Fig. 9: Example 4- Variation of Load

Multipliers with connection strength
(Table 7)
Fig. 10: Example 4-Load-deflection

result of rigid frame
Fig. 11: Example 4-Variation of weight with connection strength (Table 8)

a -Theoretical weight; b-Real weight of optimal-limit
Remark: In the case of small connection strengths or symmetric loading (seismic action in the
example 1, horizontal load with load domain b of example 4), the load multipliers determined by
shakedown analysis are the smallest (alternating plastic occurs). In the design problems, the
member plus connections costs is minimum at some values of connection strength (it is s=0.7 in the
example 4, see Fig.11b), this value depends on the determination of the conventional length. On the
other hand, the determination of the conventional length depends on a lot of parameters: the cost of
material, the cost of fabrication (the cost of the labour), and these parameters may depend on
country.
Conclusions
The CEPAO envisages the solutions for various loading conditions (proportional, repeated,
alternative), for various aspects (direct analysis and design, step-by-step analysis) of the
181
behaviour of the frames. An optimal solution may be found according for each practical case.
CEPAO may be useful for the practice in civil engineering.
CEPAO is also an auto-controlled algorithm. Indeed, we can verify easily the results by using
resident equivalent procedure. For example, limit analysis and analysis hinge-by-hinge must lead to
the same limit load factor, while they are based on two dual methods (kinematical method and static
method). After the optimal plastic design (or shakedown plastic design), the plastic analysis (or
shakedown analysis) may be operated to reanalyse for checking if the limit load factor would be
equal to 1 as theoretically expected. On the other hand, we can observe that the two modules of
elastic analysis (by equilibrium-stiffness method) and limit analysis (by LP method) constitute the
fundamental modules, which may lead to more complicated implementations, more realistic
computations. The present extensions suggest that CEPAO constitutes a source for future
implementations and researches in civil engineering practices.
In the near future, we hope to present the new version of the CEPAO for the space frameworks.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
Nguyen-Dang Hung, CEPAO-an automatic program for rigid-plastic and elastic-plastic, analysis and
optimization of frame structure, Eng. Struc. Vol 6, 33-50, 1984.
Nguyen-Dang Hung, Gry de Saxc, Analyse et dimensionnement plastique des structures barres dans les
conditions de stabilit, Construction mtallique, N3,1981.
Nguyen-Dang Hung, Sur la plasticit et le calcul des tats limites par lments finis, Thse de doctorat
spcial, lUniversit de Lige, 1984.
Nguyen-Dang Hung, Sur lutilisation du simplexe dans le CEPAO 82, Rapport interne n133 du Laboratoire
de Mcanique des Matriaux et de Stabilit des Constructions, lUniversit de Lige, Belgique, 1983.
Gry de Saxc, L. M. Ayina ohandja, Une mthode automatique de calcul de leffet P - Delta pour lanalyse
pas - - pas des ossatures planes, Construction mtallique, n3, 1985.
Chen, W.F., Lui, E.M., Stability design of steel frames, Boca Raton, FL: CRS Press, 1991.
Chen, W. F, Yoshiaki G., Richard, J. Y., Stability design of semi-rigid frames, John Wiley & sons, inc,
1996.
Jaspart, J. P., Etude de la semi - rigidit des nuds poutre - colonne et son influence sur la rsistance et la
stabilit des ossatures en acier, Thse de doctorat de lUniversit de Lige, Belgique, 1991.
F. Tin Loi, V. Vimonsatit, Shakedown of frame with semi-rigid connections, Journal of Structural
Engineering, 119(6), 1694-1711, 1993.
F. Tin Loi, V. Vimonsatit, Nonlinear analysis of semi-rigid frames: a parametric complementarity approach,
Engineering Structure, 18(2), 115-124, 1996.
Bjorhovde, R., Colson, A., and Brozzetti, J., Classification system for beam-to-column connections, Journal
of Structural Engineering, 116(11), 3059-3076, 1990.
A. Colson, R. Bjorhovde, Intrt conomiques des assemblages semi-rigides, Construction mtallique, n2,
1992.
Stphane Guisse, Quelle conomie attendre de la mise en oeuvre de noeuds semi-rigides?, Construction
mtallique, n3, 1993.
Lei Xu and al., Computer- Automated Design of Semi-rigid Steel Frameworks, Journal of Structural
Engineering, 119(6), 1741-1760, 1993.
182
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
L. M. C. Simoes, Optimization of frames with semi-rigid connections, Computers & structures, 60(4), 531539, 1996.
M. S. Hayalioglu and al., Minimum cost design of steel frames with semi-rigid connection and column bases
via genetic optimization, Computers & structures, 83, 1849-1863, 2005.
Eurocode 3- design of steel structures, Part 1.1: General rules and rules for buildings, European prestandard.
ENV 1993-1-1, February 1992.
J. M. Cabrero, E. Bayo, Development of practical design methods for steel structures with semi-rigid
connections, Engineering Structures, 27, 1125-1137, 2005.
Hodge. P. G., Plastic analysis of structures, McGraw Hill, New York, 1959.
Neal. B. G., The plastic method of structural analysis, Chapman & Hall, London, 1970.
Massonnet, Ch., Save, M., Calcul plastique des constructions, Volume 1, Nelissen, Belgique, 1976.
Cohn, M. Z, Maier, G., Engineering Plasticity by Mathematical Programming, University of Waterloo,
Canada, 1979.
Milan Jirsek, Zdenk P. Baant, Inelastic Analysis of Structure, John Wiley & Sons, LTD, 2001.
Nguyen-Dang Hung, Hoang-Van Long, On the change of variables in the calculation plastic of structures by
LP in the CEPAO program, Internal Report of Division LTAS-Fracture Mechanics, University of Lige, 2006.
Raffaele Casciaro, Giovanni Garcea, An iterative method for shakedown analysis, Computer methods in
applied mechanics and engineering, 191, 5761-5792, 2002.
S. Baset, D.E. Grierson and N.C.Lind, Second Order Collapse Load Analysis: a LP Approach, Solid
Mechanics Division University of Waterloo, Canada, N117, 1973.
183
Modified 8-Node Quarter-Point Element in Calculating Stress Intensity

Factors of 2D Crack Problems
Le-Hoai Long* and Nguyen-Dang Hung+
*
Faculty of Civil Engineering,
National University of HCMC, University of Technology, VietNam
+
Abstract
8-node quarter-point element was adjusted to satisfy: the coefficient of linear term in the
distance to the tip must vanish. In this paper, we will compare this modified element with a
broadly used element collapsed 8-node quarter-point (CQ8) element. We employed the
Displacement Correlation Technique to assess the stress intensity factors (SIFs) and Ansys
software was used to compute the CQ8 results.
Keywords: fracture mechanics, finite element method, stress intensity factors.
Introduction
The finite element method (FEM) has become firmly established as a versatile tool for the
numerical solution of engineering problems and would at first sight appear to be an ideal
method of studying crack behaviour in materials.
The fracture behaviour of cracked structures is dominated by the near-tip stress and strain
fields. In numerical modelling of fracture problems, a correct representation of local stress
and displacement fields in crack tip region is essential to accurately evaluate SIFs.
In 2D linear elastic fracture mechanics [1,6,7,9], for crack geometry, Williams proved that
the form of displacement u = {uk}, k = 1, 2 in the neighbourhood of the tip (Figure 1) is:
(1)
u k ( r , ) = a k + bk ( ) r 1 / 2 + c k ( ) r + d k ( ) r 3 / 2 + O (r 3 / 2 )
with:
r: is the distance to the crack tip.

: is the polar angle with respect to the crack tip.
x2
x1
Figure 1: coordinate system at crack tip.

L.J.Gray and Paulino [1] proved that, with traction free boundary condition on the crack
surfaces, the series expansion in Equation 1 must have c2() = c2(-) = 0 and c1() = c1(-)
184
on the crack surface. The mathematical crack results when the interior angle occupied by the
material is 2 so = and the crack surfaces correspond to = .
Long [21] modified the 8-node quarter point element proved by Barsoum to satisfy above
constraints. And also in this document, SIFs in 2D problems obtained from modified quarter
point (MQP) element are better than these from unmodified quarter point (UQP) element.
In this paper, we will investigate and compare the SIFs of MQP with a very broadly used
element, collapsed 8-node quarter point element (CQ8 element made by collapse side 1-3 or
4-7 of original 8-node QP element). Ansys software is used to compute SIFs of CQ8
element.
8-node quarter-point elements
2.1 MQP and UQP elements

7
6
5
3Le
4
x2
Le
4
x1
Le
4
3L e
4
Figure 2: 8-node quarter point element

The geometry and displacements of 8-node element in Figure 2 are interpolated by:
8
x k = N i ( , ) xi
i =1
8
u k = N i ( , )u i
(2)
i =1
k = 1,2
The crack faces are oriented along the line 1-3 ( = -1). Barsoum [2] proved that
+ 1 = 2(r L )
and so the strain singularity along the line 1-3 (or line 1-7) is therefore
1 r , which is the required singularity for elastic analysis. The next term ( + 1) 2 will
be ~ (r L ) .
The SIFs can then be evaluated by:
(
2
1
)
cos
cos
2
2 = 4 2 4u 2 u 3 3u1
K1
L 4v 2 v3 3v1
(2 + 1) sin sin 3
2
2
(3)
and
185
(2 + 3) sin 2 sin 2
2 4u 2 u 3 3u1
K2
= 4
L 4v 2 v3 3v1
(2 3) cos + cos 3
2
2
(4)
With MQP, the approximation of geometry xk is kept as in Equation 2, so that the property
+ 1 = 2(r L )
remains, and the interpolation of the geometry remains quadratic. With the
displacements, cubic terms are added into shape functions as follow:
N i* = N i + (1 )(1 + )
(5)
with:
- i = 1, 3 = 1 ; = 1
6
- i = 2 = 1 ; =1
3
- i = 4 8 = 0
- if the crack tip at node 1, = 1 and if the crack tip at node 3, = -1.
The SIFs can then be evaluated by:
and
(2 1) cos 2 cos 2
4
2 7u1 8u 2 + u 3
K1
=
3
3
L 7v1 8v 2 + v3
(2 + 1) sin sin
2
2
(6)
(
2
3
)
sin
sin
2
2 = 4 2 7u1 8u 2 + u3
K2
3
3
L 7v1 8v2 + v3
(2 3) cos + cos
2
2
(7)
2.2 Collapsed 8-node quarter point (CQ8) element

This element has the singularity property along two edges (1-3 and 1-5) and along the ray
emanating from the tip to the midside node on the opposite edge. The mesh grid using this
CQ8 element need the input parameters as: number of element around the tip, radius of
rosetteIn some cases, with incompatible input parameters, the mesh can not be done. This
makes some inconvenience.
x2
5
x1
6
7
8
1
2
Le/4
3Le/4
Figure 3: collapsed 8-node quarter point element
186
The formulas for assessing SIFs of this element will be the same as ones of UQP (above).
In this paper, some examples will be carried out to survey the effectiveness of MQP
compared with CQ8 in computing SIFs.
Numerical examples
For all examples, consistent units are used and the following material constants are
employed: Young's modulus E = 10000, Poisson's ratio = 0.3.
The Ansys software is used to apply CQ8 in calculating SIFs. The input parameter
geometries are:
Ansys 4E : 4 CQ8 elements around crack tip, radius of rosette is 2 (coarse).
Ansys 8E : 8 CQ8 elements around crack tip, radius of rosette is 1 (quite coarse).
Ansys 16E : 16 CQ8 elements around crack tip, radius of rosette is a/8 (fine) (a : half
crack length).
With MQP and UQP, an example meshgrid is shown in Figure 4. Le is crack tip element
edge length (see Figure 2, 3)
Figure 4: An example meshgrid of MQP and UQP.
3.1 3-point bending beam

This example deals with bending deformation. A 3-point bending beam under opening and
sliding mode is shown in Figure 5. This situation can be checked in plane strain.
3.1.1 Mode I case (d = 0)
SIF of mode I is evaluated for a range of geometry of crack (a/W = 0.4 to 0.7) using Le = 2
(coarse mesh), L = 50, W = 10 and P = 100. Table 1 lists the errors compared to the
analytical solutions proposed by Guinea et al [8,9].
187
W
a
L/2
L
Figure 5: 3-point bending beam

Figure 6 plots the absolute deviation of SIF of MQP, UQP and Ansys 4E, 8E, 16E from
the exacts.
a/W
MQP
UQP
Ansys 16E
Ansys 8E
Ansys 4E
0.4
0.5
0.6
0.7
0.9721
0.9775
0.9862
0.9994
0,8339
0,8167
0,7947
0,7733
1.0110
1.0069
1.0048
1.0078
1.0203
1.0088
1.0053
1.0060
0.9252
0.9122
0.9089
0.9090
Table 1: K1/K1exact of 3 point bending beam mode I.

We can see that, MQP solutions are in good agreement with exacts if compare with Ansys
fine mesh (Ansys 16E and 8E). The large deviation of MQP may be because the ratio Le/a (or
the fineness of the mesh grid) is not the best as proposed in [21]. But when we use MQP with
the coarse mesh, we can get a very big reduction of error.
25,0
% error of K1
20,0
MQP
15,0
ANSYS 16E
ANSYS 8E
ANSYS 4E
10,0
UQP
5,0
0,0
0,4
0,5
0,6
0,7
a/W
Figure 6: Comparison with Ansys software

3.1.2 Mixed mode case (d0)
Beam dimensions, P = 100 and a = 5 are reused here. Load locations change d/W = 0.25; 0.5;
1; 1.25. In this situation, we just do a comparison between Ansys 16E and the results of
MQP, UQP with Le/a = 0.3.
188
The quantity G is used in this example because the SIF of mode II is rather small than that
of mode I, the strain energy release rate G is considered to show the combined effect in 2D.
In two dimensions, this quantity is:
G=
+1 2
( K 1 + K 22 )
8
The results obtained here will be compared with data from [4] (Table 2) and plotted in
Figure 7 as the deviation from [4].
The MQP solutions for mixed mode case has mightly less agreement with [4] than Ansys
fine mesh, but their errors are always less than 5% with coarser mesh. And mind that,
reference [4] is just another numerical analysis, it is a good but not an absolutely exact
reference. This might serve as an indication to revise the proposed expression in [4].
d/W
MQP
UQP
Ansys 16E
0.25
0.5
1
1.25
0.9569
0.9736
1.0072
0.9849
0.7508
0.7511
0.7553
0.7538
0.9972
1.0091
0.9967
1.0012
Table 2: G/G[4] of 3 point bending beam mode I.

2
Deviation of G (%)
-2
-6
ANSYS 16E
-10
MQP
UQP
-14
-18
-22
-26
0,25
0,5
1,25
d/W
Figure 7: Comparison with Ansys software mixed mode
3.2 Centre crack

Consider a finite plate containing a centre crack, as in Figure 8, subjected to the uniaxial
tension = 100 applied along the y-axis.
189
H
2a
Figure 8: Centre crack

The plate dimensions are 2H = 50, 2W = 20 and thickness t = 1. The crack half length
changes: a = 2, 3, 4, ,6.
With MQP and UQP, the crack tip is discretized with the same crack tip element edge
length (Le) equal to 2.
The absolute errors of both two types of element and Ansys 8E, 16E compared with the
exact ones are plotted in Figure 9. It can be seen from Table 3 that the solutions using MQP
element are in better agreement with the analytical solution (exact solution) than Ansyss
though Ansys mesh grids are very fine.
With Ansys solutions, the mesh Ansys 16E gives better results than 8E.
Le/a
MQP
UQP
Ansys 16E
Ansys 8E
0.2
0.3
0.4
0.5
0.6
0.9937
0.9905
0.9924
1.0262
1.0533
1,1361
1,0501
1,0132
1,0376
1,0569
1.0070
1.0231
1.0328
1.0543
1.0808
1.0145
1.0237
1.0399
1.0584
1.0840
Table 3: % error of K1 of centre crack
190
16
14
12
% error of K1
10
MQP
8
ANSYS 16E
ANSYS 8E
UQP
4
2
0
-2
0,2
0,3
0,4
0,5
0,6
a/w
Figure 9: Error of centre crack pure mode I case
3.3 An infinite row of collinear cracks
2a
2H
2c
Figure 10: an infinite row of collinear cracks

This is an example of an infinite row of interactive cracks and these cracks are separated by
the gap 2c (Figure 10).
The plate can be considered as an infinite domain with 2H = 50, a = 5, Le/a = 0.4. The
ratio c/a will change from 0.2 to 4. This plate is subjected to uniaxial tension = 100 applied
in the direction perpendicular to the cracks. The SIF of mode I is considered in the following.
The error (%) of K1 results of UQP, MQP and Ansys 16E compared with the exact ones
are shown in Table 4. Figure 11 plots the relative errors of UQP and MQP results with
respect to the exact solutions. Regardless the ratio c/a, MQP solutions always reach very
accurate values. These errors are less than 1% even when the ratio c/a is less than onequarter. When the distance between cracks is wide (more than 3) MQP results are also better
than Ansyss.
191
6,0
% error of K1
5,0
4,0
MQP
3,0
ANSYS 16E
UQP
2,0
1,0
0,0
0,2
0,5
0,8
1,6
2,0
3,0
4,0
c/a
c/a
Figure 11: Effect of crack interaction

MQP
UQP
Ansys 16E
0.20
0.50
0.80
1.60
2.00
3.00
4.00
0.9921
0.9941
0.9973
0.9992
0.9943
1.0011
1.0051
1,0515
1,0277
1,0269
1,0261
1,0166
1,0230
1,0227
0.9963
0.9992
1.0021
1.0030
1.0039
1.0105
1.0202
Table 4: K1/K1exact of collinear cracks

With a small modification over 8-node quarter point element, the obtained results are very
good. The errors decrease very much if they are compared with the original element (UQP).
Moreover, using MQP can make the effect also better than Ansys software using collapsed 8node quarter point element with such a simple coarse mesh that can be meshed by hand.
Conclusions
The MQP element leads to more accurate computation of SIFs than UQP's. This situation is
maintained even with relatively coarse mesh, and so with very simple displacement
determination of SIFs. It appears that MQP element may be very useful in simulating the
crack propagation because there is no need to proceed with refined meshing and complex inraising-mesh-grid element, CQ8 element
With the coarse mesh and simple SIFs assessment, MQP element may simulate cracked
responses without using any expensive computation.
As the modified element tries to improve the local COD solution near the tip, its impact on
non-local SIF assessment methods like J-integral is limited. Extension to 3D will be
considered in a future paper.
192
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
L.J. Gray and Glaucio H. Paulino, Crack tip interpolation, revisited, September 3, 1996.
Roshdy S. Barsoum, On the use of isoparametric finite elements in linear fracture mechanics,
International journal for numerical methods in engineering, Vol 10, 1976.
Gery de Saxce and Nguyen-Dang Hung, Modified Formula for three-dimensional Direct StressIntensity Factor Computation, Communication. in Appl. Num. Meth, Vol. 6, pp 223-230, 1989.
T.FETT, Crack intensity factors and weights functions for special crack problems,
Forschhungszentrum Karlsruhe, Technik und umwelt Wissenschaftliche Berichte, FZKA 6025, 1998.
L.J. Gray, A-V. Phan, Glaucio H. Paulino, T. Kaplan Improved quarter point crack tip element,
Pergamon Engineering fracture mechanics, 2003.
Abdel-Rahman Ragab, Salah Eldin Bayoumi, Engineering solid mechanics, fundamentals and
applications, CRC Press, 1999.
J.R. Barber, Elasticity, Kluwer academic publishers, 1992.
T.L. Anderson, Fracture mechanics Fundamentals and applications, CRC Press, 1995.
G.C. Sih, Mechanics of fracture, Methods of analysis and solutions of crack problems, Noordhoff
international publishing leyden, 1973.
D.R.J. Owen, A.J. Fawkes, Engineering fracture mechanics: numerical methods and applications,
Pineridge Press Ltd, 1983.
M.H. Aliabadi, D.P. Rooke, Numerical fracture mechanics, Kluwer academic publishers, 1991.
Carlos A. Felippa, Introduction to finite element methods, Course at the Aerospace Engineering
Sciences Department of University of Colorado, 2003.
M.J. Fagan, Finite element analysis, Theory and Practice, Longman Scientific & Technical, 1992.
R.D Cook, D.S. Malkus, M.E. Plesha, R.J. Witt, Concepts and applications of finite element analysis,
John Wiley & Son Inc., 2002.
Klaus-Jurgen Bathe, Finite element procedures, Prentice-Hall International Inc., 1996.
Young W. Kwon, Hyochoong Bang, The finite element method Using MATLAB, CRC Press, 2000.
Yan Ai Min and Nguyen-Dang Hung, Stress intensity factors and crack extension in a cracked
pressurised cylinder, Engineering Failure Analysis, Vol 1, Issue 4, pp 307-315, December 1994.
Yan Ai Min and Nguyen-Dang Hung, Elastic analysis of the multiple-cracked structure by DBEM,
Compte rendu du 3me Congrs national belge de Mcanique thorique et appliqu. Lige, May, pp 14, 1994.
Yan Ai Min and Nguyen-Dang Hung, Multiple-cracked fatigue crack growth by BEM,
Computational Mechanics, Vol. 16, No. 5, pp273-280, 1995.
Tran-Thanh Ngoc and Nguyen-Dang Hung, Analysis of cracked plates and shells using "metis" finite
element model, Finite Elements in Analysis & Design (Elsevier), Volume 40, Issue 8, pp. 855-878,
May 2004.
Le-Hoai Long, Modification of a quarter point element in calculating fracture mechanics, Master
thesis, 2005.
193
Dual Analysis for Finite Element Solutions of Kirchhoff Plates

J.F. Debongnie*, Nguyen-Xuan Hung+ and Nguyen-Huy Cung++
*
Manufacturing Laboratory, University of Lige, Belgium
+
Division of Computational Mechanics, Department of Mathematics and Informatics,
University of Natural Sciences-VNU-HCMC, Vietnam
++
Department of Technical Structures, University of Transport -HCMC, Vietnam
Abstract
This paper presents the application of the dual analysis concept to plate bending. In this
method, a same problem is analyzed parallely by a displacement and an equilibrium model.
The energetic distance between these two models is the sum of both global errors and
consequently, an upper bound of each of them. After an exposition of the two models,
numerical examples are presented, which illustrate the high obtainable accuracy of the
method.
Keywords: plate bending, equilibrium element, conforming element, dual analysis
Introduction
Most of todays engineering analyses deals with problems involving differential equations
which are too difficult to be solved analytically. Currently, one of the most popular numerical
methods for solving these problems is the Finite Element Method (FEM) which has become a
very effective and powerful tool for numerical simulations in engineering and science.
The best known family of finite element models is the displacement model, in which the
compatibility equations are a priori verified, from the fact that the variables are
displacements. The solution of the problem then leads to weak forms of the equilibrium
equations. But there exists a dual family of finite elements, which are called equilibrium
elements. In these elements, an equilibrated stress field is used, and the result of the
computation consists in weak compatibility equations.
A same problem may be treated by both methods. An interesting fact is that the
comparison of the results obtained by these two approaches leads to a useful error measure.
This is the so-called dual analysis. Its initial form, which was developed by Fraeijs de
Veubeke in the 60s [1], is based on the fact that under some restrictive conditions on the data
(homogeneous prescribed displacements or zero load), the two models give an upper and a
lower bound of the energy, respectively. The difference between these energies is an
energetic measure of the sum of the errors of both models. The above mentioned restrictions,
the lack of equilibrium models in most FEM codes and the fact that at this time, performing
two analyses of the same problem was considered as too time-consuming were at that time
obstacles to a wide use of the method.
The huge improvement of computers in the recent times significantly modified the
situation. By now, a double analysis is no more an obstacle. Moreover, newer investigations
194
[5, 7] reformulated the dual analysis in a more general way, where energy bounds do no
longer play a central role. The result of this reconsideration is that in fact, dual error bounds
do exist for all cases, not for the strain energy, but for the total complementary energy. This
all gives to the dual analysis a renewed attractiveness.
This paper is devoted to the application of the dual analysis to plate problems. After a
presentation of the problem and a recall of the dual analysis principles, an equilibrium
element is presented, which allows the appliance of a pressure field. Finally, numerical
examples illustrate the estimation of the error by dual analysis as a comparison of total
energies.
General notations and the spaces of admissible fields
In what follows, the plate will be described as a plane domain with a Lipschitz-continuous
boundary . A pressure p is acting within the domain. Boundary is split in two parts,
namely 1 where kinematical conditions (u and un) are prescribed, and 2 where loads are
applied.
Using the notation ij for the moment field, one can write the internal equilibrium
equations in the following form,
Dij ij + p = 0
in
(1)
2
= ij are the partial derivatives of second
where p is the transverse load and Dij =
xi x j
order.
The compatibility equations consist to say that the curvatures ij derive from a transverse
displacement field u ,
ij = Dij u
(2)
Finally, the constitutive equations may be written as
ij = H ijkl kl
(3)
where H represents the plate Hooke matrix.

The following boundary conditions will be assumed, as they are sufficiently representative,
though not the most general ones,
on 1 , u = u , u n = u n
on 2 , n = n , K n = K n , Z i = Z i
(4)
In these expressions, u is the deflection, un is its normal derivative, n is the normal moment,
Kn is the Kirchhoff load, and the Zi are the corner loads. Defining for n, Kn and Zi, reader can
refer to Reference [5].
The purpose of problems in strong form is to find exact solutions satisfying all the
equations from (1) to (4). As such a solution can only be obtained for academic problems, the
practical way is to find an approximate solution of the weak problem as expressed by a
variational principle. In the displacement formulation, it is the principle of minimal total
195
energy. In the equilibrium formulation, it is the principle of minimal total complementary

energy. Let us introduce the function spaces of kinematically and statically admissible fields.
Let two spaces of kinematically admissible displacements, denoted by V and V0 ,
respectively, be defined by
V = {u H 2 (), u = u , u n = u n on 1 }
V0 = u H 2 (), u = u n = 0 on 1
(5)
(6)
Here, H 2 () denotes the Sobolev space of order 2 [8]. Obviously, V0 contains all differences
between two elements of V, that is to say, it is the linear space of admissible displacement
variations. These spaces lead to a bounded energy,
H ijkl ij (u ) kl (u )d <
(7)
From condition (7), both V and V0 may be equipped by the energetic norm
u
( H
ijkl
ij (u ) kl (u )d
1/ 2
(8)
where H is a bounded uniformly positive definite matrix.

Similarly, the two spaces of statically admissible stress fields, E , and statically admissible
stress variations, E 0 , are defined by
{
= {
E = ij L2 (), i, j = 1,2 : Dij ij + p = 0, K n = K n , n = n , Z i = Z i on 2

E0
ij
L2 (), i, j = 1,2 : Dij ij = 0, K n = n = Z i = 0 on 2
(9)
(10)
The complementary energy of their elements is bounded,
1
H ijkl
ij kl <
(11)
Both E and E 0 may be equipped by the energetic norm
E =
1
H ijkl
ij kl d
1/ 2
(12)
which is physically equivalent to norm (7) when ij = H ijkl kl .
Dual analysis
3.1 The displacement approach

The displacement approach consists in finding a displacement field u V for which stresses
are in equilibrium. The weak form of this condition is
nC
H ijkl ij (u ) kl (v)d = pvd + ( K n v n v n )d + Z i vi , v V0 (13)
i =1
where Z i , i = 1,..., nC are corner loads.

We here recognize a variational problem of the classical form,
196
Find u V such that

aV (u , v) = f V (v), v V0
(14)
where
aV (u , v) = H ijkl ij (u ) kl (v)d ,
nC
f V (v) = pvd + ( K n v n v n )d + Z i vi , v V0
i =1
Problem (14) has a unique solution, from a classical inequality of Sobolev spaces. It may
also be presented as the solution of the following minimization problem: Find u V such
that
(u ) = inf (v), v V
(15)
where
(u ) =
1
aV (v, v) f V (v)
2
Functional is called the total potential energy.
3.2 The equilibrium approach

Here, equilibrium is supposed to hold a priori. The equilibrium method is to find an
equilibrated stress field that verifies the so-called compatibility condition [9], which in weak
form writes as
nC
1
H ijkl
ij kl d = ( K n ( )u nun )d + Z i ( )ui , E0
1
(16)
i =1
This leads to the following variational problem,

Find E such that
aE ( , ) = f E ( ), E0
(17)
where
nC
1
a ( , ) = H ijkl
ij kl d , f E ( ) = ( K n ( )u nun )d + Z i ( )ui
1
i =1
The solution of this variational problem exists and is unique. It is equivalent to solve the
following minimization problem: Find E that minimizes the total complementary energy
( ) = inf ( ), E
(18)
where
( ) =
1
aE ( , ) f E ( )
2
This principle is the basis of the equilibrium approach and solving it leads to compatibility
equations.
197
3.3 The general dual analysis

Let u h Vh V be the discrete solution of (15), that is the element that minimizes the total
energy in some finite element subspace Vh of V. It is easy to prove [7] that the energetic norm
of the error may be calculated by
2
= u uh
= 2[ (u h ) (u )]
2
V
(19)
Similarly, let h Eh E be a strictly admissible approximate stress field (discrete solution

of (18)). The energetic norm of the stress error may be reckoned as
2
E
= h
= 2 [ ( h ) ( ) ]
2
E
(20)
From some elementary properties of the exact solution [5], the following relationship
between (u ) and ( ) may be obtained, which is
(u ) + ( ) = 0
(21)
Adding this result to relations (19) and (20) directly leads to the fundamental result of the
general dual analysis concerning the errors,
u
2
V
= 2 [ (uh ) + ( h ) ]
(22)
This error measure only requires very simple computations from the results. Moreover, if
one considers the generalized total complementary energy defined as being the total
complementary energy in equilibrium models, and minus the total energy for displacement
models, one finds that displacement models converge to the exact solution from below, while
equilibrium models converge to the exact solution from above [5]. The distance between the
two curves measures the global added error of both models, see Figure 1.
Practically, it is preferable to work with the square root of (22) and to compare it with an
evaluation of the energetic norm of the true solution, namely
[ aV (u, u )]
1/ 2
= [ aE ( , ) ]
1/ 2
1/ 2
1
1
aV (uh , uh ) + aE ( h , h )
2
2
(23)
from which follows a useful relative error measure

1/ 2
u 2 + 2
V
E
RE =
2
2
u V + E
1/ 2
(uh ) + ( h )
U (uh ) + V ( h )
(24)
where
1
1
aV (u h , u h ) , V ( h ) = aE ( h , h ) is the strain energy and the complementary
2
2
energy, respectively.
U (u h ) =
This relative error measure only requires very simple computations from the results. One
may naturally object that two finite element models are necessary to obtain such an error
measure. But the present proof never used the assumption that uh and h should be RayleighRitz approximations. The only requirement is that uh and h have to be admissible fields. As
198
an example, after a displacement finite element analysis, one may construct a statically
admissible h field, inspired from the displacement analysis, and use the above results. This
way is the Ladevze method [10]. The symmetrical construction, which could be named dual
Ladevze method, relates to the compatibility error [6]. As a conclusion, Ladevzes
approach may be considered as a special form of the general dual analysis.
In the particular case of structures where
-
one type of boundary conditions is homogeneous,

The approximate homogeneous fields are obtained by a Rayleigh-Ritz procedure,
one can prove the existence of upper and lower bounds of the exact energy. This is the
classical dual analysis as proposed by Fraeijs de Veubeke. In this case, the cumulated error is
thus measured by the difference between the strain and complementary energies.
Choosing finite elements for plate bending
Here, we would like to choose appropriate finite elements for the displacement approach and
the equilibrium approach.
In the displacement model, continuity of deflection at the interfaces is the first requirement
to satisfy. The next requirement is the continuity of normal slope between adjacent elements.
Thus, C1 - continuity requirement has to be obtained. Another reason for trying to a rigorous
enforcement of normal slope continuity is to guarantee that the direct influence coefficients
are actually lower bounds to the true ones in case of the homogeneous displacement
boundary. From the above strict conditions, the conforming elements of assembled triangle
(HCT) or assembled quadrilateral (CQ) [5] should be chosen. Moreover, in reformulating
these elements, the advantage of using the area coordinates is useful when assembling and
calculating the fields [13]. The reliability of this way has been well tested in our package.
Figure 1: Convergence and error of both types of approximation

In the equilibrium approach, a triangular equilibrium plate element with degree zero was
first introduced by L.S.D. Morley [3]. A regular family of equilibrium triangles and
199
rectangles was still developed into a high level [4] but will not be considered here. A
drawback of the constant moment field is the impossibility of obtaining exact equilibrium in
the presence of a constant pressure, which is a severe limitation. In order to solve this
problem, it is necessary to add a special mode in which the pressure is equilibrated by corner
loads only. For this purpose, a particular system of axes will be chosen as follows. Let us call
1, 2, 3 the nodes, as taken in the counter clockwise sense. Let side 1-2 be the X axis. Y axis is
perpendicular to it, passed through node 1 and orientated in such a manner that Y3 is positive.
Let c1(X,Y), c2(X,Y) and c3(X,Y) be the three area coordinates and ci = 0 be the equation of
the side which is opposite to node i. The new moment field can be expressed as follows
= +
(25)
where = [ ij ], i, j = 1, 2,3 is a constant matrix, = [ 1
3 3 ]
are unknowns,
amplitude refers to pressure, and A is the area of the triangle.

The following special mode T has been obtained by the first author
X X2
X (X X 2 )
1 X
1
T11 = 3 c1 + 3
c2 3 3
( X 2 X 22 c 2 X 32 c3 )
c3 +
Y3
3 Y3
2A
X 2Y3
1 Y
1
(Y 2 Y32 c3 )
T22 = 3 c3 +
3 X2
2A
(26)
2X 3 X 2
1
1
T12 = c1 + c 2
c3 + ( XY X 3Y3 c3 )
6
X2
A
The complementary field in (26) has been added to the basic field (constant field) in order to
give the general field. To specify that our element contains this special mode, we will call it
enhanced Morley (EM) element. Results from this new element were reported in earlier
papers [9, 12].
Numerical results
5.1 A clamped square plate

Consider a square plate, edge length L =10m, thickness t = 0.1m, Youngs modulus E =
2.05x1011Pa, Poissons ratio = 0.3. It is clamped on all edges with a uniform distributed
load, p = -1000Pa. Only a quarter (upper-right) of the plate is modelled due to the symmetry
of the geometry and the boundary condition, see Figure 2. An initial coarse mesh is created
with 8 triangles. Meshes of M x M elements over one quarter are uniformly refined, with M =
4, 8, 16 and 32. The exact strain energy is Uex = 10.363879 [4].
In the case of a clamped plate, one type of boundary conditions is homogeneous (Fraeijs
de Veubekes particular case). Therefore, the relative error can be rewritten as
1/ 2
V ( h ) U (uh )
RE =
V ( h ) + U (uh )
(27)
The error is thus measured by the difference between the two obtained values of the
elastic energy.
200
Figure 2: Square plate and initial mesh

Two different bending plate elements are used for the analyses. The first one is the HCT
conforming triangle which has reformulated in [13] with three degrees of freedom (D.O.F.)
per node. The other one is our enhanced Morley equilibrium triangle with 1 D.O.F. per node
and 1 D.O.F. per edge.
Table 1: The results on relative error of conventional dual analysis
2x2
S.A.Model
(EM element)
V ( h )
25.53458
K.A. Model
(HCT element)
U (u h )
8.19668
4x4
14.99747
9.77068
45.94
8x8
16x16
11.47289
10.55812
10.23024
10.33456
23.93
10.34
32x32
10.36521
10.35726
1.96
Mesh
R.E (%)
71.69
Herein, both approaches converge when the mesh is refined. The distance between the two
curves is a measure of convergence. Based on the global error (R.E), this method leads to a
reliable estimation. The convergence behaviour of the strain energy in terms of the number of
elements is illustrated in Figure 3.
5.2 L-shaped plate

Consider the L shaped plate with a uniform pressure and clamped on a part of its boundary,
cf. Figure 4a. Data of problem: edge length L = 5m, the quantitative remainders are the same
of the first problem. The meshes will generally be composed of 3 node or 6 node triangles
with two different levels of refinement. Figure 4b is an example of a uniform mesh.
201
Figure 3: Convergence curve for classically dual analysis of the square plate
(b)
(a)
Figure 4: L shaped plate (a) and mesh of 1024 elements (b)

Table 2: The results on relative error of classically dual analysis
Finite element model
EM element
Mesh
D.O.F
1
2
3
4
5
36
136
528
2080
8256
V ( h )
3.359811
2.126611
1.981689
1.865485
1.820613
HCT element
D.O.F
30
108
408
1584
6240
U (u h )
1.372400
1.618722
1.711413
1.751100
1.769728
Relative
error
R.E (%)
64.81
36.82
27.05
17.78
11.91
202
Figure 5 plots the convergence of the energy of L shaped plate in dual analysis.
Figure 5: convergence curve for L shaped plate

The convergence of lower bound and upper bound of strain energy in this problem is slow.
This fact is not very surprising, as the problem is singular, due to the re-entrant angle. In such
cases, an adaptive mesh refinement procedure would lead to better results.
b)
a)
c)
Figure 6: Square plate, boundary conditions and mesh forms
203
5.3 A square plate with load at the centre

Consider a square plate loaded at the centre, F = 1000 N, edge length L =10m, thickness t =
0.1m, Youngs modulus E = 2.05x1011Pa, Poissons ratio = 0.3. It is clamped on one edge,
the opposed edge being loaded with a prescribed non-zero transversal displacement
u = 0.01 m (Figure 6a). The finite element meshes are illustrated with triangular elements
(Figure 6b) and quadrilateral elements (Figure 6c).
In this problem, the displacement element is the HCT conforming triangle as reformulated
in [13] with three degrees of freedom (D.O.F.) per node. The equilibrium is the classical
Morley equilibrium triangle with 1 D.O.F. per node and 1 D.O.F. per edge [3].
The calculated results from HCT element and Morley element are summarized in the
following tables
Table 3: Convergence of the energies
Finite element model
K.A. Model (HCT element)
Nel
D.O.F
8
32
128
512
15
55
207
799
U (u h )
28.630708
27.797573
27.533898
27.440801
S.A.Model (Morley element)
f V (u h )
3.140623
3.199027
3.214918
3.220142
D.O.F
17
67
263
1039
V ( h )
21.124286
25.3217398
26.783227
27.218831
f E ( h )
38.422033
47.223287
50.285300
51.198337
Table 4: Convergence of the errors in dual analysis

Nel
2 [ (uh ) + ( h ) ]
8
32
128
16.384676
5.393998
1.633815
49.754994
53.119313
54.659632
57.39
31.87
17.34
512
0.482306
54.317125
9.39
U (uh ) + V ( h )
Relative error (%)
Results of tables 3 and 4 provide a global view based on the total energy and the upper bound
of global error estimation. The relative error corresponding to the final mesh is still a large
value (9.39%) [7]. Therefore, an improved solution is necessary. For this reason, we apply
the CQ element to the displacement model.
Table 5: Convergence of the errors in dual analysis is improved significantly when using
the conforming quadrilateral (CQ) element
D.O.F
15
55
207
U (u h )
27.677173
27.495156
27.433853
f V (u h )
3.197004
3.214167
3.219642
799
27.408949
3.221671
Relative error (%)

54.25
30.02
16.21
8.72
204
The relative error corresponding to the final mesh of using CQ element gets a better result
than the previous estimation (8.72%). The convergence behaviour of total energy in dual
analysis is illustrated in Figure 7.
Conclusions
The application of dual analysis to plates is shown to work as an efficient error measure. Its
classical form involving strain energy comparison is limited to homogeneous boundary
conditions. But its more evolved version based on the total complementary energy works in
any case.
Figure 7: Convergence curve for the generally dual analysis

The classical Morley element has been completed in order to make it able to exactly
equilibrate a pressure, a fact that allows us to treat more realistic problems. This enhanced
element is found to work fairly.
It has to be mentioned that, in the frame of dual analysis, there is no need to know the
exact solution, as it is always comprised between two the convergence curves. So, the
Richardsons extrapolation is not necessary in the error evaluation.
Our results only derive from an evolution of the energy bounds as computed by a uniform
refinement of the mesh. Singular problems are more difficult to treat with uniform meshes
because the convergence is slow. Thereby, a sound knowledge is necessary to generate finite
element meshes based on cost-effective and accurate solutions. We thus should combine our
method with an adaptive local refinement procedure in order to improve the cost
effectiveness of the error bound evaluation. The result of this investigation will be shown in a
further paper.
References
[1]
B.M. Fraeijs de Veubeke, Displacement and equilibrium models in the finite element method, in
"Stress analysis", edit. Zienkiewicz O.C., Holister G.S., Wiley, London, 1965.
205
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
B.M. Fraeijs de Veubeke, A conforming finite element for plate bending, Int.J.Solids Structures, 4,
pp.95 to 108. Pergamon Press. Printed in Great Britain, 1968.
L.S.D. Morley, The triangular equilibrium element in the solution of plate bending problems, Aero
Quart., 19: 149-168, 1968.
G. Sander, Applications de la methode des elements finis a la flexion des plaques,
Coll.Pub.Fac.Sc.Appli. Univ.of Lige, N015, 1969.
J.F. Debongnie, Fundamentals of finite elements, Les Editions de lUniversit de Lige, Lige, 2003,
ISBN 2-93032254-3.
J.F. Debongnie and P. Beckers, On a general decomposition of error an approximate stress field in
elasticity, Comput. A.M.and Eng. Sciences, 8: 261-270, 2001.
J.F. Debongnie, H.G. Zhong and P. Beckers, Dual analysis with general boundary conditions,
Comput. Methods Appl. Mech. Engrg., 122:183-192, 1995.
J.F. Debongnie, Some aspects of the finite element errors, Univ. of Lige,Report LMF/D42, October
2001.
J.F. Debongnie, Applying pressures on plate equilibrium elements, Univ. of Lige, May 2004.
P. Ladevze and D. Leguillon, Error estimate procedure in finite element method and application,
SIAM J. Numer. Anal., 20: 485509, 1983.
A. Mashaie, E. Hughes and J. Goldak, Estimated error bounds for finite element solutions of elliptic
boundary value problems, Comput. Methods Appl.Mech.Engrg. 130, 17-31, 1996.
Nguyen-Xuan Hung and J.F. Debongnie, The equilibrium finite element model and error estimation
for plate bending. International Congress Engineering Mechanics Today 2004, Ho Chi Minh City,
Vietnam, August 16-20, 2004.
Nguyen-Xuan Hung, The conforming finite element model and error estimation for plate bending,
Proceedings of the 8th National Conference on Mechanics, Hai Phong City, Vietnam, August 27-28,
2004.
206
Equilibrium Model in the Element Free Galerkin Method

Bui-Quoc Tinh*, Marc Duflot+, Stphane Bordas++ and Nguyen-Dang Hung+++
*
ILSB, Vienna University of Technology, Vienna, Austria
+
CENAERO, Avenue Jean Mermor, 30 B-6041 Gosselies, Belgium
++
EPFL, Station 18 EPFL-LSC, Ecublens CH-1015, Lausanne, Switzerland
+++
Abstract
In this paper, we have applied the element free Galerkin (EFG) method to solve problems in
solid mechanics. The meshless equilibrium model is applied instead of the traditional
displacement model. The stress field is expressed by the values of Airy stress function.
Moving least square (MLS) approximation be employed for construction of the shape
function and using the Lagrange multipliers to impose the essential boundary condition.
These results are compared together with the classical displacement model and the analytical
solution
Keywords: equilibrium model, stress approach, meshless, meshfree, MLS, EFG.
Introduction
Nowadays, meshless methods are new numerical methods for computational mechanics.
Meshless methods are used to establish a system of algebraic equations for the whole
problem domain without using a predefined mesh. These methods use a set of nodes scattered
within the problem domain as well as a set of nodes scattered on the boundaries of the
domain to represent the problem domain and its boundaries. These sets of scattered nodes
have been employed in building a field variable approximation.
The EFG method, first developed by Belytschko et al [1] is now well known and has been
widely used during the last decade. The purpose of this paper applied equilibrium model in
the EFG method. Fraeijs de Veubeke introduced the concept of equilibrium model in the
framework of the FEM [9]. This concept was obviously the main guideline of the research
performed at the Aerospace Laboratory of the Lige University (LTAS) during the sixties and
seventies. Details can be found in [4]. The analysis in this paper is based on the more recent
and more general formulation of Debongnie et al. [5]. Together the main ideas of this paper
are based on the ideas of Duflot and Nguyen-Dang as in [6], [7] and [8].
The outline of this paper goes as follows. In Section 2, MLS approximations to construct
the shape functions are presented. In Section 3, we established meshless equilibrium model
with using the Lagrange multipliers. Two numerical examples illustrate the proposed
equilibrium model in Section 4. Finally, in Section 5, we discuss the advantages and
drawbacks of equilibrium model over traditional displacement model and future works.
207
Moving least square (MLS) approximation
This method employs MLS approximants to approximate the function f (x) with f h (x) . The
approximation of f (x) at point x is denoted f h (x) below. We can see [2], [8] for details.
n
f h ( x) = I ( x) f I
(1)
I =1
where f I is the value of the field f at x I and I is the shape function of node I , given by
I (x) = c (x)p( x I ) wI (x)
(2)
where p( x) is a set of basis functions, usually the set of monomials up to a given order,
wI (x) is a weight function associated with node I and
c(x) = A 1 (x)p(x)
(3)
with
n
A(x) = w(x)p( x I )p ( x I )
(4)
I =1
A support domain I containing x I is associated with node I such that wI (x) and, as a
results, I (x) equal zero outside I . This choice is made in order to provide the
approximation with a local character and to restrict the sums in equations (1) and (4) to a few
terms. In this paper, we use the quartic spline weight function:
1 6 s 2 + 8s 3 3s 4 if | s | 1
(5)
w( s ) =
0
if | s |> 1
with s =
x xI
is a radius of influence; and dm I is a size of the support domain of the I th

dmI
node, with dm = d c ; where is a scaling coefficient and d c is a characteristic length that
relates to the nodal spacing near the point of interest. In equilibrium model, we need the
expression of the first- and second-order partial derivatives of this weight function in twodimension with respect to circle domain follows
12s + 24 s 2 12s 3 if | s | 1
w,k ( s ) =
0
if | s |> 1
12 + 48s 36 s 2 if | s | 1
w,kl ( s ) =
0
if | s |> 1
if x = x I we have w( x I ) = 1 ; wI ,k ( x I ) = 0 and wI ,kl ( x I ) =
(6)
(7)
12 kl
dm I
We also need the expression of the first- and second-order partial derivatives of this shape
function in two-dimension. The first-order derivatives of the shape function is given by
I ,k ( x) = c ,k ( x)p( x I ) wI (x) + c (x)p( x I ) wI ,k (x)
(8)
and second-order derivatives
208
I ,kl (x) = c ,kl (x)p( x I ) wI (x) + c ,k ( x)p( x I ) wI ,l (x) +

+ c,l (x)p( xI ) wI , k (x) + c (x)p( xI ) wI , kl (x)
(9)
with
c ,k (x) = ( A 1 ) ,k (x)p(x) + A 1 (x)p ,k (x)
= A 1 (x) A ,k (x) A 1 (x)p(x) + A 1 (x)p ,k (x)
c ,kl (x) = A 1 (x)[p ,kl (x) A ,k (x)c ,l (x) A ,l (x)c ,k (x) A ,kl (x)c(x)]
(10)
(11)
and
n
A ,k (x) = wI ,k (x)p( x I )p ( x I )
(12)
A ,kl (x) = wI ,kl (x)p( x I )p ( x I )
(13)
I =1
n
I =1
The derivatives of shape and weight functions can be shown the following in figure 1
Figure 1: The shape and weight functions and their derivatives

209
Meshless equilibrium model and Lagrange multipliers
For the sake of completeness, we recall the basic equations in two-dimension of solid
mechanics as follows: Lets consider a linear elastic solid that occupies a domain bounded
by subject to the body force b in , to the surface tractions t on t and with prescribed
displacements u on u with t I u = and t U u = . We have the relations between
the displacement fields u , the strain field and the stress field are
1. The compatibility relations
1
2
u i = u i on u
(15)
ij = Dijkl kl in
(16)
ij = (u i , j + u j ,i ) in
(14)
2. The constitutive relations
3. The equilibrium equations
ij , j + bi = 0 in
ij n j = t i on t
(17)
(18)
The Galerkin weak form are described in detail as in [2], [6] and [8] by using the Lagrange
multipliers. From the equilibrium equations (17), (18), we can be given follows.
D 1d T nu d T (n T t )d T nd = 0
u
(19)
where are Lagrange multiplier functions which have the physical meaning of boundary
displacements on t. Then if we consider the internal statically admissible stress field, which
is in equilibrium with zero body forces. This model is restricted to problems with a body
force deriving from a potential:
b = V
(20)
The difficulty in the equilibrium model is to satisfy equation (17) in the whole domain. To
that end, we decide to express the stress field as
= ( + V ) I T
(21)
where is the Laplacian T , I is the identity matrix and is called the Airy stress
function. For two-dimensional problem, we can re-write the equation (21) below
x , yy

(22)
y = , xx
xy , xy

The Airy stress function ( x) is approximated in the moving least-squares sense by

h ( x) given by
n
h ( x) = i ( x) i
(23)
i =1
210
where the I ( x) is the shape function (2). Since we chose a quartic spline weight function,
which possesses continuous second-order derivatives, as the weight functions, h (x) also
possesses continuous second-order derivatives and h (x) is continuous. The essential
boundary conditions in equations (18) on boundary t are enforced with the help of
Lagrange multipliers, demonstrated in [2], [6] and [8].
n
( x) = N I (r ) I with x t
(24)
the variation of Lagrange multipliers

n
( x) = N I (r ) I with x t
(25)
where n is the number of nodes for this interpolation, r is the arc-length along the essential
boundary and N I (r ) is a Lagrange interpolate. The final discrete system equations can be
obtained from the Galerkin weak form (19) as follows.
G d
=
0 q
(26)
Fij = C iT D 1C j d
(27a)
F
G
and F is the nodal flexibility matrix
d i = C iT nu d
(27b)
Gij = N IT n T C j d
(27c)
q I = N IT t d
(27d)
For two-dimensional problems, we have
i , yy
C i = i , xx
i , xy
(28a)
N
NI = I
0
(28b)
nx
n=0
n y
0
N I
ny
n x
(28c)
for the plane stress condition
211
E
D=
1 2
1
0 0
0
1
2
(28d)
with E and are Youngs modulus and Poissons ratio, respectively.

In order to obtain the integrals in equations (27a) - (27d), a cell structure which is
independent of the nodes in be used. In each cell, Gauss quadrature is used. The number of
quadrature points depends on the number of nodes in a cell. We have used nQ nQ Gauss
quadrature where
nQ = m + 2
(29)
and m is the number of nodes in a cell.

We use a quadratic basis ( p(x) = [1, x, y, x 2 , y 2 , xy ]T ) to be able to exactly represent a
constant stress field. We choose the number of points where the boundary conditions are
enforced to be equal to the number of nodes on t .
Numerical results
4.1 Cantilever beam

The cantilever beam shown in figure 2 is considered [1,2,3,10]. The beam is of length L and
height D subjected to a parabolic traction at the free end. The beam has a unit thickness and
a plane stress problem is considered. This is a benchmark problem because the analytical
solution is available. The analytical displacements are given by
2 D 2
y
(
6
3
)
+
(
2
+
)
L
x
x
P 2
D2x
+
+
+ (3L x) x 2
uy =
3
y
(
L
x
)
(
4
5
6 EI
4
ux =
Py
6 EI
(30)
(31)
and the stress components are
xx ( x, y ) =
P( L x) y
I
yy ( x, y ) = 0
(32a)
(32b)
P D2
(32c)
y 2
2I 4
The following parameters were used for the cantilever beam problem: the dimensions of
the beam model are L = 8 and D = 4 . The material properties are: Youngs modulus
E = 3.10 7 , Poissons ratio = 0.3 and the parabolic shear force P = 250 . Regular
distributions of nodes are employed.
xy ( x, y ) =
212
L
Figure 2: A cantilever beam and boundary conditions
Figure 3: Stresses ( xx ), ( xy ) of the beam along x = L using 45 nodes

2
Figure 4: Stresses ( xx ), ( xy ) and ( yy ) along x = L using 1225 nodes

2
213
In figure 3, we can see that the displacement model seems better than equilibrium model
in the coarsest set of nodes (45 nodes). But figure 4, when we increase the number of nodes
(1225 nodes), the comparison between the stresses ( xx ), ( xy ) and ( yy ) at the section
x=L
calculated analytically, displacement and equilibrium models. The results are very
2
good agreement with analytically as well as displacement model is given by Belytschko et at
[1].
Solution with different support sizes
Figure 5: Stresses ( xx ), ( xy ) along x = L with different support sizes

2
The radius of support domain affects the solution of problem, so we studied the support
domain according to each scaling coefficients . The solutions of total complementary
energy with different support sizes are obtained in table 1, figure 6 and stresses shown in
figure 5. The results are in close agreement with the exact result 0.038291666 as in [2]. The
equilibrium model needs more nodes inside the support domain. These results were quite
agreement compare with analytical solution. The values of the scaling parameter from 2.5
to 4.0 are given good results and the support size cannot over the value above.
Number
of nodes
17 x 9
33 x 17
49 x 25
65 x 33
Analytical
solution
0.03829166
0.03829166
0.03829166
0.03829166
Complementary strain energies

= 2 .5
= 2 .8
= 3 .2
0.041042444
0.040817679
0.040719856
0.040195350
0.040142104
0.040120325
0.039948811
0.039924601
0.039924321
0.039928110
0.039925817
0.039917563
Table 1: Results of the clamped beam problem with different support sizes
Figure 6: Convergence rate of the total complementary energy

214
4.2 Clamped beam with uniform force

The final example is the cantilever beam which clamped on one edge, the opposed edge being
loaded with a constant shearing P = 93.75 , as shown in figure 7. There is no body force and
plane stress conditions are considered. The numerical values are L = 8 , D = 4 and the
parameters of material: Youngs modulus is E = 3.10 7 , Poissons ratio is = 0.3 . The nodes
of the meshless approximation are regular spaced
y
x
L
Figure 7: The clamped beam submitted to a uniform force
In this example, the exact solution is not available. The estimated exact energy is
0.088851269 as in [2]. In figures 8 is also illustrates the distribution of the normal and shear
stresses on the cross section at x = L 2 of the beam. Both the displacement and equilibrium
models solution are plotted together for comparison. Very good agreement is observed
between the stresses calculated by the two models.
Figure 8: Normal and shear stresses at middle beam

The solution with different support sizes are obtained in table 2 and showed in figure 9 for
this problem. The same of the example above, these results were quite agreement compare
with the estimated energy solution
Number of
nodes
17 x 9
33 x 17
49 x 25
65 x 33
= 2 .5
0.092929131
0.089969937
0.089256500
0.088994983
Complementary strain energies

= 2 .8
= 3 .2
0.0919024565
0.091589030
0.0896845090
0.089608060
0.0891669294
0.089165915
0.0889749100
0.088974910
= 3 .5
0.091068271
0.089487926
0.089201713
0.089049171
Table 2: Results of the clamped beam problem depends on each by the equilibrium model
215
Figure 9: Convergence rate of the total complementary energy
Conclusions
In this paper, we presented a meshless method to solve elastostatic problems based on an

equilibrium model. The equilibrium equations and the constitutive equations are satisfied a
priority and the approximation is realized on the compatibility equations. To satisfy the
equilibrium equations at each point of the domain, we derive the stresses from an Airy stress
function. This function is approximated by a linear combination of meshless shape functions.
It's noted that the smoothness of the meshless shape functions is an essential property that
permits to conceive this method. The drawback of this method when comparing with the
traditional displacement model is that it requires the second-order derivatives of the meshless
shape functions, and mainly that a quadratic basis must be used. However, an advantage is
that there is only one degree of freedom associated with each node while in the traditional
method, the number of degrees of freedom for each node equal to the number of space
dimensions.
Furthermore, in the equilibrium model, the values of took from 2.5 to 4.0 are gave good
results. If we take this interval outside, the results of problem will not good. This model
seems to need more nodes inside the support domain than the displacement model. This is a
drawback of the equilibrium model. Because the number of nodes inside the support domain
is larger, so it will take a lot of the time for calculation.
In this respect, the numerical examples show that the meshless equilibrium model fulfils
our objectives. The method was illustrated in two-dimensional linear elasticity but it can be
extended. Firstly, the method can be applied in non-linear elasticity. Secondly, the method
can be applied in three-dimensions, where the reduction of the number of degrees of freedom
between the equilibrium and the traditional methods is 3 instead of 2. The method in threedimensions is similar to what was done in this paper: Equation (21) is indeed still valid to
obtain an equilibrated three-dimensional stress field and Equation (23) is also an appropriate
expression for the meshless approximation of the stress function.
References
[1]
T. Belytschko, Y. Y. Lu and L. Gu, Element-free Galerkin method, International Journal for

Numerical Methods in Engineering; 37:229-256, 1994.
216
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
Bui-Quoc Tinh, Application of the element free Galerkin method for dual analysis, European Master
Thesis, 2005.
Bui-Quoc Tinh and Ngo-Thanh Phong, Application of equilibrium model in the element free Galerkin
method, Vietnam Journal of Mathematical Applications, 2005.
J. F. Debongnie and P. Beckers, Revisiting early works of the aerospace laboratory of Liege dual
analysis, 60s and 70s, Colloquium International. Nhatrang, 2000.
J. F. Debongnie, H. G. Zhong and P. Beckers, Dual analysis with general boundary conditions,
Computer Methods in Applied Mechanics and Engineering; 122:183-192, 1995.
M. Duflot and Nguyen-Dang Hung, Dual analysis by a meshless method, Communications in
Numerical Methods in Engineering 18:621631, 2002.
M. Duflot and Nguyen-Dang Hung, Global error estimation with displacement and equilibrium
meshless method, Computational mechanics WCCM VI in conjunction with APCOM04, Sept. 5-10,
Beijing, China, 2004.
M. Duflot, Application des mthods sans maillage en mechanique de la rupture, PhD thesis,
University of Liege, 2004.
B. Fraeijs De Veubeke, Displacement and equilibrium models in finite element method, In Stress
Analysis, Zienkiewicz OC, Holister G (eds). John Wiley and Sons, 1965: chapter 9, 145-197. Reprinted
in International Journal for Numerical Methods in Engineering; 52:287-342, 2001.
G. R. Liu, Mesh Free Methods-Moving beyond the Finite Element Method, CRC Press, 2003.
217
Active Balancing of Monorotor during Operation

*
Le-Dinh Tuan*
Department of Naval Architecture and Marine Engineering,
Ho Chi Minh City University of Technology, Vietnam
Abstract
In the field of rotating machines, mass unbalance is inherent to the considerable vibrations
and noise which must be reduced. This work, which follows an overview on classical
balancing techniques, concerns the active balancing of monorotors. The unbalance is detected
and corrected in real time, during operation in steady state. The method consists of generating
a correction force using two moving weights, namely the satellites, fixed on the same plane
and moving at a constant radius of the rotation axis. The procedure of balancing is based on
the influence coefficients method that is most exploited in calibrated balancing machines.
Specific measurement and control systems are also built up for the active experiments.
A first experiment on active balancing is also presented. The test bench is based on the
hard-type balancing machine HnB100 developed and commercialized by the team is also
investigated. The experiments show that the vibration due to unbalance is significantly
decreased.
Keywords: active balancing, influence coefficients method
Introduction
1.1
Dynamic balancing
Unbalance in rotating machines is an important factor which must be taken into

consideration, especially in modern machines that require high speed and/or high reliability.
In general, the unbalance causes damage in machineries, so it must be detected and
decreased when necessary. That is dynamic balancing. The unbalance correction consists of
proposing solutions of fully or partly correction of complementary reaction forces at the
supports or bearings. This means to avoid serious damages due to fatigue in the related
structures, to reduce noise and vibration, so, to extend the life of the machine and enhances
its reliability.
1.2
Active balancing
Besides the optimization of shapes and of materials from the design, different techniques
have been developed to control the vibration behaviour of rotating machines. One of the
recently developed means is active control: starting from measurements carried out in real
time and from a controller, an active system supplies energy to the structure by the
intermediary of actuators to reduce its vibrations according to predetermined objectives [2,5].
The balancing technique using active control aims also at the direct reduction of vibration
218
sources. The technique consists in adding or removing materials so as to bring both inertial
axe and rotating axe closer. Not like in traditional way, this technique permits to realize the
whole balancing process automatically without stopping the machine and furthermore in real
time [2].
1.3
Test bench HnB100J
An initial experimental application in this research is the realization a test bench HnB100J
[2]. This test bench permits the active experiments on monorotors of up to 1.4 m in length,
900 mm in diameter, 150 kg in mass. The technology and process efficiency can be then
evaluated during the research. They are also exploited for normal balancing techniques.
The reaction on the supports due to unbalance of rotor is measured thanks to piezoelectric
sensors well integrated on the machine. They form, together with the
preamplifiers/amplifiers, ADC, phase sensors, the basis of a measurement system. This
system is only completed with the analyser ActiveHnB writen on standard C.
Balancing of rotor
2.1
Types of unbalance
Three unbalances need to be distinguished [4]:

- Static unbalance: the centre of gravity of rotor and the rotating centerline do not coincide.
The unbalance masses can be replaced by an equivalent mass m at a radius r from the rotating
centerline (Fig. 1, a). At an angular frequency , it produces a centrifugal force P = mr 2 .

(a)
(b)
Figure 1: Static and couple unbalances.
- Dynamic / couple unbalance: the unbalance masses are replaced by two equal masses in
two different radial planes at exactly opposite angular positions (Fig. 1, b). At an angular
frequency , the centrifugal forces of the two unbalances produce a couple M = mra 2 .
- Mixed unbalance: the centrifugal forces due to unbalances are equivalent to a force P and a
couple (F,F) (Fig. 2), in other words, a combination of a static unbalance and a couple
unbalance. Such an unbalance is most general.
Unbalance can be corrected by adding counterweights in certain planes. The centrifugal
forces generated by them balance ones by equivalent masses. The choice of balancing masses
and their attachment on rotor is called balancing.
219

Figure 2: Mixed unbalance.
2.2
Influence coefficients method
In rigid rotor balancing, the influence coefficients method is most adapted to in-situ balancing
or calibration-typed balancing since it uses only experimental information. The rotor is
described by the measurement planes and balancing planes (correction planes). The
sensitivity of the measurement planes with respect to the balancing planes characterizes the
responses of unbalance for a given rotating speed. This method implies the assumption of
linearity of the system. In general case, when adding a trial weight on one end of rotor also
changes considerably the vibration measured on the other end, it has to employ the
simultaneous two-plane balancing technique. The most adequate method is influence
coefficients method (ICM) [1, 4]. The balancing process requires the simultaneous vibration
measurements due to unbalances B1, B2 in two balancing planes i, ii (Fig. 3) thanks to the
sensors integrated on the V-bearing supports 1 and 2 [2].
Figure 3: Simultaneous balancing on two planes.

The vibrations are then measured after placing alternately trial weights in two balancing
planes. The assumption of linearity of the system yields:
V10 11 12 B1
=

V20 21 22 B2
(1)
where Vkl denotes the vibration measured at the bearing k at the run l, the value l = 0
corresponds with a run of rotor at initial configuration. The influence complex coefficients
ik are the functions of the rotating speed. One carries out three successive runs:
(1) Rotor is in initial configuration with the unbalances B1 and B2 to determine in the
planes i and ii. One measures the vibrations V10 and V20 at the bearings 1 and 2.
(2) One installs then a trial weight M1 on the plane i, and measure the vibrations V11 and
V21.
220
(3) Remove M1, one finally installs a trial weight M2 on the plane ii, and measure V12,
V22.
Taking into account the linearity of the system, and on the condition of carrying out the three
runs at the same rotating speed, one can also write:
and
V11 11 12 B1 + M 1
V21 21 22 B2
(2)
V12 11 12 B1
V22 21 22 B2 + M 2
(3)
(1), (2) and (3) permit to compute ij and so B1 and B2. If one calculates Vkl - Vk0 due to only
trial weights, one obtains successively:
kl = (Vkl Vk 0 ) / M l
k , l = 1..2
(4)
and one deduces from them the unbalances:

V
B1
= [ ]1 10
B2
V20
(5)
The application of ICM remains simple but comprises a disadvantage because of the number
of starting-stops necessary to the characterization of the machine [2]. The technique amounts
to the determination of correction weights which are manually attached to the rotor when
stopped. The correction is thus fixed and might become ineffective in the case of
evolutionary unbalances. It has therefore become essential to think of an ability to correct the
unbalances during operation. The rotating machines in exploitation can present a degradation
of the vibratory levels caused by the modification of the state of balance of the machines due
to the wear, the deformation, the sliding of the assembled elements,...Regular processes of
balancing are thus programmed and require a complete stopping of the whole machines.
Within this framework, active balancing would make it possible to correct the unbalances at
any time during operation and to thus avoid the stopping of the machines for the usual
interventions of maintenance.
Active balancing
3.1
Principle
Active balancing consists of generation and control of unbalance forces at certain positions
on rotating shaft. In the principle of operation, two mobiles of the same mass, called
satellites, are compelled to move in the same plane at a constant distance from the rotor axis
[2]. They thus generate an unbalance force of variable amplitude and phase (Fig. 4).
221
B1
B2
m
m
B2
B1
B1+ B2 = 0
B = B 1+ B 2
Figure 4: Generation of an unbalance from two masses.
3.2
Optimal movement
3.2.1 Correction positions

Following the application of the influence coefficients method, the system of active balancing
must generate an unbalance Bc. If B1, B2 are unbalances generated by the satellites 1 and 2,
one has:
Bc ei = B1ei1 + B2ei 2
(6)
Both satellites have an identical mass ms and are located on the same radius r. The amplitudes
of the unbalances generated by the satellites are thus identical:
B1 = B2 = ms r = b
(7)
Given 0<< /2, one has:
1, 2 = m
(8)
m
B2
B1
Bc
Figure 5: Positions of correction.

The expression of the angle permits to position the satellites according to the generated
unbalance:
= arccos(Bc / 2b)
(9)
The unbalance Bc is calculated by ICM.
222
3.2.2
Criterion of displacement of the satellites
It has to determine the mode of displacement of the satellites from their initial positions
towards the positions of correction so as to ensure a fast correction, to minimize the
perturbations during this transient stage [2].
Several strategies of piloting of the satellites are possible (Fig. 10). One possibility consists of
modification of the phase of the satellites by maintaining them in phase opposition, then
adjusting the amplitude of the resulting unbalance. A second possibility is to modify
simultaneously the phase and the amplitude of the unbalance. The second solution is adapted
to a fast correction. To minimize the perturbations sustained by the rotors means the
diminution of the vibratory amplitudes right from the starting of the satellites. Taking into
account the assumption of linearity of the system, it guarantees thus the diminution of the
vibration.
Bfinal
S2
S2
S1
B initial
S1
m
S2
+
S1
B initial
S2
S1
B final
Figure 6: Configurations of displacement.
3.3
Measurement modules
The determination of the unbalance correction by ICM requires the precise knowledge of the
rotating speed and vibratory state of the structure.
3.3.1
Positions of the satellites and rotating speed
The angular positions of the satellites are determined in comparison with a reference (Fig.7).
The reference phase is obtained from the real-time measurements.
Figure 7: Determination of location of the satellites and the reference phase.
223
The principle of determination of the satellite phases and the rotating speed is shown in Fig.
8. The measurement of m, 1, 2 gives:
= 60 / m
( RPM )
1, 2
1, 2 = 360
(10)
(11)
(o )
3.3.2 Treatment of the vibration signals

The parameters of signal treatment are inherent to the rotating speed . It is therefore
necessary to build an specific signal processing module to determine the vibration amplitudes
and phases at any [3]. The principle is to acquire whole periods synchronized compared to
the reference. The signal is discretely sampled to obtain a fixed number of acquisition points
per period. The determination of vibration amplitude, reference phase, satellite phase,
rotating speed is all based on this information. The synchronization of sampling with the
reference and with a whole number of rotations, the measurement of vibration are
continuously carried out for each analysis loop.
The DFT (Discrete Fourier Tranform [4]) in real time permits to compute vibration
amplitudes and phases at a predetermined speed. It acts as a numeric filter of the signals
coming from the vibration sensors.
Stage I: Synchronous reference phasor
reference signal
Reference
Reference
time
S1 S2
S1 S2
satellite signals
time
Stage II: Decoding of signals
Satellite S1
time
Satellite S2
time
Satellite S1
/ Reference
Satellite S2
/ Reference
Stage III: Timing
time
2
time
m
Period of rotation
of Monorotor
time
Figure 8: Determination of the satellite phases and the period of rotation.
224
3.4
Organization of the balancing phases
As an balancing technique, the active balancing enhance the vibration behaviour but, more
active, while a machine is in operation regardless of unknown and varying residual
unbalances on the rotor [2].
3.4.1 Correction strategy
The solution is to measure continuously the vibrations, and then by a comparison between the
influence-coefficient-based computed unbalances and permissible residual ones we decide
whether to move satellites or not [2].
3.4.2
Two main phases
The influence coefficients method requires a phase of characterisation of the machine, i.e.
calibrating phase. This phase is realized before the unbalance correction prepared by the
satellites in the next phase, exploitation phase, and at active mode.
3.4.3
Active system
The synoptic scheme of active balancing is given in Fig. 9. A monorotor (7) is installed on 2
vibratory supports on which integrated 2 vibration sensors (9). On monorotor there are 4
satellites (6) positioned thanks to satellite phasors (8). An electrical motor (1) rotates the
monorotor through the intermediary of a belt (2) and a universal joint (5). The reference
phase of monorotor is determined by a fixed reference disk (3) and phase sensor (4). The
vibration signals from the supports, the reference phase, satellite phases are analysed in real
time by the analyser ActiveHnB through an data acquisition card (11). The speed of
monorotor can be controlled through the intermediary of an inverter (13) while the satellites
can be driven thanks to driver (10).
Figure 9: Structure of the active balancing.
Experimental validation
In order to verify the capacities of the balancing system, many balancing tests were realized.
The mechanical designs for large-sized rotors, the measurement and control systems, the
225
active balancing technique and the process efficiencies are important to optimize the existing
technology. The limits of the system is then remarked [2].
The computation of influence coefficients ij is carried out at the calibrating phase and stored
in a look-up table [4].
The verification of ICM is also realized with different rotors and afterwards with the
monorotor on the test bench HnB100J. The measured vibration amplitudes before and after
active balancing are given in Fig. 10, 11, 12, 13 where monorotor is at 500 rpm.
The residual unbalances are all smaller than permissible ones (ISO 1940, at grade G 6.3 [4]).
In fact, the equivalent mass generated from a couple of satellites shows that the residual
unbalance 12 g 65 mm is easily obtained.
4
1.5
x 10
channel 1
channel 1 (filtered)
channel 2
Voltage *5/216 [V]
0.5
0.5
1.5
50
100
150
200
Time [ms]
250
300
350
400
Figure 10: Vibrations before active balancing.

4
1.5
x 10
channel 1
channel 1 (after active balancing)
Voltage *5/216 [V]
0.5
0.5
1.5
50
100
150
200
Time [ms]
250
300
350
400
Figure 11: Vibration at the support 1 before and after active balancing.
226
x 10
1.5
channel 2
Voltage *5/216 [V]
0.5
0.5
1.5
50
100
150
200
Time [ms]
250
300
350
400
Figure 12: Initial vibration at the support 2.

The results also show the effectiveness of specific measurement system, signal treatment,
and active balancing technique.
4
1.5
x 10
channel 2
channel 2 (after active balancing)
Voltage *5/216 [V]
0.5
0.5
1.5
50
100
150
200
Time [ms]
250
300
350
400
Figure 13: Vibration at the support 2 before and after active balancing.
Conclusion
This paper shows an active balancing technique applying to a monorotor as an initial

experiment It is quite evident that the vibration due to unbalance can be reduced significantly.
The limit of the system developed in this research is the exploitation of the influence
coefficients method which imposes a characterization of the machine in steady operation and
an intervention at predetermined speeds. It seems interesting to determine a method of
characterization of the influence coefficients on any range of speed.
Though the concept of active balancing developed in this study was applied well on a test
bench and with a monorotor, it can be extended to other machines by means of an appropriate
adaptation. Moreover, the active balancing devices can be integrated in the certain machines
at the design stage.
227
References
[1]
[2]
[3]
[4]
[5]
Dyer S. W. and Ni J., Adaptive Influence-Coefficient Control of Single-Plane Active Balancing

Systems, Manufacturing Science and Engineering, ASME-IMECE 1999, MED-10, pp.747-755, 1999.
Le-Dinh Tuan, Experiments in active balancing of monorotor during operation, VNUHCM-HCMC,
project B2004-20-10, 2004.
Le-Dinh Tuan, Dynamics of mechanical structures, EMMC Center- HCMUT, 2004.
Le-Dinh Tuan, Manual of the balancing machine HnB100, HCMC University of Technology, 2003
(in Vietnamese).
Zhou S. and Shi J., Active Balancing and Vibration Control of Rotating Machinery: A Survey, Shock
Vib. Dig., 33(5), pp. 361-371, 2001.
228
A Refinement Criterion with p-Version in the Error Estimate

and the Convergence Rate of the Finite Element Analysis
*
Nguyen-Hoai Son*
University of Technical Education Ho Chi Minh City, Vietnam
Abstract
The goal of this study is to further investigate and to develop a more efficient way in the
error estimate and the rate of the convergence for the adaptive mesh p-refinement procedure
in the finite element analysis for two-dimensional and three-dimensional elastostatic
mechanics problems. The oscillation of the stress field around singularity points is also
considered in the refinement process. In this paper, an adaptive p-refinement procedure with
a uniform p analysis capability is called a fully p-refinement procedure. We also establish the
refinement criteria for the adaptive strategy with p-version based on the ratio of error
indicator or the enrichment indicator.
Keywords: estimate, refinement, error indicator, adaptive.
Introduction
In this paper, an adaptive p-refinement procedure with a uniform p analysis capability is

called a fully p-refinement procedure. We have established a criteria refinement procedure in
the error estimate and the rate of convergence based on the ratio of error indicator. Several
error estimates have been developed for engineering use. However, they are computationally
intensive and still need to be validated for accuracy in practical engineering analysis [1],[3].
Besides, the mentioned error estimates are in the form of an energy norm. This quantity
differs from the quantities of interest in solid mechanics analysis, which are usually
displacements or stresses at particular points inside a problem domain. Also, the energy
convergence may not guarantee stress convergence. The development of an accurate error
estimate for stress will result in a more efficient way of controlling the refinement process
[2],[4] and will provide valuable information for checking the convergence of final stress
results.
2
The energy functional and the error equation in the energy
norm
Find u V such that the Dirichlets boundary condition satisfies the following equations:
v V
B (u , v) = L(v)
1
J (u ) = B(u, u ) L(u )
2
B (e, v) = B (u EX , v) B (u FE , v) = L(v) B (u FE , v)
(1)
(2)
229
e = u EX u FE
where
J (u ), e
E ( )
E ()
= u EX u FE
(3)
E ()
, u EX , u FE : the energy norm, the error energy norm, the exact energy and the
approximation element finite energy, respectively.
The criterion of the convergence

- The algebraic rate of convergence:
e
E ()
= u EX u FE
E ()
k
N
(4)
k
exp(N )
(5)
- The exponential rate of convergence:

e
where
E ()
= u EX u FE
E ()
k , , , , N : the positive constants and the degree of freedom, respectively.
A strategy of the enrichment indicator
4.1
p-refinement
In order to select h-refinement and/or p-refinement, we establish a ratio between two p and
p-1 level by enrichment indicator for j element as follows:
j(pj)
C =
, if j ( p j 1 ) 0
(6)
j ( p j 1 )
where p j , j : the order of the basis polynomial and the error indicator of the jth element,
respectively.
When p j > 1, j ( p j 1) can be computed by considering the part of U (energy
norm) involving the shape functions of degree equal to p j , if p j 1 , the expression (6) can
not be computed.
* if C < 1 the strategy refined on the jth element will be the p-refinement
* if C 1 the strategy refined on the jth element will be the h-refinement, because the
error indicator j ( p j ) > j ( p j 1 ) ,i.e. the error estimate increase when the order polynomial
basis hierarchic increase, this is not effective.
230
Table 1: a new enrichment indicator in computing of the error criteria for the L-shape 2-D by
(**) and (*) [5], [6].
p
C =
p
p
ees
ees
j 1
p j 1
pj
U
U
pj
(e
(e
p j 1
ex
+U
2
ex
+U
)
)
1/ 2
p j 1
1/ 2
2
pj
(**)
0.91904844443771
0.83452098125554
0.77782631745437
0.62711167259153
0.45999378454578
0.38613868694310
0.26321887715915
1
2
3
4
5
6
7
es
(*)
hC l
0.82173888888889
0.59132222222222
0.50557222222222
0.38033333333333
0.31075000000000
0.26523333333333
0.24467222222222
0 < C = 0.25 < 1, l 3% C > 0.25 ( p refinement is necessary)
where ees , U , eex , es , hC , l : the norm of the estimate error, the norm of the strain energy,
the norm of the exact error, the estimate error indicator, the length of the longest edge of jth
element and the previous tolerance, respectively.
4.2
h-refinement
e R max j
(7)
j =1,..., N
By the other way in the computation for the enrichment indicator, we have
es
max
p
U h 2 + ees
hnew
i
R ees l
N
h
1/ 2
= eN
(8)
e
R Ni
ees
0 < R 1, l = 4%, R = 0.82
R < 0.82, hnew > 0.67 (h-refinement is necessary)
Table 2: a new size of the element and the enrichment indicator in computing of the error
criteria for the L-shape 2-D by (8) and (7) [5], [9].
#Element
hnew
75
192
300
675
1200
1875
3352
5.0000
3.1250
2.5000
1.6667
1.2500
1.0000
0.8929
0.21389655791820
0.29742892449463
0.34221505046118
0.44306729695172
0.33296603779371
0.61536514766244
0.66194147175037
0.24213368200584
0.34700835422613
0.40865949469253
0.55090348171000
0.55140470348212
0.80372930397042
0.82500021777001
231
Application: L-shape plate problem
Let us, consider an unstressed square hole in an infinite plate subjected to unidirectional
tension as show in figure 1. According to the symmetry of problem, the finite element
analyses are done in the case of plane stress on a quarter of the plate with a = 50, = 1
and the material characteristics are E = 10 5 and = 0.3 . The exact strain energy is given by
2
[7],[8]. u E ( ) = 0.311332
Figure 1: L-shape plate.
The results of error estimation.
Table 3: Error estimation and effective index results for h and p refinement of quadrilateral
element.
- p uniform mesh (rc =0.38), p=1,2,,8
#dof
192
450
682
1472
2562
3952
4930
Uh
0.302606
0.307048
0.308255
0.309621
0.310189
0.310491
0.310610
e es
es
e ex
ex
Time(s)
0.093037
0.064915
0.054836
0.040509
0.032756
0.027766
0.025534
0.147913
0.106438
0.091003
0.068460
0.055935
0.047742
0.044041
0.093413
0.065452
0.055470
0.041364
0.033808
0.029000
0.026870
0.169812
0.118119
0.099909
0.074337
0.060702
0.052044
0.048212
0.7164
0.5352
0.6740
0.5883
0.5305
0.3928
0.5880
1.531
2.273
4.531
9.654
19.564
39.523
75.895
- p graded mesh (rc =0.53), p=1,2,,8

#dof
532
872
1362
2002
2792
3732
4822
Uh
0.310675
0.310822
0.311001
0.311113
0.311178
0.311218
0.311245
e es
es
e ex
ex
Time(s)
0.025612
0.022561
0.018165
0.014764
0.012369
0.010630
0.009273
0.045902
0.040434
0.032557
0.026461
0.022168
0.019051
0.016620
0.025632
0.022583
0.018193
0.014798
0.012409
0.010677
0.009327
0.045986
0.040506
0.032623
0.026531
0.022246
0.019139
0.016718
0.999238
0.999019
0.998488
0.997714
0.996747
0.995604
0.994236
1.438
2.297
4.578
9.781
20.781
40.687
77.032
232
- h-p graded mesh (rc=0.4), p=1,2,3,4,5, udiv=10,18 vdiv=10,..,18

#dof
192
532
1226
2578
4930
Uh
e es
es
e ex
ex
Time(s)
0.107694
0.026095
0.020784
0.010620
0.008122
0.168714
0.044968
0.036088
0.027327
0.021464
0.107582
0.025632
0.020199
0.014832
0.011180
0.196497
0.045986
0.036224
0.026592
0.020041
0.8586
0.9779
0.9962
1.0276
1.0710
0.828
1.328
4.640
23.938
121.937
0.299758
0.310675
0.310924
0.311112
0.311207
Figure 2: Graph log( e
Figure 3: Graph
es
) (left) and effective index (right) versus # DOF .
versus # DOF (left) and CPU times versus # DOF (right).
Table 4: Error estimation and effective index results for h-p refinement of triangular element.
- p non-uniform mesh (rc=0.445)
p
#dof
1
2
3
4
352
1290
2816
4930
Uh
0.299352
0.301513
0.310405
0.311213
e es
es
e ex
ex
Time(s)
0.109796
0.099469
0.031657
0.013935
0.196753
0.178248
0.056730
0.024972
0.109453
0.099090
0.030446
0.010908
0.200049
0.180459
0.054648
0.019554
0.9835
0.9877
1.0381
1.2771
3.078
11.14
48.093
161.526
233
- h-p non-uniform mesh.

p
#dof
1
2
3
4
192
962
2816
4930
Uh
0.291431
0.303209
0.310940
0.311234
e es
es
e ex
ex
Time(s)
0.140985
0.089994
0.019183
0.008602
0.214399
0.143518
0.033387
0.015210
0.141070
0.090127
0.019798
0.009899
0.261318
0.163676
0.035506
0.017744
0.999396
0.998521
0.968904
0.868966
1.703
5.578
47.953
178.765
Figure 4: Displacement and von mises stress field for p-refinement of the triangular element.
Table 5: Relative error estimation: comparison between the computation and FESTA
software with the SPR and REP Technique.
Computation
h-uniform
h-non uniform
#dof (%) #dof (%)
FESTA software [85]

Uniform SPR
Adaptive REP
Adaptive SPR
#dof (%) #dof# (%) #dof (%)
es
es
es
192
682
2562
4930
22.58
192
25.25
14.27
682
14.68
8.76
2562
8.15
6.87
4930
6.09
Computation
p- non uniform
h-p non uniform
#dof (%) #dof (%)
96
720
1920
2736
96
16.56
518
6.67
1756
4.29
3986
1.33
FESTA software [85]
Uniform SPR
Adaptive REP
#dof (%) #dof
es (%)
es
96
670
1516
2366
352
1290
2816
4930
96
720
1920
2736
96
670
1516
2366
es
es
es
19.67
17.82
5.67
2.49
es
192
962
2816
4930
21.43
14.35
1.97
0.98
12.60
5.33
3.89
3.49
12.60
5.33
3.89
3.49
96
518
1756
3986
16.56
6.67
4.29
1.33
12.60
5.27
2.83
1.38
Adaptive SPR
#dof (%)
es
12.60
5.27
2.83
1.38
234
Figure 5: Graph log( e
es
) (left) and effective index (right) versus # DOF .
Figure 6: Graph U versus # DOF (left) and CPU times versus # DOF (right).
Conclusion
By using an error tolerance max = 5% . For comparable error levels, the table 3
shows that the permissible error is achieved after 5 refinement steps, i.e. for mesh #6.
The p and h-p graded mesh is obtained desirable result only after 2 refinement steps.
The refinement is continued leading to a final error of 4% for the h-refine and 1%-2%
for p-refine.
Table 3 show that the difference between relative error estimator in energy norm and
exact relative error is small. The p and h-p refinement convergence its fast toward the
actual value of U h
up to 3 decimal digits accuracy.
The comparison between the quadrilateral mesh (table 3) and the triangular mesh
(table 4) shows that the relative error estimation of quadrilateral mesh is better than
the triangular mesh at the same degree of freedoms.
The comparison relative error estimation between the computation and FESTA
software with the SPR table 5 (superconvergent patch recovery) and REP (recovery
Equilibrium patch) Technique gives the results reliability.
Figures 2, 3, 5 and 6 shows that the rate of convergence use by energy norm as
relative error in p and h-p version is faster than in h-version. The Effective index is
oscillation around value 1.
235
From table 1, 2 and the equations (**), (8) we observe that 1
esp
esp1
, the error
estimate decreases with increasing polynomial degree. If enrichment were indicated

on element e, h-refinement would be the preferred strategy and if
esp
esp1
< 1 this
would suggest p-refinement as the preferred strategy.
The maximum von-mises stress oscillates approximate 0.4% between two hierarchic
solutions p, p+1 of an element around the singularity points.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
Szabo B.A., Mesh design for the p-version of the finite element method, Computer Methods in
Applied Mechanics and Engineering, Vol. 55, pp. 181-197, 1986.
Zienkiewicz O. C. and Zhu J. Z., Adaptive techniques in the finite element method, Communications
in Applied Numerical Methods, 4:197-204, 1998.
Cugnon, F. and Beckers, P, Error estimation for h and p methods, 8th Mechanical Engineering
Chilean Congress, Conception, 27-30 October 2004, pp.737-744.
Nguyen-Hoai Son and Mai-Duc Dai, The error estimate for finite element analysis with h-p version in
the linear elasticity 2-D, 3-D, International conference 8-2004 French-Vietnam.
Flaherty J. E., Finite element analysis, Troy, New York, 2002.
Aiffa M., Adaptive hp-refinement methods for singularly-perturbed elliptic and parabolic systems,
PhD thesis, Rnsselaer Polytechnic Institute, Troy, 1997.
Cugnon F., Automatisation des calculs lment finis dans le cadre de la methode-p, these de doctorat,
2000, ULG.
Paulino G. H. et al, A Methodology for adaptive finite element analysis towards an integrated
computational environment, Computational Mechanics 23 (1999) 361-388 @ Springer-Verlag 1999.
Adjerid, S., A posteriori Finite Element error estimation for second-order hyperbolic problems,
March 7, 2004.
236
A Hybrid Stress Approach for High - Accuracy of Stresses

and Low Sensitivity to Mesh Distortion
Dang-Dinh Thi* and Nguyen-Dang Hung+
Department of Mechanical and Aerospace Engineering,
University of Florida, Gainesville, FL 32611, USA,
+
*
Abstract
We present in this paper a special class of hybrid stress elements, namely the metis
displacement finite element (MFE). The developed theory is based on extension of the hybrid
stress model previously proposed. The formulations and applications of the metis
displacement finite element model to plane elasticity problems are examined with the aid of a
MFE program coded in FORTRAN. In order to illustrate the approach, an 8-node plane
isoparametric metis displacement finite element which is invariant and of correct rank is
developed, and it is compared with the pure displacement elements and with the recently
developed models with emphasis on the accuracy of stress computation. Considerations of
sensitivity to mesh distortion are investigated more rigorously with simple mesh distortion
measures defined in this paper. The numerical results show that the accuracy of the current
element model is better than that achievable with the cited finite element models, especially
in computing the stresses. The metis displacement finite element is notable for low sensitivity
to mesh distortion and in high-accuracy of stresses.
Keywords: hybrid stress element, metis displacement element, pure displacement model,
mesh distortion, high- accuracy.
Introduction
It was recognized that the popular finite element models based on element stiffness matrices
may be classified into three models: (1) compatibility model which is based on the
assumption of a displacement field continuous over the entire field; (2) equilibrium model
which is based on an assumed equilibrating stress field; (3) hybrid model in which
compatible displacement functions are assumed along the inter-element boundary in addition
to the assumed equilibrating stress field in each element. The last model is, of course, the
hybrid stress model which was proposed in 1964 by Pian [1]; its results can provide more
accurate stress estimation than the assumed displacement models.
In the early stage of development, the hybrid stress elements are based on the principle of
minimum complementary energy and assumed stresses need to be defined in a Cartesian
coordinate system for satisfaction of the equilibrium conditions. Although this hybrid model
is rational, the performance of the resulting assumed stress elements is affected very much by
mesh distortion. And it is not easy to satisfy coordinate invariance and equilibrium conditions
simultaneously as shown by Yeo et al [2]. In general, the hybrid stress finite element model
has no monotonic convergence and the accuracy of their responses are recognized by many
practical users of the finite element method, this work may be regarded as an attempt to relate
237
the former ones to the latter ones. Recently, great efforts have been devoted to developing
high- performance stress assumptions; the main issues have been coordinate invariance,
satisfaction of equilibrium conditions and sensitivity to mesh distortion [2-9].
The present paper deals with a special finite element model named metis element in our
terminology, which was proposed earlier by Nguyen- Dang Hung [10, 11]. The current model
is based on the original hybrid stress element and has monotonic convergence. The boundary
displacement field of this finite element is unisolvent (defined and continuous within the
domain, satisfying conformity between elements), and has been fully exploited in recent
years [12-16]. In this paper, an 8-node plane isoparametric metis displacement finite element
is developed in which the equilibrated stress field within the element is derived from the
stress function approach, and is the complete cubic stress field known as the best field. The
invariant condition as well as correct rank of stiffness matrix are guaranteed and possess
nearly the minimum number of stress parameters. Its results are compared with the pure
displacement element solutions and the recent developed model solutions.
The paper is organized as follows: Section 2 presents formulation of the problem, twofield variational formulations, strong and weak form as well as finite element discretization.
Section 3 gives results and discussions. Conclusions are summarized in section 4.
Formulation of the problem
2.1 Two-field variational formulation

Consider a linear elastic solid wherein geometry is described by Cartesian coordinates xi, the
tensor of stress is ij, the tensor of strain is denoted by ij, the prescribed surface traction is ti
at the external surfaces S ,the prescribed displacement is ui at the external surfaces Su, the
body force per unit volume is fi .
The relations between ui, ij and ij are presented by the following equations:
Equilibrium equations
ij , j + fi = 0
inV
Boundary conditions
(1)
ij n j = ti
on S
(2)
ui = ui
on Su
(3)
1
2
in V
(4)
Strain- displacement relations
ij = (ui , j + u j ,i )
Constitutive relations
ij =
A
ij
(5)
where A is the strain energy density.

Let ijs be a statically admissible stress field, by definition a stress field which satisfies
both the equilibrium equations (1) and the boundary conditions (2). And let uik be a
238
kinematically admissible displacement field, by definition a displacement field which

generates a strain field in such a way that both equations (3-4) are verified. Consider ije an
admissible stress field which satisfies only equation (1), and similarly u%i an admissible
displacement field which satisfies only (3).
According to the kinematically admissible displacement field uk, we have the total
potential energy functional:
(uk ) = ( k ) ( f , uk ) (, uk ) S
(6)
where
( k ) = A( ij k )dV ; A(ijk) : strain energy density

v
( f , u ) = fi uik dV ; f : applied body forces

k
( , u ) S =
k
n
j
u k dS ; n j ji : applied distributed loads on S
ij i
The pure displacement finite element models are derived from the variation of the
functional (6). Assumptions are made on the displacement field uk where the requirements of
unisolvence are satisfied a priori. We consider now the following hybrid stress functional:
( e , u% ) = ( e ) ( e , u% ) S + (, u% ) S
(7)
If u% is defined as the kinematically admissible displacement field uk, we obtain the

following functional:
( e , uk ) = ( e ) ( e , uk ) S + (, uk ) S
(8)
where
( e ) = B( ij e )dV , B(ije) : complementary strain energy density

V
( , u ) S = n j ij eui k dS
e
The models based on the variation of (8) create a special class of the hybrid stress finite
elements, which is the so-called Metis displacement element. Where the admissible stress
field e is in equilibrium within the element and the admissible displacement field u% in (7) is
replaced by the kinematically admissible displacement field uk in (6). The displacement field
defined only on the boundary in (7) must be the same as in (6), this means that unisolvent
conditions are assumed to be verified. This requirement may be satisfied by taking the same
nodal displacement for the metis element as for the pure displacement element.
Lemma 2.1: for all the stress fields which are statically admissible and unisolvent, the
following inequality always exists
(uk ) ( e , uk )
for any
(9)
239
Proof: Let c be an arbitrary deformation field which is compatible within V, the

compatibility equations (4) are verified. We have the following inequalities and equations for
all the corresponding deformation tensors ijc and ijc :
The Green formula for all the deformation tensors ijc
n
j
u dS = ij ijc dV + ij , j uic dV
c
ji i
(10)
A(ijc) is recast using the Lagrange transformation as

A( ijc ) = ijc ijc B ( ijc )
(11)
and the convexity of B ( ij ) leads to

B ( ije ) B ( ijc ) ( ije ijc ) ijc
(12)
We consider the sum of equation (6) and equation (8), and take into account (11)
( e , uk ) + (uk ) B( ije ) B( ijk ) dV + ijk ijk dV n j eji uik dS fi uik dV
V
(13)
Using equation (12), equation (13) leads to
( e , uk ) + (uk ) ije ijk dV n j eji uik dS fi uik dV
V
(14)
Using Green formula (10), and using (1) for ije , we have the right hand side of the
inequality (14) vanishes, and finally we obtain
( e , uk ) (uk )
(15)
The inequality states that the current model will yield a structure which is more flexible
than the compatible model of the same boundary displacement approximation. The dual
properties and the monotonic convergence of the current model are proven and given in detail
in Reference [12].
2.2 Finite element discretization

Consider the solid to be discretized into finite elements Vn, the body forces are not
considered, B ( ij ) is the complementary energy density, and ti are the imposed tractions
on S n . The functional with two independent fields ij and uj as proposed in (8) may be
written in the form
= B( ij )dV n j ji ui dS + ti ui dS
n Vn
Vn
S n
(16)
= Dijkl ij kl dV ti ui dS + ti ui dS
n Vn
Vn
S n
(17)
or
240
where Dijkl are the material's elastic compliance coefficients, Vn the boundary of the n-th
element, ij the stress field in equilibrium defined within the element, and ui the displacement
field defined uniquely on the boundary.
The current paper uses the stress functions approach in order to automatically generate a
discrete stress field in equilibrium without body loads. The given stresses are complete
polynomials, and this choice guarantees the necessary condition for the invariance of the
element. In plane elasticity, the derivation of the equilibrated stress field is based on an Airy
stress function with degree s
s
s ( x, y ) = mn x m n y n
(18)
m=0 n =0
In order to obtain the approximation functions for the stresses up to nd, the Airy stress
function is approximated by monomials xm-nyn [18]. The corresponding stress distributions are
directly calculated as
s
m
2s
x =
= (n 1)n mn x m n y n 2
2
y
m=2 n=2
s m2
2s
y =
=
(m n)(m n 1) mn x m n 2 y
2
x
m=2 n =0
2
s m 1
s
xy =
= (n m)n mn x m n 1 y n 1
x y m =1 n =1
(19)
Therefore equation (19) can be rewritten in the form

x ....
ni (ni 1) x mi ni y ni 2

m n 2 n
y = .... (mi ni )(mi ni 1) x i i y i

ni (ni mi ) x mi ni 1 y ni 1
xy ....
[3 x 1]
[3 x ND]
.... 1
.... M
.... ND
[ND x 1]
(20)
or
= S
where
T = { x y xy } ; T = {1
(21)
2 .... i .... ND } ; S T = {S x
Sy
S xy }
(22)
The index i varies from 1 to ND, we can find the relations between mi and ni in terms i as
shown in Reference [12]
1
(23)
ni = i + 2 0.5mi (mi + 1); mi = integer part of 1.99 +
25 + 8i 5
2
Remark 2.2. Pian [1] showed that the necessary condition so that the element stiffness
matrix is regular is
ND N q N r
(24)
241
where ND is the number of parameters , the Nq is the number of nodal displacement and Nr
is the number of rigid body modes. The current paper uses a typical 8-node isoparametric
element with 16 degrees of freedom. The minimum value of ND is 13, therefore the complete
polynomial base is minimum cubic in order to satisfy the invariant condition, and thus the
optimal number of parameters is 18 (see Reference [12, 13] for the details).
The displacement field uj must be kinematically defined only on the edge of the element.
In the present paper, the displacement field is in the natural coordinate system. The
displacements uT={u v}, in the x- and y- directions, respectively, are expressed in Vn in terms
of nodal displacement as
8
u = N i ( , )ui
i =1
(25)
v = N i ( , )vi
i =1
where xi and yi are the Cartesian coordinates of nodes of elements, Ni(,) are the appropriate
shape functions in the mapped coordinate system (,). The displacement vector is presented
in the matrix form as follows
u1
v
1
u2

u N1 0 N 2 0 N 3 0 ...... N8 0 .
(26)
u= =

v 0 N1 0 N 2 0 N 3 ...... 0 N8 .
.

u8

v8
u = [ N ]{q}
where
q T = {u1 v1 u2
v2
u3
v3 ..... u8
(27)
v8 }
(28)
As per standard isoparametric elements, the actual element geometry is mapped into a
normalized coordinate system using the same shape functions as used for displacement
interpolation. This coordinate transformation is given by
8
x = N i ( , ) xi
i =1
8
y = N i ( , ) yi
(29)
i =1
where Ni ( i = 1..8) are shape functions
242
1
1
N1 ( , ) = (1 )(1 )(1 + + ); N 2 ( , ) = (1 2 )(1 )
4
2
1
1
N 3 ( , ) = (1 + )(1 )(1 + ); N 4 ( , ) = (1 + )(1 2 )
4
2
1
1
N 5 ( , ) = (1 + )(1 + )(1 + + ); N 6 ( , ) = (1 2 )(1 + )
4
2
1
1
N 7 ( , ) = (1 )(1 + )(1 + ); N8 ( , ) = (1 )(1 2 )
4
2
(30)
The Jacobian matrix writes
y
Ni
x
i
=
y i =1 N i
xi

x

J=
x

x
x
-1
-1
J
J
=
=
;

y

y
Ni
y
i
Ni
y
i
(31)
y
y

1
x
=
det J x x

y

(32)
where J-1 is the inverse of the Jacobian matrix, det J is the determinant of the Jacobian of the
co-ordinate transformation. The first term of the functional can be presented in matrix form
1
1
Dijkl ij kl dV = T DdV
2 Vn
2 Vn
(33)
Substituting (21) into (33) we obtain

1
1
Dijkl ij kl dV = T H e
2 Vn
2
(34)
where matrix H e is defined as

He =
Vn
1 1
DSdV = t S T DSdetJdd
(35)
-1 -1
and in plane stress conditions, matrix D is defined as

1
1
D =
E
0
1
0
0
0
2(1 + )
(36)
in plane strain conditions, matrix D is defined as

1
(1 + )
D=
1
E
0
0
0
0
2
(37)
243
Taking into account the equilibrated stress field (ij,j = 0). The second term of the
functional (17) can be written in matrix form as
tiui dS = n j ijui dS = ( ij ui ), j dV = ( ij , j ui + ij ui, j )dV = ij ui, j dV (38)
Vn
Vn
Vn
Vn
Vn
u
v
x
v
v u
x
u
,
,
t
u
dS
=
+
+
+
=
+
{
}
{
}
v
i
i
x
y
xy
x
xy
xy
y
x
y
x y
Vn
u

y
y
(39)
Moreover we have
N1
u
u
u1
-1
-1
u = J = J

u
N1 u

1
y
N8
N2
u2 .....
u
N8
N2
u .....
u
2
8
(40)
N3
N3
0
u1
v
1
u2

0 v2

.
0 .

.
u
8
v8
(41)
N1
-1
=J
N1

N2
N2
0
N8
N8
0 .....
0 .....
u
x
-1
u = J Pq

y
(42)
and we also have
, xy } = { x y
1 0
1 0
1 0
xy } 0 0 = T 0 0 = TS T 0 0
0 1
0 1
0 1
(43)
Similarly, we have
v
x
-1
v = J Qq

y
(44)
and
244
xy
, y } = { x y
0 0
0 0
0 0
T
T T
xy } 0 1 = 0 1 = S 0 1
1 0
1 0
1 0
(45)
Substituting (42-45) into (39) we obtain
t u dS = Gq
(46)
1 0
0 0
T
G = S 0 0 J -1 P + 0 1 J -1Q dV
Vn
1 0
0 1
(47)
i i
Vn
where matrix G is in the form
Taking into account (27), the third term of the functional (17) can be presented in matrix
form as
t u dS = u
i i
S n
tdS =
S n
N T tdS = q T g
(48)
S n
where the generalized force vector g is computed as follows

g=
tdS
(49)
S n
and the consistent nodal forces of node i caused by boundary tractions are
x
y
Pxi = N i tt
d
tn

Se
x
y
Pyi = N i tn
d
+ tt

Se
(50)
where tn and tt are force components per unit length in the normal and tangential direction,
respectively, prescribed for a typical 8-node isoparametric element. The distributed surface
loads are treated as equivalent nodal forces (see Reference [20] for the details). Thus, the
generalized force vector g can be written as
g T = { Px1 Py1 Px 2 Py 2 .... Px8 Py 8 }
(51)
Substituting (33), (46) and (48) into (17), finally we obtain
1
= T H e TGq + q T g = 0
2
(52)
Invoking the stationary conditions of with respect to any q and , we obtain the
element level equations
H e Gq = 0
T
G + g = 0
(53)
The matrix He is symmetric and positive definite. Hence from the first equation of (53) we
have
245
= H e-1Gq
(54)
Substituting (54) into the second equation of (53) we obtain

g = G T = G T H e-1Gq
(55)
The following equation system is established
Kq = g
(56)
In addition, the element stiffness matrix K and the element stress matrix, , respectively,
are shown as below
K = G T H e-1G
(57)
= SH e -1Gq
Results and discussions
3.1 The first example - considerations of sensitivity to mesh distortion

Two well-known examples are taken to investigate the metis displacement finite element
(MFE) in view of the sensitivity to mesh distortion and the accuracy of stresses. The material
is assumed to be linear, elastic, homogenous, and isotropic.
The elements and numerical examples considered here can be summarized as
Elements
(1) Metis element (MFE)- present element
(2) Q4
: standard four-node isoparametric plane element
(3) 5-I : four-node plane hybrid stress element with five parameters
(4) RQ4 : four-node plane element with modified incompatible displacement modes
(5) OPS5 : four-node plane element via penalty-equilibrium approach
(6) M5 : four-node plane element with refined transformation matrix
To assess the MFE model and compare the performance of the cited elements, a distortion
sensitivity study involving a two-element cantilever beam under bending and acted upon by
shear forces at the tip is examined as shown in Figure 1a and Figure 1b. The degree of
geometric distortion of these elements is represented by the dimension 'e', and material
properties are E= 1500, =0.25. The description and comparison of the elements Q4, 5-I,
RQ4, M5, and OPS5 were also given in Reference [2-5]
Y
150
100
2
100
Figure 1a: Mesh distortion test

under end shear
Figure 1b: Mesh distortion test

under end moment
246
3.1.1 The cantilever beam under end shear forces
The first example is a uniform cantilever beam subjected to an end load P=150 in Figure
1a.The adopted mesh consists of two elements. The full integration is used in this example.
Figure 2a and Figure 2b illustrate the comparison of the metis displacement finite element
(MFE) and the cited elements, the calculations for the vertical displacement at point A and the
normalized bending stress at point B (x = 1, y = 2) with the computed exact solutions in
Reference [19], uex (at point A) = 102.625, ex(at point B) = -4050.
Element distortion test under end shear
Vertical displacement at point A
1.4
Metis element (MFE)
1.2
Q4
5beta-I
0.8
RQ4
0.6
M5beta
OPS5beta
0.4
Exact
0.2
0
-1
Distortion e
Figure 2a: The vertical displacement uA/uex ,versus distortion 'e'

Element distortion test under end shear
1.2
Sigma(x) at point B
Q4
5beta - I
0.8
RQ4
0.6
M5beta
0.4
OPS5beta
Metis element (MFE)
0.2
Exact
0
-1
Distortion e
Figure 2b: The normalized bending stress B / ex ,versus distortion 'e'.

Figure 2a shows the predicted displacement distributions at point A versus distortion 'e'.
The M5 element yields the best result with maximum error 4.6 %, followed by the MFE
element with maximum error 9.9 %, the RQ4 element with maximum error 14.7 %, the
247
OPS5 element with maximum error 30.2 % , the 5-I element (standard hybrid stress
element of Pian) with maximum error 34.6 %, and Q4 element yields the worst result with
maximum error 77.4 %. The results show that for the vertical displacement at point A, the
present metis element (MFE) yields reliable displacement solution even when the elements
are highly distorted.
Figure 2b shows the predicted distributions for the normalized bending stress at point B
versus distortion 'e'. The results show that the normalized bending stress of the present metis
element is nearly constant with respect to 'e', while the M5 element yields the result with
maximum error 9.2 %, followed by the RQ4 element with maximum error 10.1 %, the
OPS5 element with maximum error 14.1 %, the 5-I element with maximum error 21.1 %,
and Q4 element yields the worst result with maximum error 75.8 %. The calculated stress for
the present metis element shows excellent agreement with the exact solution by using this
simple mesh distortion.
3.1.2 The cantilever beam under end shear moment
The second example is a cantilever beam under bending shown in Figure 1b, load P=1000,
again discretized using two elements. The full integration is also used in this example.
Figure (3a) and (3b) illustrate the comparison of the metis displacement finite element
and the cited models calculations for the vertical displacement at point A and the normalized
bending stress at point B( x = 1, y = 2 ) with the exact solutions computed in [19], uex (at
point A) = 100, ex(at point B) = -3000.
Vertical displacement at point

A
Element distortion test under end moment

1.6
1.4
Metis element (MFE)
1.2
Q4
5beta-I
0.8
RQ4
0.6
M5beta
0.4
OPS5beta
0.2
Exact
0
-1
Distortion e
Figure 3a: The vertical displacement uA/uex ,versus distortion 'e'
248
Element distortion test under end moment

1.2
Sigma(x) at point B
1
Q4
0.8
5beta - I
RQ4
0.6
M5beta
OPS5beta
0.4
Metis element (MFE)

Exact
0.2
0
-1
Distortion e
Figure 3b: The normalized bending stress B /ex ,versus distortion 'e'
Figure 3a shows that the predicted distribution for uA of the MFE is in excellent
agreement with the exact solution in any variation of eccentricity 'e', it shows almost no
variation with element distortion, while the M5 element yields results with maximum error
15.3 %, followed by the RQ4 element with maximum error 20.2 %, the 5-I element with
maximum error 35.2 %, the OPS5 element with maximum error 49.8 %, and finally the Q4
element yields the worst result with maximum error 93 % which means that mesh distortion
makes the Q4 solution erroneous.
Similarly, in Figure 3b, it is found that the predicted distribution for the normalized
bending stress of the metis displacement finite element yields the best result. It shows almost
no variation with any element distortion with maximum error 2.2 %.
Throughout the two examples investigated, we conclude that the present metis element
yields reliable solutions even when the elements are highly distorted for both types of
loading. The numerical results show that the MFE element is not sensitive to mesh distortion
and yields highly accurate stresses.
3.2 The second example problem - a thick walled cylinder under internal
pressure
The second example problem is an axisymmetric thick-walled cylinder under internal
pressure. This problem is selected to assess the metis displacement finite element
performance in a more general isoparametric shape, and to show that the MFE is to be
particularly accurate in axisymmetric problems. The current element is compared with the
exact solution and the standard 8-nodes isoparametric displacement element.
Due to the symmetry of the body we only consider part of the body from = 0o to = 90o,
and symmetry boundary conditions are imposed at = 0o and = 90o as shown in Figure 4.
249
2a
2b
Figure 4: A thick walled cylinder under internal pressure

The problem is discretized by 9 elements as illustrated in Figure 4 and assumed to be in a
state of plane strain. For this case the inner radius is 5 and the outer radius is 20. The applied
internal pressure is p= 10, and the material properties are E= 1000, =0.3, thickness t= 1. The
3x3 reduced integration rule is used in calculating matrices. The exact solution to this
problem is given by Timoshenko and Goodier [19].
Figure 5a and Figure 5b illustrate the comparison between the radial stress r and the hoop
stress that are computed at Gauss points versus the angle for both models (the MFE and
the pure displacement finite element) and the exact stresses re, e (see Reference [12] for
details).
Figure 5a shows the radial stress as a function of along R= 6.667 which passes through a
middle line of 3x3 Gauss points. The stress should be independent of as shown in [19].
The stress distribution of the MFE model is a sinusoidal curve which oscillates about the
exact solution- axis while the stress distribution of the standard 8- node isoparametric
displacement model is linear but situated far from the exact solution axis.
This shows that the MFE model yields better distributions for the radial stress with less
than 3.39 % error at all locations. In contrast, the standard 8-node isoparametric displacement
model yields a less accurate distribution with 12.74 % error at all locations, compared with
the exact solution.
Similarly, Figure 5b shows the hoop stress as a function of along R= 6.667 which passes
through a middle line of 3x3 Gauss points. The hoop stress distribution for both models are
sinusoidal curves, and it is seen that the MFE model yields the better distribution with less
than 1.69 % error at all locations, while the standard 8-node isoparametric displacement
model yields 4.46 - 4.64 % error.
Figure 6a and Figure 6b illustrate the comparison between the predicted stresses (radial
stress r and hoop stress ) which are computed at Gauss points along the line =750 for
both models (the MFE and the pure displacement finite element) and the exact stresses re
,e. The distribution of the radial and hoop stresses r and as a function of radial location
along the line = 750 which passes through a line of 3x3 Gauss points are shown in Figure 6a
and Figure 6b. The best distributions for stresses are obtained by using the MFE model which
shows excellent agreement at all locations along the line. The hoop stresses are obtained by
using the displacement model in Figure 6b which are in good agreement at all locations along
the line. And its radial stresses distributions are slightly more accurate than the exact solution
in Figure 6a.
250
Radial stress distribution

1.14
1.12
sigma(r)/sigma(exact)
1.1
1.08
1.06
MFE
1.04
Displacement model
1.02
Exact
1
0.98
0.96
0.94
-20
20
40
60
80
100
Angle
Figure 5a: Radial stress distributions versus for the thick cylinder problem with
9 elements along the line R= 6.667 which passes through Gauss points
Hoop stress distribution
sigma(theta)/sigma(exact)
1.03
1.02
1.01
1
MFE
0.99
Displacement model
0.98
Exact
0.97
0.96
0.95
-20
20
40
60
80
100
Angle
Figure 5b: Hoop stress distribution versus for the thick cylinder problem with 9 elements
along the line R= 6.667 which passes through Gauss points
251
Radial stress, versus radial position R

1.2
1
MFE
Sigma(r)/p
0.8
Displacement model
Exact
0.6
0.4
0.2
0
0
10
15
20
25
Figure 6a: Radial stress distribution r versus R for the thick cylinder problem
with 9 elements along the line = 750 which passes through Gauss points
Hoop stress, versus radial position R

1.2
Sigma(theta)/p
1
0.8
MFE
Displacement model
0.6
Exact
0.4
0.2
0
0
10
15
20
25
Figure 6b: Hoop stress distribution versus R for the thick cylinder problem with 9
elements along the line =750 which passes through Gauss points
Conclusions
From the results and discussions we can reach the following conclusions:
The basis theory of the metis displacement finite element which does not seem to
have been documented well in the literature hitherto is introduced in the paper.
252
An 8- node isoparametric metis finite element is also developed in detail with the
plausible equilibrated stress field. It possesses the desirable element characteristics of
an ideal finite element: (a) free from zero-energy or kinematic deformation modes; (b)
invariant with reference coordinates; (c) accurate stress evaluation.
The current approach is notable for low sensitivity to mesh distortion and in highaccuracy of stresses. Reliability of stress calculation in the case of distorted meshes is
important in stress analysis problems, especially in the geometric and non-linear
problems, and fracture/ failure criteria in composite materials where the solutions are
strongly dependent on the accuracy of stresses. From this point of view, the current
model would be more attractive than the pure displacement models and it is not too
much to say that the main issue of the current model has been the development of
stress assumptions which can ensure the accuracy of stresses even when meshes are
distorted. This superiority is particularly evident when the element is used in a more
general shape such as in thick cylinder problems. The calculated stresses are
accurately represented at all locations.
Acknowledgements
The first author would like to dedicate his work to Professor Nguyen- Dang Hung for his 65th
anniversary.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
T.H.H. Pian, Derivation of element stiffness matrices by assumed stress distributions, AIAA. J. 2,
1333-1336, 1964.
S.T. Yeo and B.C. Lee, New stress assumption for hybrid stress elements and refined four-node plane
and eight-node brick elements, Int. J. Numer. Methods Engrg. 40, 2933-2952, 1997.
K.Y. Yuan, J.C. Wen and T.H.H. Pian, A unified theory for formulation of hybrid stress membrane
elements, Int. J. Numer. Methods Engrg. 37, 457-474, 1994.
M.C. Bouzeghoub and M.J. Gunn, On stress interpolation for hybrid models, Int. J. Numer. Methods
Engrg, 37, 895-904, 1994.
K.Y. Yuan, Y.S. Huang and T.H.H. Pian, New strategy for assumed stresses for 4-node hybrid stress
membrane element, Int. J. Numer. Methods Engrg. 36, 1747-1763, 1993.
C.C. Wu, Dual zero energy modes in mixed/hybrid element-definition, analysis and control, Comput.
Methods Appl. Mech. Engrg. 81, 39-56, 1990.
W. Feng, S.V. Hoa and Q. Huang, Classification of stress modes in assumed stress fields of hybrid
finite elements, Int. J. Numer. Methods Engrg. 40, 4313-4339, 1997.
J.P. Moitinho de Almeida and O.J.B. Almeida Pereira, A set of hybrid equilibrium finite element
models for the analysis of three-dimensional solids, Int. J. Numer. Methods Engrg. 39, 2789-2802,
1996.
Z.P. Jiao, T.H.H. Pian and Sheng Yong, A new formulation of isoparametric finite elements and the
relationship between hybrid stress element and incompatible element, Int. J. Numer. Methods Engrg.
40, 15-27, 1997.
Nguyen-Dang Hung, "Sur un classe particulire dlments finis hybrides: Les lment mtis", in: Proc.
The First International Congress of GAMNI, (Dunod Technique, Paris) 5363, 1979.
Nguyen-Dang Hung, On the monotony and the convergence of a special class of hybrid finite element:
The metis elements, in: Proc. The IUTAM Symposium Variational Methods in the Mechanics of
Solids, edited by S.N. Nasser, (Northwestern University, Evaston, IL, USA) 208-213, 1980.
Dang-Dinh Thi, A metis displacement finite element analysis, Master thesis, University of Liege,
1999.
Dang-Dinh Thi and Nguyen-Dang Hung, Invariant Isoparametric Metis Displacement Elements, in:
Proc. International Colloquium in Mechanics of Solid, Fluids, Structures and Interactions, (Nha Trang,
Vietnam) 468-489, 2000.
253
[14]
[15]
[16]
[17]
[18]
[19]
[20]
Nguyen-Dang Hung and Tran-Thanh Ngoc, Analysis of cracked plates and shells using Metis finite
element model, Finite Elem. Anal. Des. 40, 855-878, 2004.
Nguyen-Tien Duong and Nguyen-Dang Hung, Direct and indirect metis element methods for edge
delamination in laminates, Comput Mech, WCCM VI in conjunction with APCOM04, (Beijing,
China), Tsinghua University Press & Springer-Verlag, 2004.
Nguyen-Tien Duong and Nguyen-Dang Hung, Regular and singular metis finite element models for
delamination in composite laminates, Finite Elem. Anal. Des. 42, 650-659, 2006.
Nguyen-Tien Duong, "Dlaminage des composites sous chargements complexes par lments finis
mtis", Ph. D. Thesis (in French), University of Lige, 2005.
L.E. Malvern, Introduction to the Mechanics of a Continuous Medium, Prentice-Hall, 1969.
S.P. Timoshenko and J.N. Goodier, Theory of elasticity, 3rd ed, New York, McGraw-Hill, 1970.
E. Hinton and D.R.J. Owen, Finite element programming, Academic Press, 1977.
254
Puncture Mechanics of Rubber Membranes

Nguyen-Chien Thang*, Vu-Khanh Toan*, Jaime Lara+
*
Dpartement de gnie mcanique
Universit du Qubec / cole de technologie suprieure
1100, rue Notre-Dame Ouest
Montreal, Qubec, Canada, H3C 1K3
+
Institut de recherche Robert-Sauv en sant et en scurit du travail
505 de Maisonneuve Ouest
Montreal, QC, Canada, H3A 3C2
Abstract
Puncture resistance is one of the major mechanical properties of rubber membranes. The
intrinsic material parameters controlling the puncture are still unknown. The standard ASTM
F1342 experiment uses a conical puncture probe to evaluate puncture resistance. This test is
commonly used for any type of protective clothing. This work aims to investigate the
quantitative material parameters that control the puncture resistance of thin rubber
membranes. Three commercial rubbers are investigated. The results demonstrate that the
probe-tip geometry strongly affects the results in puncture characterization. The maximum
puncture force depends on the contact surface between the elastomer membrane and the
probe tip. Using the Mooney strain-energy function, the indentation force was calculated for
elastomer membranes with large deformations in the absence of friction. The puncture
strengths of elastomer membranes are much lower than their tensile and biaxial strengths.
The results show also that the puncture depends on the maximum local deformation and its
independent of the indentor geometry.
Keywords: puncture resistance, failure strain, rubber, elastomer membrane, protective
clothing.
Introduction
Puncture resistance is one of the major mechanical properties of rubber membranes. The
intrinsic material parameters controlling the puncture are still unknown. Various
investigations have been performed on specific cases involving different materials. However,
the reported investigations are either qualitative, and do not provide a fundamental
understanding of the mechanisms controlling puncture, or are not applicable to the highly
elastic elastomer membranes. The following section outlines the reported works on puncture
involving different behaviours in various materials.
The puncture resistance of protective gloves to surgical needle was studied in [1 - 3]. In
these works, 19 commercially available surgical glove liners were ranked according to a
measurement of the puncture force. However, the thickness is not taken into account in these
works, and the results are only qualitative for purposes of comparison. The puncture
behaviour of rigid plaques such as polycarbonates and acrylics is reported in [4, 5]. In these
works, the maximum load was found to vary linearly with both the plunger diameter and the
255
sample thickness, but no quantitative analysis of the results was performed. More
fundamental investigations on puncture have been carried out on rubber blocks by fracture
mechanics [6]. Using fracture mechanics, a method has been developed to calculate the
fracture energy in puncture. However, the rubber-block situation involves surface indentation
and is not applicable to the case of puncture of elastomer membranes. The quantitative
characterization of puncture resistance has also been developed for geotextiles and
geomembranes. In these materials, a correlation was found between the puncture force and
the tensile strength for probes greater than 20 mm in diameter [7-10]. Considering a loading
state of pure axisymmetric tension, the puncture resistance of geotextile membranes was
calculated in term of tensile wide-width strength. However, the results are only applicable in
the case of linearly elastic deformation. Thin rubber membranes are for their part
hyperelastic, and highly nonlinear.
Presently, to evaluate the puncture resistance of elastomer membranes, the standard
ASTM F1342 and ISO 13996 tests are the most commonly used methods. These tests were
developed to evaluate the puncture resistance of thin flexible-materials. The ASTM F 1342
[11] is designed for any type of protective clothing, including coated fabrics, laminates,
textiles, plastics, elastomeric films or flexible materials. This test method determines the
puncture resistance of a material specimen by measuring the maximum force required for a
conical puncture probe to penetrate through a specimen clamped between two plates with
chamfered holes not less than 10 mm in diameter. This work aims to investigate the material
parameters that control the puncture resistance of thin rubber membranes in the ASTM
standard F 1342. Both the test and the mechanisms of puncture are analyzed in order to point
out the intrinsic parameters controlling puncture. Investigations are also performed to
determine the effects of probe geometry and the correlations with other mechanical
properties.
A better understanding of puncture mechanics will be helpful to develop suitable methods
to evaluate puncture resistance, to predict the failure of protective clothing materials as well
as to develop new protective materials for better puncture resistance.
Experimental procedure
2.1 Materials
Three types of commercial rubbers commonly used for protective gloves, neoprene, nitrile,
and natural rubber, were investigated. Neoprene sheets with three different thicknesses, 0.40,
0.78 and 1.57 mm, were obtained from Fairprene Industrial Products. The 0.30 mm thick
nitrile samples were cut from Nitrile Gloves manufactured by Ansell Co. The 1.0 mm thick
natural rubber samples were cut from NR Gloves, manufactured by Sandstrahler Co.
2.2 Puncture test

The puncture test was carried out on an Instron 1137 universal-testing machine. The puncture
probes were held by a pin chuck mounted to the load cell. The elastomer sample was clamped
between two steel plates under pressurized air, as shown in Figure 1. The hole of the lower
plate was chamfered to avoid stress concentration.
256
Figure 1: a) Puncture test device
b) Sketch of sample holder
Conical probes with different cone angles, and diameters are investigated in this work, as
shown in Figure 2. These probes were manufactured from stainless steel.
Figure 2: Conical probes using in puncture study
2.3 Tensile test

The uniaxial tensile test was performed according to the D 412 ASTM standard [12], using
the C-type dog-bone sample. The test was carried out at room temperature, with a crosshead
speed of 100 mm/min. The elongation was measured by an MTS LX 500 laser extensometer.
2.4 Biaxial test

The balloon test [13, 14] was used to determine the failure stress and strain in an equibiaxial
state (Figure 3). The pressure was measured using a pressure gauge while the maximum
diameter of the balloon before burst was recorded through an MTS LX 500 laser
extensometer.
257
Figure 3: Balloon type equi-biaxial extension equipment.
FF
Figure 4: Typical diagram of force vs. vertical displacement for 0.3 mm thick Nitrile samples.
20
Cylin. probe d = d2
->
Puncture force (N)
15
10
Conical probe:
d1 = 0,05 mm
d2 = 1,05 mm
<- Cylin. probe d = d1
0
0
20
40
60
80
100
Probe tip angle (degree)
Figure 5: Variation of maximum puncture force in 0.78 mm thick Neoprene as a function of

probe-tip angle.
258
Table 1: Failure true stress of tensile and puncture tests (in parenthesis: SD - standard
deviation)
Material
Nitrile
Neoprene
NR
Failure true stress (MPa)

Tensile
210 (22)
71 (7)
325 (31)
Puncture
128 (18)
45 (7)
104 (11)
Table 1 shows the comparison between the measured tensile stress and the maximum stress in
the puncture of elastomer membrane, calculated from maximum puncture force. The
maximum stress in puncture appears to be much smaller than the maximum tensile stress.
Observations of the deformations of the sample before puncture revealed that at the top
surface of the probe tip, the deformation is at a maximum, and is in an equibiaxial state
(Figure 6), there is no concentration of stress in this experiment due to the probe tip has
hemispherical shape. Since puncture involves a biaxial state of stress, it is relevant to verify
whether these discrepancies can be due to the effect of biaxial stress. To determine the
maximum equibiaxial strength, the elastomer membranes were air-inflated until they burst.
The pressure was measured and controlled using a pressure gauge while the maximum
diameter of the balloon before burst was recorded through an MTS LX 500 laser
extensometer. The true failure strain and stress were calculated by [13, 14]:
2 = Sball/S0 and
= 2 RP/2t
(1)
where R is the maximum radius of the balloon at burst, P is the maximum pressure, t is the
initial thickness, Sball is the maximum surface area of the balloon at burst, and S0 is the initial
surface of the sample.
Figure 6: Biaxial deformation at probe-tip surface.
259
Table 2: Failure true stress of equibiaxial and puncture tests (in parenthesis: SD)
Material
Nitrile
Neoprene
NR
Failure true stress (MPa)

Biaxial
270 (29)
73 (9)
317 (35)
Puncture
128 (18)
45 (7)
104 (11)
Table 2 shows the comparison between the equibiaxial and the calculated maximum puncture
stresses of different elastomer membranes. The results point out that the puncture strength is
also smaller than the biaxial strength.
The comparison between the maximum deformations measured in puncture, tensile, and
equibiaxial tests is shown in Table 3. The results once again confirm that puncture is
controlled by a maximum local deformation of the membrane at the probe tip. This
deformation is much smaller than the maximum tensile and equibiaxial deformations.
Table 3: Failure engineering strain and true strain of tensile biaxial balloon tests compared to
puncture tests (in parenthesis: standard deviation)
Material
Nitrile
Neoprene
NR
Failure engineering strain (%)

Tensile
410 (22)
376 (18)
971 (51)
Biaxial
239 (17)
186 (16)
493 (42)
Puncture
132 (11)
146 (7)
318 (23)
Failure true strain (%)

Tensile
163 (10)
156 (8)
237 (14)
Biaxial
122 (9)
105 (9)
178 (15)
Puncture
84 (8)
90 (5)
143 (12)
260
Figure 7: Deformation of elastomer membranes using puncture.

Observations revealed that the elastomer membrane adheres to the probe over a specific
distance from the probe tip before puncture (Figure 7). The measured puncture force thus
results from the contact force at the probe. From the deformations of the elastomer membrane
shown in Figure 7, it could be assumed that the contact force would include both normal and
tangential forces due to friction. In order to verify the contribution of the tangential force due
to friction, puncture tests with lubricated probe tips were performed, and the results revealed
that friction played only a minor role in puncture. The maximum puncture force in the
presence of a lubricant was only slightly reduced (less than 6%). The maximum force
measured in the puncture tests was therefore due to the normal pressure of the elastomer
membrane on the probe tip.
Large deformations in the indentations of circular elastic membranes by a spherical
indentor in the absence of friction have been analyzed in [15]. Using a similar approach, the
theoretical puncture force will be assessed in terms of the conical probe-tip geometry. A point
P on the deformed membrane can be described by two coordinates: , the distance from the
point to the axis of symmetry, and , the meridian arc length between the centre of the
membrane and the point shown in Figure 8. The solution of the deformed membrane is in the
form:
261
Figure 8: Coordinates in indentation of circular elastic membrane with conical probe.

= (r) and = (r)
(2)
where r is the coordinate of the point P on the undeformed membrane (corresponding to the
point P on the deformed membrane). The meridian and circumferential stretch-ratios of the
deformed membrane are thus, respectively, 1 = d/dr and 2 = /r. There are two distinct
regions on the deformed membrane, the region which is in contact with the puncture probe,
called the contact region, and the other, called the non-contact region.
In the non-contact region, since it has no external load, the equations of equilibrium in the
meridian and normal directions are:
dT1 1
+ (T1 T2 ) = 0 ,
d
K 1T1 + K 2T2 = 0
(3)
where T1 and T2 are respectively the stress resultants per unit edge length in the meridian and
circumferential directions, and K1 and K2 are the principal curvatures of the arcs in the
corresponding directions. Since elastomer membranes are elastic, isotropic and
incompressible, their mechanical property can be described by the well-known Mooney
strain-energy function [16]:
W(I, II) = C1(I-3) + C 2(II-3) = C1(I-3 + (II-3))
(4)
where the strain invariants are I = 12 + 22 + 12 22 , II = 12 + 22 + 12 22 , III = 123 = 1,

and C1, C2 are material constants, and = C2/C1. From this function, the stress as a function
of stretch-ratio can be derived, respectively:
262
1 =
W
W W I W II
W
=
+
=
21 213 2 2 +
213 + 2122
I 1 II 1
1
I
II
W
3 3
2
+ 22
= 2 2 1 13 2 3
= 2 2 1 1 2 C1 + 2 C 2
II
I
2
(5)
The stress resultants per unit edge length in the meridian and circumferential directions are
respectively:
h
h
T1 = 1true hdeformed = 11 h3 = 11
= 1 ; (III = 123 = 1 -> 3 = 1/ 12)
12
1h
= 2hC1 1 13 2 3 (1 + 22 )
2
2
and
T2 =
2h
= 2hC1 2 13 2 3 (1 + 12 )
1
1
(6)
where h is the undeformed thickness of the membrane.

By introducing three variables x = d/dr, 1 = d/dr and 2 = /r, and substituting these
equations into Equation (3), the governing differential equations for the variables, x, 1, 2 in
the non-contact region can be obtained.
In the contact region, the membrane conforms to the surface profile of the probe, which is a
known surface. Hence, and are related by the equation of the probe surface. The puncture
force includes:
F = F1 + F2,
where F1 is due to the pressure under the probe tip, and F2 is due to the pressure between the
conical surface of the probe tip and the membrane (the region between point A and point B
in Figure 7). F1 and F2 are calcculated as follows:
The contact region under the probe tip is in an equibiaxial stress state: 1 = 2 = 0, and r at
the point A (Figure 7) is determined by: r1 = R/0 (2 = /r = R/r). The pressure distribution
between the probe tip and the membrane in this region is:
pn = K1T1 + K2T2
(7)
where K1 = K2 = 1/R;
1
(8)
and T1 = T2 = C1 2h1 6 (1 + 20 )
0
The force F1 can thus be expressed as a function of the stretch ratio 0 under the probe tip by:
1
(9)
F1 = 4R2pn = 16R C1 h1 6 (1 + 20 )
For F2, the vertical force component due to the pressure in the zone AB (Figure 7) is:
F2 = A pn ave. sin = (R22 R 2 ) pn ave.
(10)
where A is the area of the conical surface (zone AB in Figure 7): A =
(R
sin
2
2
R2
263
The average pressure pn ave in the zone AB is:

pn ave. = K 1T1 + K 2T2
where K 1 = 0, K 2 = 2cos/(R2 + R) > pn ave. = 2cos/(R2 + R) T2
and T2 is the average values of T2 in the zone AB:
1
T2 = C1 2h 2 3 3 (1 + 12 )
1 1 2
(11)
(12)
where 1 , 2 are the average values of 1 , 2 in the zone AB (Figure 7).

In the contact region BC (between point B and point C in Figure 7), there results:
= R2
x = d/dr = 0
2 = /r = R2/r
K1 = 0 and K2 = 1/R2
(13)
The total load at each cross-section remains unchanged, and is equal to the total load F
exerted by the probe. Since T2 is perpendicular to F, by substituting 2 = /r = R2/r, this leads
to:
3
2
r
R2
3 r
(14)
F = 2 T1 = 2R22h C1 1 1 1 +
r
R2
R2
where F = F1 + F2 (Eqs. 9 and 10).
Figure 9 shows the variation of 1 and 2 with r. At r = a, the clamping sample holder
(Figure 8), 2 (r = a) = 1. The radius r3 in the underformed membrane, corresponding to the
point C after deformation, can be determined by iteration with the continuity conditions:
(14)
Eq. (3)
Non-contact region
Eq. (3)
Non-contact region
r3
A
r3
B C
B C
Figure 9: Schematic variation of 1 and 2 with r.

Eqs. (14) and Eqs. (3) give:
Eqs. (13) and Eqs. (3) render:
1(r3)c = 1(r3)nc
2(r3)c = 2(r3)nc
(15)
(16)
264
where the c and nc subscripts denote the quantities being obtained from the contact and noncontact regions, respectively. For an approximate value of 10 (1 at r = a), we can find r3a
and r3b that satisfy Equations 15 and 16 respectively. If |r3a r3b|/r3a > 0.001, take r3 = (r3a +
r3b)/2, recalculate Equations 3 and get the new 10, the iteration was performed until |r3a
r3b|/r3a < 0.001.
When r3 is determined, the radius r2 in the undeformed membrane, corresponding to the point
B after deformation, and 1 , 2 in Equation 9 can be calculated. The correct value of F2 is
obtained by Equation 10. Then the puncture force can be computed by F = F1 + F2 where F1
and F2 are calculated by Equation 9 and 10 respectively or F can be obtained by Equation 14.
40
Puncture force: F (N)
35
Measurements
30
Calculated from tensile
data
25
20
15
2 mm
10
13o
0
0
1
Probe tip diameter: d (mm)
Figure 10: Calculated and measured puncture force as a function of probe-tip diameter in 0.3
mm thick Nitrile sample (C1 = 902 kPa, = 0.28).
265
For the conical probes with = 13o, d2 = 2.0 mm, the calculated puncture force and
experimental results of these probes for Nitrile and NR are shown in Figures 10 and 11
respectively.
35
Measurements
Puncture force: F (N)
30
Calculated from tensile

data
25
20
15
2 mm
10
5
13
0
0
Probe tip diameter: d (mm)
Figure 11: Calculated and measured puncture force as a function of probe-tip diameter in 1.0
mm thick natural rubber sample (C1 = 70 kPa, = 2.96).
12
Stress (Mpa)
10
Experimental
Data
6
4
Mooney Model
2
0
0
Strain
x 100%
Figure 12: Stress-strain behavior of neoprene using tensile test.

The calculated puncture force of the conical probes is smaller than that obtained from
experimental results (7% - 15%). This is probably because the Mooney function is only
accurate for strains less than about 100% [16]. Figure 12 shows the comparison between the
Mooney function and the experimental data of tensile tests on neoprene samples. It can be
seen that above about 100%, the discrepancy between Mooneys function and experimental
data strongly increases with stretch ratio. Since the failure strain in the puncture is about
135% to 320% for the rubbers studied, the error of Mooneys function becomes significant.
Other material models for rubber are more precise (up to 250% of strain), but they have many
parameters (> 2), therefore the computation process will be more complicated. In the future
works, Mooneys model will be replaced by other precise material model.
266
Conclusions
It has been demonstrated in this work that the geometry and dimensions of the probe used in
the F1342 ASTM standard strongly affects the results of the puncture characterization of
elastomer membranes. The maximum puncture force depends on the contact surface between
the elastomer membrane and the probe tip. The indentation force has been calculated for
large deformations in the absence of friction using the Mooney strain-energy function. The
calculation agrees with experimental measurements within the range of accuracy of the
Mooney strain-energy function. The puncture strength is much smaller than both the tensile
and the biaxial stresses. The maximum stress in puncture corresponds to the maximum strains
measured from the top surface of the probe. The puncture of the elastomer membrane
depends on the maximum local deformation and its independent of the indentor geometry.
A better understanding of puncture mechanics will be helpful to develop suitable methods
to evaluate puncture resistance, to predict the failure of protective clothing materials as well
as to develop new protective materials for better puncture resistance.
References
[1] L.F. Leslie, J.A. Woods, J.G. Thacker, R.F. Morgan, W. McGregor, R.F. Edlich, Needle puncture
resistance of medical gloves, finger guards, and glove liners, J. Biomedi. Mater. Research 33, 41 46,
1996.
[2] J.A. Salkin, S.A. Stuchin, F.J. Kummer, R. Reininger, The effectiveness of cut-proof glove linears: cut and
puncture resistance, dexterity, and sensibility, Orthopedics: Thorofare 18, 1067 1071, 1995.
[3] Daniel J. Hewett, Protocol for the puncture resistance of medical glove liners, National Institute for
Occupational Safety and Health, USA, 1993.
[4] G.R. Tryson, M.T. Takemori, A.F. Yee, Puncture testing of plastics: effects of test geometry, Amer. Soc.
Mech. Eng. - AMD 35, 638 647, 1979.
[5] L.M. Carapellucci, A.F. Yee, Some problems associated with puncture testing of plastics, Poly. Eng. Sci.
27, 773 785, 1987.
[6] A. Stevenson, Kamarudin Ab Malek, On the puncture mechanics of rubber, Rub. Chem. Tech. 67, 743
760, 1994.
[7] V.P. Murphy, R.M. Koerner, CBR strength (Puncture) of geosynthetics, Geotech. Test. J. 3, 167 172,
1988.
[8] D. Narejo, R.M. Koerner, R.F. Wilson-Fahmy, Puncture protection of geomembrances, Part II:
experimental, Geosynthetics Inter. 3, 629 653, 1996.
[9] R.F. Wilson-Fahmy, D. Narejo, R.M. Koerner, Puncture protection of geomembrances, Part I: theory,
Geosynthetics Inter. 3, 605 628, 1996.
[10] Tushar K. Ghosh, Puncture resistance of pre-strained geotextiles and its relation to uniaxial tensile
strain at failure, Geotextiles and Geomembrances 16, 293 302, 1998.
[11] ASTM F1342 Standard Test method for Protective Clothing Material Resistance to Puncture, Annual
Book of ASTM Standards 11.03, USA, 1999.
[12] ASTM D412 Standard Test Methods for Vulcanized Rubber and Thermoplastic Elastomers - Tension,
Annual Book of ASTM Standards 9.01, USA, 1998.
[13] Y. Yoshihiro, K. Sueo, T. Yuji, S. Wataru, Biaxial Extension Strength of Carbon Black Reinforced
Rubber, Nihon Gomu Kyokaishi 67, 576 587, 1994.
[14] Sueo Kawataba, Fracture and mechanical behavior of rubber-like polymers under finite deformation in
biaxial stress field, J. Macromol. Sci. - Phys. B8(3), 605 630, 1973.
[15] W.H. Yang, K.H. Hsu, Indentation of a circular membrane, J. App. Mech. 38, 227 -230, 1971.
[16] M. Mooney, A theory of large elastic deformation, J. App. Phys. 11, 582 592, 1940.
267
Prediction of Delamination Strength at Interface

between Thin Film and Substrate by Cohesive Zone Model
Do-Van Truong*, Hiroyuki Hirakata** and Takayuki Kitamura**
* Graduate School of Engineering Physics and Mechanics,
Kyoto University, Japan
** Department of Engineering Physics and Mechanics, Kyoto University, Japan
Abstract
An electronic device consists of multi-layered submicron-thick films, and delamination often
takes place at an interface edge because of the stress singularity near the edge. Since the
stress singularity at an interface edge depends on the edge shape, the fracture mechanics
concept cannot be used to compare the delamination strength between the components with
different shapes. This paper aims to predic the delamination strength at the interface edge
with arbitrary shape using a cohesive zone model. Two different experiments are conducted
for a gold thin film on a silicon substrate to calibrate the cohesive law. The validity of the
approach is then discussed.
Keywords: Interface strength; cohesive zone model; delamination; criteria
Introduction
Delamination is one of the mechanical failures often met in microelectronic devices and it
sometimes brings about fatal malfunction of the system. Therefore, the quantitative
comparison of interface strength between bi-materials of different combinations is a
necessary task to choose the suitable materials and improve the reliability of the devices.
In terms of mechanical strength, delamination is mainly caused by the concentrated stress
near the interface edge due to the mismatch of the elastic property. According to Bogy [1],
the stress field near the interface edge is expressed by the following equation.
K
r
(1)
Here, K is the stress intensity parameter, r is the distance from the edge and is the stress
singularity. Crack initiation is governed by the singular stress field and the strength is
characterized by K [2-5]. The criterion of crack initiation is described using K as,
K = Kc,
(2)
where, Kc, is the critical stress intensity parameter which is characteristic for each interface
edge. However, because the dimension of K, MPa.m-, depends on the stress singularity, ,
namely the material combination and the edge shape, we cannot use Kc to compare the crack
initiation strength between different material combinations and between different edge shapes
even in the same material combination.
268
Recently, the cohesive zone model approach has emerged as a powerful tool to simulate
the fracture behavior [7-14] with an emphasized feature as removing the singularity by
considering a cohesive zone ahead of a crack tip [15]. In the cohesive zone, the cohesive
traction is related to the separation displacement under the cohesive law evaluated by fitting
experimental data. The parameters of the cohesive law are specific values to materials
(interfaces). By using a cohesive zone approach, we might evaluate the ad hoc interface
strength of bi-materials and compare the strength between them.
In this study, the method for evaluating the crack initiation strength from interface edges
between thin films and substrates using a cohesive zone model is developed. The cohesive
law for an interface between a gold (Au) thin film and a silicon (Si) substrate is calibrated by
crack initiation and propagation tests. Then, the crack initiation strength at the interface edges
with different shapes are estimated by finite element method (FEM) with the cohesive zone.
Finally, the interface strength between different materials is quantitatively compared on the
basis of the crack initiation stress
2.1 Materials and specimen

Figure 1 shows a cross-section of a tested material combination. An Au film with the
thickness of 200 nm is deposited on a Si substrate of thickness 500 m by the evaporation
method under a pressure of 1.9 103 Pa. Two different edge angles of 180o/180o (TypeA)
and 90o/90o (Type B) as shown in Fig 2 are prepared to calibrate the parameters of the
cohesive law.
Au
Si
400 nm
Fig. 1 Cross-section of a test material Au/Si
180o
180
Au
Si
a. Type A with the edge angle

of 180o/180o
90o
Au
90o
Si
b. Type B with the edge angle

of 90o/90o
Fig. 2 Edge shape of interface
269
Figure 3 shows the specimens of Type A and Type B and the loading system. A square
coupon is cut from a plate of the material with the Au film and the Si substrate. A cantilever
of stainless steel with a polished surface is glued on the coupon by standard epoxy. Then, the
Au film outside of the glue region is removed for avoiding the effect of film fracture during
the test. For Type A specimens, a pre-crack is introduced by the following method shown in
Fig. 4. The substrate and the cantilever are fixed together by a drop of epoxy. Then, a load is
applied to the cantilever edge. At a low load, a crack is initiated at the left edge of the
interface and stops at the point of the drop. Finally, the drop of epoxy is carefully removed
from the specimen. Two and three specimens are prepared for Type A and Type B,
respectively, to check the repeatability and the sizes of the specimens are listed in Table 1.
Crack propagation and initiation tests are conducted for the specimens A and B by a
remodelled micro-Vickers hardness tester (Shimadzu; MCTE-500). The load, P, is applied at
the end of the cantilever at a constant loading rate of 0.02 N/s by an electro-magnetic actuator
which is connected to a loading tip with a cone shape, and the displacement at the loading
point, uy, is monitored during the test. All the tests are conducted at a room temperature in
laboratory air.
Table 1 Specimen sizes and critical loads
Type A
Type B
A-1
A-2
B-1
B-2
B-3
L1
(mm)
L2
(mm)
H1
(mm)
H2
(mm)
Pre-crack
length a, mm
Critical
load Pc,
N
6.21
7.01
8.18
7.36
8.83
5.91
5.48
1.70
2.31
1.93
0.96
0.96
0.98
0.98
0.97
1.96
1.87
2.10
1.94
2.07
2.70
1.40
0
0
0
0.50
0.47
0.59
0.74
0.55
270
Detail of
Epoxy (thickness:
Thin film, Au
Substrate,
H2
Type B (without
Cantileve
Substr
H1
A
P
L
Crack length a
Type A (with
Fig. 3 Specimens and loading
Cantileve
Substra
H1
P
L1
L2
Fig. 4 Crack initiation
2.2 Experiment results

In all the tests, the relationship between P and uy is almost linear. At the critical value, Pc,
at which uy remarkably increases, the crack begins to propagate in Type A tests or a crack is
initiated at the interface edge in Type B tests along the interface between the Au film and the
Si substrate. The resulted critical loads Pc of all the specimens are listed in Table 1.
After the tests, the facture surfaces of both the film and substrate sides are examined by
Auger electron spectroscopy. Figures 5 (a) and (b) show the spectra on the film and the
substrate sides of a Type B specimen. Only Au peaks (69, 2024 and 2111 eV) are confirmed
271
Intensity
on the film side while Si peaks (92 and 1619 eV) are recognized on the substrate side. This
indicates that the delamination occurs at the clear interface between Au and Si. For the other
specimens, the similar spectra are observed.
A
u
A
0 u
50
100
150
A
200
Intensity
Energy of electron,
eV
a. Film (Au)
side
S
i
S
i
50
100
150
200
Energy of electron,
eV
b. Substrate (Si)
side
Fig. 5 Auger electron spectra on the fracture surfaces of e-Au/Si
Cohesive zone model approach
3.1 Idea of the approach

The cohesive zone approach is proposed to remove the singularity by considering a
cohesive zone ahead of a crack in which the atomic attraction acts between the separating
surfaces. The traction, , in the zone is assumed as a function of the separation distance, ,
and this relationship is considered as a cohesive law. The work of the interface separation per
unit area, , is calculated as follows.
m
= ( )d
0
(3)
Here, m is the critical separation distance corresponding with the vanishment of traction. For
an interface with a specific cohesive law, for example a bi-linear cohesive law shown in Fig.
6, the parameters of , m and m (the maximum traction) are specific values to the interface
and can be considered like material constants.
272
3.2 Cohesive zone in finite element analysis

On the basis of the above idea, Needleman [7] introduced the cohesive surface concept in
the FEM, in which the predicted crack growth path is inserted by a layer as a cohesive zone.
The material behavior in this zone is described by a cohesive law that relates the cohesive
traction to the separation displacement between the upper and lower cohesive surfaces. The
crack growth occurs only when the separation displacement at the tail of the cohesive zone
reaches a critical value, at which the cohesive traction vanishes.
There are many proposed cohesive laws such as bi-linear [11-13], trapezoidal [10, 12] and
exponential [16] laws. In this study, the bi-linear cohesive law as shown in Fig. 6 is chosen.
The bi-linear cohesive law is characterized by four parameters as the work of interface
separation per unit area, o, the maximum traction, m, the maximum separation
displacement, m, and the initial slope, c. o can be set at the value equal to the critical energy
release rate Gc (or critical J-integral, Jc) for an interface crack (edge angle: 180o/180o) in a
elastic bi-materials [17], while the other parameters are evaluated by fitting the critical load
obtained by the numerical analysis to that obtained by the experiment.
Traction,
m
Separation,
Fig. 6 Bi-linear cohesive law
To eliminate the effect of the thickness of the cohesive layer on the result, the value of
thickness is set at one unit length in the FEM [18]. This choice ensures that the strains are
equal to the relative separation displacements in the cohesive zone. Kitamura [19] indicated
that the stress concentrated region near the interface edge between the Au thin film and the Si
substrate is on the order of 10 nm. Thus, the smallest size of the cohesive element near the
interface edge is set at 10 nm in the FEM.
Calibration of the parameters of the cohesive law

The cohesive law shown in Fig. 6 can be expressed by a following relationship.
o =
1
1
m m = m ( 1 + 2 )
2
2
(4)
273
or
o =
1 m
+ 2
m
2 c
(5)
Among the four parameters, only three of them, for example o, m, and c, are
dependent because 2 can be obtained from Eq. (5).
The calibration process of the parameters of the cohesive law in Type A is started by a
following scheme. The work of the interface separation per unit area o is set at the value
equal to the critical J-integral, Jc. This is justified by the lack of occurred plasticity anywhere
in the specimen (in this case, Gc = Jc). Jc is calculated by a commercial FEM code, ABAQUS
6.5, under the plane strain condition. The FEM mesh is shown in Fig. 7, in which a ring of
collapsed quadratic quadrilateral elements is used at the crack tip. The critical load, Pc,
obtained from the experiment is applied to the edge of the cantilever. The materials are
assumed to be linear elastic and isotropic, and the elastic constants used are listed in Table 2.
The obtained values of Jc in each specimen are listed in Table 3, and o is set at the average
value of Jc = 0.165 J/m2.
.....
Epoxy
Au
200 nm
Si
Pre-crack
Crack tip
Fig. 7 Mesh division for FEM analysis
Table 2 Elastic constants used in FEM analysis

Material
Youngs modulus E, GPa
Poissons ratio
Si
167
0.30
Au
83
0.44
Epoxy
2.5
0.30
Stainless steel
200
0.30
Table 3 Critical J-integral value

Specimen
Critical J-integral Jc, J/m2
Average value of Jc, J/m2
A-1
A-2
0.150
0.180
0.165
274
Next, the other parameters m and c are calibrated in the FEM with the cohesive zone.
Because the initial slope, c, is insensitive to o in the range of from 106 GPa/m to 108 GPa/m,
c is chosen at 107 GPa/m in this study. The last parameter, the maximum traction, m, is
calibrated at 1.0 MPa which is the best match to the measured delamination load. Therefore,
the parameters o, c and m of the interface Au/Si are successfully calibrated at 0.165 J/m2,
107 GPa/m and 1.0 MPa, respectively.
The parameters of the cohesive law calibrated above are then validated by the Type B
results. Figure 8 shows the FEM model of a Type B specimen in which the cohesive zone is
inserted between the Au film and the Si substrate as a single layer of cohesive elements. The
crack initiation load is evaluated when the first cohesive element fails. The crack initiation
loads obtained from the cohesive zone approach and from the experiments are listed in Table
4. Both the results agree fairly well with each other. This signifies that the parameters of the
cohesive law calibrated for the Au/Si interface are valid and we can ad hoc evaluate the crack
initiation load for the interface edge with arbitrary edge shape using the cohesive zone
approach.
Epox
Au
Cohesive
Si
Fig. 8 Mesh division for FEM
Table 4 Crack initiation load
Specimen
Pcex, N
B-1
0.59
B-2
0.55
B-3
0.74
PcCZM, N
0.52
0.51
0.80
Crack initiation strength
5.1 Comparison of the crack initiation strength among different edge

shapes
In this section, the comparison of the crack initiation strengths for the interface edges with
different angles on the basis of the cohesive zone model approach. Figure 9 shows the FEM
275
models of Au interconnects on Si substrates with the edge angles of 60o/180o, 90o/180o and
120o/180o. The bottom face of the Si substrate is completely fixed and the symmetry
boundary condition is applied on the right face. Uniform gross stress, g, is applied on the
upper surface of the Au part and is monotonically increased. The crack initiation stress, gc,
is evaluated when the first cohesive element fails. The obtained crack initiation stresses are
listed in Table 5. The model with the larger angle of the Au edge has the lower crack
initiation strength.
60o
Au (or Cu)
Si
250 nm
(a) 60o/180o
90o
Au (or Cu)
Si
250 nm
(b) 90o/180o
120o
Au (or Cu)
Si
250 nm
(c) 120o/180o
Fig. 9 FEM models of Au/Si (Cu/Si) bi-materials with different edge angles
276
Table 5 Crack initiation stress of Au/Si interface edge

Edge angle
60o/180o
90o/180o
120o/180o
Crack initiation stress gc,

MPa
2.3
2.0
1.2
5.2 Comparison of the crack initiation strengths for different bi-materials

The crack initiation strength is compared for two different bi-materials. The experimental
result for a sputtered copper (Cu) thin film on a Si substrate reported by Hirakata et al. [20] is
used as the counterpart. In this experiment, almost the same specimen and the loading system
as the Type A test in the present study were used. The sizes and the critical loads at the crack
propagation of two specimens are listed in Table 6. The parameters of the cohesive law of the
Cu/Si bi-material are evaluated by the cohesive zone model approach presented above. The
elastic constants used are listed in Table 7. The values of o, c and m are 0.970 J/m2, 107
GPa/m and 5.0 MPa, respectively. The crack initiation stresses of the Cu/Si interface edges of
60o/180o, 90o/180o and 120o/180o are evaluated by the same manner as that used for the Au/Si
interface edges and the results are listed in Table 8. The results indicate the similar tendency
to the Au/Si models that the model with the larger angle of the Cu edge has the lower crack
initiation strength. Because the specimen geometry is the same for each combination, the
crack initiation strength can be quantitatively compared. The strength of the Cu/Si with the
angle of 60o/180o is 7.4 times higher than that of the Au/Si. The strength of the Cu/Si with the
angles of 90o/180o and 120o/180o is 5.0 and 4.7 times higher than that of the Au/Si,
respectively. The difference in the crack initiation strength between the Au/Si and Cu/Si is
larger for the smaller edge angle.
Table 6 Specimen sizes of bi-material Cu/Si
Type A
L1 (mm)
L2(mm)
H1(mm)
H2(mm)
10
10
2.22
2.69
1.8
1.8
2
2
Pre-crack
length a,
mm
0.51
0.68
Critical
load Pc, N
1.36
1.46
Table 7 Elastic constants used in FEM analysis

Material
Youngs modulus E, GPa
Poissons ratio
Si
Au Epoxy
167 Cu
2.5
0.30 0.34 0.30
Stainless steel
200
0.30
Table 8 Crack initiation stress of Cu/Si interface edge

Edge angle
60o/180o
90o/180o
120o/180o
Crack initiation stress gc,

MPa
17.0
10.0
5.6
277
Conclusions
(1) A method for evaluating the crack initiation strength from interface edges between thin
films and substrates using a cohesive zone model is developed.
(2) The parameters of the cohesive law for a gold (Au) thin film and silicon (Si) substrate
are calibrated by crack propagation tests along the interface. The crack initiation loads
obtained from the cohesive zone approach and from crack initiation experiments agree
fairly well with each other. This signifies the validity of the cohesive law model.
(3) The crack initiation stresses at the interface edges with different shapes are estimated on
the basis of the cohesive zone approach. The model with the larger angle of the Au edge
has the lower crack initiation strength.
(4) The interface strength of the Au/Si is quantitatively compared with a different bimaterial of a sputtered copper (Cu) on a Si substrate by the cohesive zone model
approach. The strength of the Cu/Si with the angle of 60o/180o, 90o/180o and 120o/180o
is 7.4, 5.0 and 4.7 times higher than that of the Au/Si, respectively. The difference in the
crack initiation strength between the Au/Si and the Cu/Si is larger for the smaller edge
angle.
References
[1] Bogy DB., Edge bonded dissimilar orthogonal elastic wedges under normal and shear stress, J Appl Mech
1968;35:14654
[2] H. Hirakata, T. Kitamura, Y. Yamamoto, Evaluation of interface strength of micro-dot on substrate by
means of AFM, Int. J. Solids Struct. 41 (2004) 3243.
[3] T. Kitamura, H. Hirakata, T. Itsuji, Effect of residual stress on delamination from interface edge between
nano-films, Engng. Fract. Mech. 70 (2003) 2089.
[4] Kitamura, T., Shibutani, T, Ueno, T., Crack initiation at free edge of interface between thin films in
advanced LSI, Engng Fract Mech 2002;69:128999
[5] F. Shang, T. Kitamura, H. Hirakata, I. Kanno, H. Kotera, K. Terada, Experimental and theoretical
investigations of delamination at free edge of interface between piezoelectric thin films on a substrate, Int. J.
Solids Struct. 42, 2005. 1729.
[6] C. F. Shih., R. J. Asaro, Elastic-plastic analysis pf cracks on bi-material interfaces: Part I-Small scale
yielding, J. Applied Mech. 55, 1998. pp 299-316.
[7] Needleman, A., A continuum model for void nucleation by inclusion debonding, ASME Journal of
Applied Mechanics 54, 525531, 1987
[8] Tvergaard. V., Effect of strain-dependent cohesive zone model on predictions of crack growth resistance,
J. Solid Structures., 1996., 33: 3297-3308
[9] Tvergaard V., Hutchinson JW., On the toughness of ductile adhesion joints, J. Mech. Phys.
Solids.,1996.,5:789-800
[10] Wei Y., Hutchinson JW., Interface strength,work of adhesion and plasticity in the peel test, Int. Journal of
Fracture, Vol.93, pp.315-333,1998.
[11] Mohammed, I. and Liechti, K. M., Cohesive Zone Modeling of Crack Nucleation from Bimaterial
Corners, J. Mech. Phys. of Solids 48, pp. 735-764 2000
[12] Z.-H. Jin, C.T. Sun., Cohesive fracture model based on necking, Int. Journal of Fracture, Vol.134,
pp.91108, 2005
[13] Z.-H. Jin, C.T. Sun., Cohesive zone modeling of interface fracture in elastic bi-materials, Eng. Frac.
Mech., 72, pp 1805-1817-2005
[14] C.R. Chena., O. Kolednik., J. Heerense., F.D. Fischerf., Three-dimensional modeling of ductile crack
growth: Cohesive zone parameters and crack tip triaxiality, Eng. Frac. Mech., 72., pp 2072-2094, 2005
[15] Brenblatt G. I., Mathematical Theory of Equiblirum Crack In Brittle, Adv. Alpp. Mech. Vol 7,1962
278
[16] M. Lane., H. R. Dauskardt., Plasticity Contribution to Interface Adhesion in Thin Film, Interconnect
Structures. J. Mater. Res.,15:2758-2768, 2000
[17] J. R. Rice., A path Independent Integral an the Approximate Analysis of Strain Concentration by Notches
and Cracks, J. Appl. Mech. 1968; 379-386
[18] ABAQUS 6.5, Analysis Users Manual, Vol. 4, pp 18.5.1-18.5.6
[19] Kitamura, T., Hirakata,H., Do-Van Truong., Initiation of interface crack at free edge between Thin Films
with weak stress singularity, Thin Solid Films, (under reviewing)
[20] H. Hirakata, T. Kitamura, T. Kusano, Pre-cracking technique for fracture mechanics experiments along
interface between thin film and substrate, Engng Fract Mech 2005;72:1892-1904
279
Creating Macro Commands by using Bacon Commands

to Simplify Meshing of 3-Dimensional Cracks
and Calculate the Lifetime of Cracked Structures due to Fatigue
Nguyen-Dang Hung*, Tran-Duc Han+,
Chau-Dinh Thanh+ and Nguyen-Van Hieu++
*LTAS, Division of Fracture Mechanics, University of Lige, Belgium
+EMMC5 students, Ho Chi Minh City University of Technology, Vietnam
++EMMC4 student, Ho Chi Minh City University of Technology, Vietnam
Abstract
The commercial finite element package SAMCEF including its post-processor SAMCEF
Postfac is one of the most powerful tools to compute stress intensity factors (SIFs) of cracked
problems. However, it is very complicated to generate a fine mesh of the structure with
cracks. Moreover, the .FI3 command of the pre-processor SAMCEF Bacon can only generate
meshing of 3D planar cracks. In this paper, by using commands of Bacon, some macro
commands have been created to generate mesh of hexagonal box with a 3D surface crack.
The box generates a very fine mesh in the vicinity of the crack front, a coarse mesh far out in
the crack front and a transfinite mesh on the outside surface of the box.
The lifetime of cracked structure is predicted by using the Paris-Erdogans law and the twoparameter model with the trapezoidal numerical integration. The results of lifetime prediction
are in good agreement with what is found in the literature.
Keywords: planar crack, surface crack, stress intensity factor, two-parameter model, lifetime
prediction, fatigue.
Introduction
The phenomenon of failure by catastrophic crack propagation in structural materials causes

problems of design and analysis in many fields of engineering. One important area when
considering such failure is the aerospace industry. The consequences of catastrophic failure
are also a prime consideration in civil and offshore engineering. Therefore the driving need
for methods which quantify the effects of the presence of cracks on material performance has
led to the evolution and development of the theory of fracture mechanics.
Cracks are present to some degree in all structures. They may exist as basic defects in the
constituent materials or they may be induced in construction or during service life. Therefore,
a fundamental requirement of fracture mechanics theory is some means of assessing the
stability of such cracks. In this respect, the most significant advance has been the introduction
of the stress intensity factor (SIF) as a single parameter for categorizing the onset of crack
propagation. The use of the SIF in examining crack stability requires an accurate knowledge
of the stress field in the vicinity of the crack front for the structural geometry, loading and
boundary conditions. Unfortunately, analytical solutions only exist for selected, relatively
simple cases, and it is in the determination of the stress and displacement fields for complex
280
practical situations that numerical technique such as finite elements and boundary integral
methods play an important role.
Over the last years, the finite element technique has become firmly established as a versatile
tool for the numerical solution of engineering problems including cracks. The available
commercial finite element packages have been widely applied in linear or nonlinear analysis
on non-cracked and cracked structures. SAMCEF (http://www.samcef.com) is one of the
most powerful tools to compute SIFs of cracked problems. However, it is very complicated to
define a very fine mesh in the vicinity of the crack front and a transition from the fine mesh
around the crack front to the coarse mesh far out in the crack front. As a result, most of the
work has been limited for cracks in the structure, since the mesh generation is still the most
time-consuming work. Especially, SAMCEF does not support the definition of 3D surface
cracks. The .FI3 command of Bacon can only generate the meshing of 3D planar cracks.
In the present work, by using .FI3 and other commands of Bacon module of SAMCEF, some
macro commands have been developed to generate automatically mesh for an hexagonal box
with a three-dimensional crack. In the hexagonal box, these macro commands generate a very
fine mesh in the vicinity of the crack front and gradually a coarse mesh in the region far out
from the crack front. Moreover, they generate a transfinite mesh on outside surfaces of the
box. Therefore, it is easier to combine the meshed cracked box with the remaining meshed
domains of the structure. Consequently, it will not take a lot of time to mesh the cracked
structures, which is the most difficult task of the finite element analysis.
The assumption that an actual part-through crack can be replaced by an equivalent elliptical
surface crack has been experimentally verified. The propagation is modelled by successive
crack extensions. The new configuration of the crack after each extension is determined by
the Paris-Erdogan equation within only the SIFs range being taken into account. As a result,
after each extension of the crack, it takes a lot of time again to re-mesh the structure.
However, this problem can be partially overcome by using the macro commands presented in
this paper, since the refinement or the re-meshing is only conducted in the pre-defined
cracked box while the remaining mesh of the structure is not modified. The lifetime of the
structure is the number of loading cycles necessary for the crack to extend from some initial
size to a maximum allowable size. This number is determined by repeatedly integrating the
Paris-Erdogran equation with a numerical integration scheme after each finite element
analysis. Some numerical examples are solved to illustrate the application of these macro
commands. The results are in good agreement with the literature.
2 Creating macro commands for a hexagonal box with an

elliptical crack surface
2.1 Decomposition of the box
We consider first the case of a cracked structure with a
plane of symmetry normal to the crack front. So, only
half of the structure must be meshed. As a result, we
create a meshed box containing a haft surface elliptical
crack as illustrated in Fig. 2.1.
Fig.2.1: Symmetric cracked box
281
In order to have a fine mesh in the vicinity of the

crack front and a transfinite mesh on the outside
faces of the box, the box is decomposed into
domains as shown in Fig. 2.2.
The domain No.1 contains a crack surface. To
use the .FI3 command, which only generates
mesh for domain with a planar crack, the domain
Fig. 2.2: Decomposition of the
No.1 is again decomposed into several smaller
symmetric cracked box
.FI3 boxes. Those .FI3 boxes are defined along a
unique final crack front, shown in Fig 2.4 and are then adjacent. The dimension of the
domain No.1 and the number of these .FI3 boxes are chosen in such a way that the surface
crack can be approximated by planar cracks. Here, to avoid Postfac post-process all these
sub-crack fronts separately, we have to use the .FIA command to modify attributes FISS,
FRAC. As a result, Postfac will only post-process the original crack front. Moreover, the
elements on the interfaces between all sub-domains are merged by the .COL command. Fig.
2.3 shows the mesh of the domain No.1 before and after applying the .COL command.
(a)
(b)
Fig. 2.3: Meshing of the domain No.1 by FI3. command

(a) Before merging elements on interfaces
(b) After merging elements on interfaces
No.4
No.2
No.1
No.3
No.5
Fig. 2.4: Meshing of domains No.2, No.3, No.4, No.5 and No.1
In the Domains No.2, No.3, No.4 and No.5 a transfinite mesh is generated with the .GEN
command. We also use the .COL command to merge the elements on the interfaces between
domains. The elements on the interface between domain No.2 and No.3 are not merged since
this interface is the crack surface. Now, we have a mesh of the domains with a crack surface
as illustrated in Fig. 2.4.
The remaining region of the box is decomposed into 6 domains, No.6, No.7, No.8, No.9,
No.10 and No.11. These domains make the box have transfinite mesh on the outside faces
and the size of the elements increasing gradually when they are far away from the crack front.
282
Then, we use the .COL command to merge elements on the interfaces between these domains
also between these domains and domains No.1, No.4, No.5.
We consider now the full box problem (no
symmetry). As done before, we decompose the
box into domains that have shapes and
positions as shown in Fig. 2.5.
In which, there are
(i)
1 cracked domain that contains a crack

front and meshed by .FI3 command.
Fig. 2.5 Decompose the cracked box into
(ii)
domains
2 sub-domains containing remain
portions of the crack surface are
meshed transfinite by .GEN command and merged with the cracked domains. But the
interfaces belonging to the crack surface are not merged.
(iii)
10 intermediate transfinite meshing domains that mesh the remains of the cracked
box.
2.2 Input data to create macro commands for generating mesh of the
hexagonal box with an elliptical crack surface
Because of the complicated shape of box and the crack surface, a lot of input data are
necessary. The data are defined based on Bacon commands. New commands are named
.FI0Sym and .FI0. The authors wrote a program, called SAMCEF SUPPORT, in C++
language to read a bankfile and return a new bankfile. This program gets data from .FI0Sym
and .FI0 commands to set a batch of correspondent Bacon commands which automatically
mesh the hexagonal box with a 3D surface crack and write then also all Bacon commands to
the new bankfile.
2.2.1 .FI0Sym command

This command is applied for the box having symmetric plane. The set of available parameters
is given below:
.FI0Sym
CRACK FRONT
CRACK BOX
FACE 1 LIGNES
FACE 2 LIGNES
FACE 3 LIGNES
FACE 4 LIGNES
DEGRE
MODIFIE 1 ELEMENTS
DISTRIBUTION
MODIFIE 2 ELEMENTS
DISTRIBUTION
MODIFIE 3 ELEMENTS
DISTRIBUTION
MODIFIE 4 ELEMENTS
DISTRIBUTION
MODIFIE 5 ELEMENTS
DISTRIBUTION
Fig. 2.6: Direction and order of lines created

for the symmetric cracked domain
283
MODIFIE 6 ELEMENTS
DISTRIBUTION
MODIFIE 7 ELEMENTS
DISTRIBUTION
INITIAL POINT
INITIAL LIGNE
INITIAL PLAN
PRECISION
One must declare all values at correct positions as defined-format. Otherwise, SAMCEF
SUPPORT will fail, not inform the error and cannot mesh the symmetric cracked domain.
The followings are defined-formats of the parameters:
CRACK FRONT Number of a line representing the crack front. It is the line number 1 in Fig.
2.6.
CRACK BOX
Number of the 18 lines defining the crack box. It is declared in order from line
number 2 to line number 19 in Fig. 2.6.
FACE 1 LIGNES Number of the 18 lines on the symmetric plan. It is declared in order from line
number 20 to line number 37 in Fig. 2.6.
FACE 2 LIGNES Number of the 12 lines on the surface normal to the symmetric plan and
containing the other crack tip. It is declared in order from line number 38 to line number 49
in Fig. 2.6.
FACE 3 LIGNES Number of the 5 lines on the surface opposite to the symmetric plan. It is
declared in order from line number 50 to line number 54.
FACE 4 LIGNES Number of the 3 lines on the last surface of the symmetric cracked domain. It
is declared in order from line number 55 to line number 57.
DEGRE Set the creation of interface nodes on cell sides. Option is as similar as the one of
Bacon.
MODIFIE 1 to MODIFIE 7 except MODIFIE 4 parameters used in conjunction with other parameters
enables the user to change the characteristics of lines number 38, 26, 28, 2, 4 and 34
respectively. Line characteristics can be a set of elements or a distribution of nodes on lines.
ELEMENTS DISTRIBUTION n r Imposes a number of elements and a distribution of (corner) nodes
on the line present in the MODIFIE number respectively.
n = 1 uniform distribution.
n = 2 distribution according to a geometric progression so that the ratio of the size of the
LAST element by that of the FIRST one should be equal to r.
n = 3 distribution according to a geometric progression so that the ratio of the size of the
FIRST element by that of the LAST one should be equal to r.
INITIAL POINT Declares a point number that is bigger than the maximum point number at
position using this command.
INITIAL LIGNE Declares a line number that is bigger than the maximum line number at position
using this command.
INITIAL PLAN Declares a plan number that is bigger than the maximum plan number at position
using this command.
PRECISION Imposes the precision used in .COL command to merge nodes on interfaces
having the same coordinate. It lets no gap in the domain except the crack surface. If one
doesnt declare this value, the command will use the default precision of Bacon.
Note that the .COL command merges two meshes having the same coordinate nodes to avoid
the presence of gaps or discontinuity of material on the interfaces between domains.
However, the nodes share the same coordinate when distance between them is not larger than
284
the pre-defined precision. This parameter takes the default value of SAMCEF or is input by
the .MODE command. It is better to use the .MODE command to set the precision especially
for complex domains. We have two ways to check whether or not there is a gap between the
interfaces of domains in SAMCEF as follow:
(i)
Using command GRAP PEAU and .DES VI.

If there is a gap, meshing line will appear at this
gaps location. On the contrary, meshing lines
are only at crack surface and out surfaces.
(ii)
Or using command GRAP PEAU and GRAP

ARRET 2; VI. If there is gap, there is a line at
this gaps location. On the contrary, lines
appear only at crack surface and intersections
between out surfaces.
Fig. 2.7: Eight groups of faces

created for the symmetric cracked
b
The .FI0Sym command creates automatically 2 groups

of elements for the crack domain and the crack front named symmetric fiss 1 and group
number 2, respectively. In order to simplify the merging of meshes, there are 6 more groups
of faces with an imposed name from group number 3 to 8 following the rule like Fig. 2.7
2.2.2
.FI0 command
This command is applied for the box with fully elliptical crack surface. The set of parameters
available is given below:
.FI0
CRACK FRONT
CRACK BOX
FACE 1 LIGNES
FACE 2 LIGNES
FACE 3 LIGNES
FACE 4 LIGNES
DEGRE
MODIFIE 1 ELEMENTS
MODIFIE 2 ELEMENTS
MODIFIE 3 ELEMENTS
MODIFIE 4 ELEMENTS
MODIFIE 5 ELEMENTS
MODIFIE 6 ELEMENTS
MODIFIE 7 ELEMENTS
INITIAL POINT
INITIAL LIGNE
INITIAL PLAN
PRECISION
DISTRIBUTION
DISTRIBUTION
DISTRIBUTION
DISTRIBUTION
DISTRIBUTION
DISTRIBUTION
DISTRIBUTION
Fig. 2.8 The direction and order of lines

created the box with full crack surface
Like the .FI0Sym command, users must declare

all values at correct positions as defined-format. If
not, the SAMCEF SUPPORT will fail and not
inform on the error. As a result, the box can not be
correctly meshed. The followings are definedformats of the parameters as .FI0Sym and
numbered as presented in Fig. 2.8.
The .FI0 command creates automatically 2 groups
of elements for the crack domain and the crack
front named fiss 1 and group number 2,
Fig. 2.9 Ten groups of faces created

for the box with full crack surface
285
respectively. In order to simplify the merging of meshes, there are 8 more groups of faces
with an imposed name from group number 3 to 10 following the rule as seen in Fig. 2.9.
2.2.3
Use the SAMCEF SUPPORT
In order to use the SAMCEF SUPPORT program to create macro commands by .FI0Sym and
.FI0 commands, one must follow the steps below:
1. Create a bankfile with all necessary commands, including Bacon, .FI0Sym and FI0
commands, to finish data for the pre-processing stage. Any word processing program
that can export a text file edits this file.
2. Run the SAMCEF SUPPORT program.
3. Choose Export Bankfile menu to select the bankfile wanted to create macro
commands in Open Dialog Box.
4. Input the file name for the bankfile containing macro commands in Save As Dialog
Box.
5. A message will inform on the success if all necessary parameters for the commands
are declared. Particularly, the program only checks the number of parameters in each
command line. If not, the program will inform not success.
6. Run SAMCEF program, add module Bacon and use the created bankfile to import
data for the pre-processing stage of SAMCEF.
Fatigue crack growth calculation
3.1 Hypotheses
y
In this paper, authors assume to know the cracks
B*
shape during fatigue growth. Therefore, a twoparameter model is used: the cracks shape will remain
B
b*
an ellipse (or projective ellipse like example 3) during
b
the propagation. This simplification reduces the
O
x
A
A*
a
number of geometrical parameters to two. It means that
a*
the crack front with semi-axes a and b (Fig. 3.1) grows
Fig. 3.1: Two parameters of the
after one cyclic loading step to the new configuration
crack shape
described by the following expression
y2
x2
(3.1)
+
=1
(a*) 2 (b*) 2
where the two unknowns a* and b* can be obtained by the condition that the coordinates of
points A* and B*, deduced from the Paris law, must satisfy equation (3.1).
The main fracture mode is usually the mode I due to the modes II, III being very small. As a
result, in the paper, we assume that the crack propagates in the tangent plane located at the
crack front according to mode I.
286
3.2 The Paris law
Begin
N = 0; a0; b0; bf, Kcr
min, max
To compute the propagation rate of the crack,

we use the Paris law:
da
= C (K eff ) m
dN
-BACON
-ASEF
-POSTFAC
(3.2)
where C and m are material properties, a is the

crack length, N is the number of loading
cycles and Keff is the effective range of SIF
defined as:
Suppose b0
calculate a0
K eff = U .K = U ( K max K min )
-BACON
-ASEF
-POSTFAC
U=
a = a + a
b = b + b
Program
KImin, KImax at points A and B
Calculate N
()
Suppose b
calculate
No
3.3
()
a = a0 + a0
b = b0 + b0
where Kmax and Kmin are the SIFs

corresponding to the maximum (max) and
minimum (min) applied loads respectively,
and U is defined as:
1
; R 0.5
1.5 R
U = 1;
R > 0. 5
K
R = min
K max
KImin, KImax at points A and B
b = bf or
K = Kcr
Yes
Calculate SIFs: Using three modules of
SAMCEF:
Ntotal = N
End
Module Bacon: pre-processing for module
Asef.
Fig. 3.2: Program of fatigue calculation
Module Asef: elastic analysis of the structure.
Module Postfac: evaluate the energy release rate by direct or VCE methods from the results
given by module ASEF.
Each elastic analysis gives the SIFs along the crack front. So we can determine the crack
propagation rate for each point of the crack front. We re-mesh the structure (only the box is
changed), compute the new SIFs and so on. The program of fatigue calculation is described
in Fig. 4.2 where the steps () and () are explained in more details as follows:
Step ():
-Suppose bi
da
db
-Calculate ( ) ai and ( ) bi by using equation (3.2)
dN
dN
da
da
-Calculate the rate ( ) ai ,bi ai = ( ) ai ,bi bi
db
db
Step ():
The number of loading cycles required to extend the crack from (i-1)-size to (i)-size (ai-1 ai
and bi-1 bi) is evaluated by integrating equation (3.2) with a trapezoidal rule.
287

ai ai 1
1
1
[
]
+
da
da
2
( ) ai 1 ( ) ai
dN
dN
b b
1
1
]
N iB = i i 1 [
+
db
db
2
( ) bi 1 ( ) bi
dN
dN
N iA =
N i =
N iA + N iB
2
The number of loading cycles: Ni = Ni-1 + Ni
3.4
3.4.1
Numerical example
b0 = 1.0 mm
a0 = 2.0 mm
2W = 80 mm
2h = 240 mm
t = 4 mm
2h
B
E = 201000 N/mm
= 0.3
2a A
Example 1
max = 500 MPa

min = 100 MPa
A structure with a semi-elliptical surface crack under

2W
t
tension (Fig. 3.3) is discretized into finite elements.
Because of symmetry, only an half of the problem is
Fig. 3.3: Plate with a semi-elliptical
studied. To compare with Ref. [3], instead of
surface crack under tension
equation (3.2) we use here the propagation law
proposed by Ref. [6]. By this law, the crack grows only when Keff is bigger than Kth.
da
= C[( K eff ) m ( K th ) m ]
dN
K m
= C[(
) ( K th ) m ]; R 0.5
1.5 R
= C[( K ) m ( K th ) m ]; R > 0.5
2.5
10
8 9
6 7
4 5
2 3
1
1.5
(3.3)
where Kth represents the threshold of
propagation. The material constants in
equation (3.3) are the same as Ref. [3] for
comparison:
C = 3.0 10 13
1 : 0 cycles
2 : 2586 cycles
3 : 4953 cycles
4 : 7130 cycles
5 : 10088 cycles
6 : 12718 cycles
7 : 15804 cycles
8 : 18488 cycles
9 : 20829 cycles
10 : 23810 cycles
0.5
0.5
1.5
3.5
Fig. 3.5: The crack shape at different numbers

of loading cycles
mm
( MPa mm ) 3
cycle
m = 3 Kth
in MPa mm )
190-144R
(unit
Fig. 3.4 shows the numbers of loading

cycles versus the crack depth b which is
in good agreement with Ref. [3]. The
total number of cycles is obtained when
the crack depth reaches the final value of
bf = 2.5 mm. The shapes of crack at some
important numbers of loading cycles are
presented in Fig.3.5.
Fig. 3.4: Comparison of our results
and the results of Ref. [3]
288
3.4.2
Example 2
We aim at calculating SIF of a round pipe containing external circumferential elliptical crack
under tension. The tension is 100 N/mm2. Material properties are: Young modulus E =
201000 N/mm2, Poision ratio = 0.3. Crack length 2a is 5.0mm. Crack depth b is 1mm. For
the geometrical dimensions, see Fig. 3.6.
Fig. 3.7: Boundary conditions and

mesh of the cracked round pipe
Fig. 3.6: Round pipe containing an external

circumferential elliptical crack
180
stress intensity factor K1
160
140
120
100
.VIK
.VCE
80
60
.EDI
Ref. [2]
40
20
0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
curvilinear coordinate along crack front
Fig. 3.8 SIF mode I obtained by local,

global methods and Ref. [2]
Due to symmetry, only half of the pipe is modelled. Fig. 3.7 shows the boundary conditions
and the mesh of the pipe obtained with the macro command for cylindrical box with crack
plane.
The result of calculating SIF mode I of the pipe is displayed in Fig. 3.8. In that, .VIK
calculates the SIFs distribution along crack front using local method based on the
displacement field near the crack tip. The .VCE performs a direct calculation of the energy
release rates according to the so-called virtual crack extension technique. The .EDI directly
computes the J-integral according to the so-called equivalent domain integral [5]. The
obtained results are in good agreement with Ref. [2]. This means that the mesh is good
enough.
289
3.4.3
Example 3
The tubular T-joint model is shown diagrammatically in Fig. 3.9. The chord is 3900mm
length, 914mm outside diameter and 32mm wall thickness. The length, outside diameter and
wall thickness of the brace are 1310, 457, 16 in mm respectively. The two ends of the chord
are restrained rigidly in all directions. The joint is axially loaded with a force of 470kN in the
brace direction. The properties of elastic material are Youngs modulus E = 2.1x1011Pa and
Poissons ratio = 0.3. The values of the C and m parameters are
3.34 x10 13
mm
( MPa mm ) 2.92 and 2.92 respectively.
cycle
When the joint was subjected to this fluctuating axial force, it was found in the test that
fatigue crack initiated at the weld toe on the chord close to the saddle point, and propagated
through the chord wall. Noordhoek and Verheul [4] measured the development of fatigue
crack growth by beach marking technique and geometry of the surface crack is shown in
Fig.3.10.
470kN
15
20
Dimensions in mm
1310
51
914
crack location
457
crack profile
Fig. 3.9: Diagram of tubular joint,

showing dimensions and crack location
32
3900
Dimensions in mm
Fig. 3.10: Typical weld profile at the

saddle point hot spot
The beach marks on the crack surface show that the profile of the crack has an elliptic shape.
The crack tip and crack depth points are the quarter points of the ellipse. Actually, the crack
front is not a plane semi-ellipse. It is a curve on a cone crack surface. We can assume that the
crack front is the projection of an ellipse on a tangential plane with the crack surface.
Therefore, we use the .FI0 command to mesh the hexagonal box containing a semi-elliptical
crack surface because of symmetry. Fig. 3.11 shows the mesh around the crack obtained with
Bacon and Fig. 3.12 displays the principle I stress along in the vicinity of the crack front
computed with Asef.
Fig. 3.11 Meshing around crack
Fig. 3.12 The principle I stress

around the crack front
290
The SIFs results of each crack extension are used to predict the fatigue crack growth with the
Paris-Erdogan law. The initial dimensions of the crack are 6.5mm of depth and 30.2mm of
length.
As presented, we assume that the crack front is a semi-ellipse curve defined by two points:
crack depth and crack tip. Therefore, we only calculate the propagation of these two points.
Since the problem includes mixed modes, the following formula is applied:
K eff = K I2 + K II2 + (1 + )K III2
K III2
1
K eff = K I2 + K II2 +
and
for plane stress condition

for plane strain condition
As the minimum value of cyclic axial loading is equal to zero, the stress intensity factor range
of the individual modes is given by K = K
max
The VCE method with perturbation 0.00001 is used to compute SIFs. The lifetime calculation
is based on the program described in Fig.3.2. We obtain the number of cycles corresponding
to the dimension of crack tip and crack depth as shown in Table 4.1.
Case
Crack
depth
(mm)
Surface
length
(mm)
6.5
Cycles (x 106)
Error
(%)
Calculated
Refer. [4]
30.2
0.883
0.883
0.0
11.2
60.9
1.083
1.077
0.5
14.3
91
1.204
1.182
1.9
18.0
123.3
1.295
1.285
0.8
24.2
202.2
1.418
1.448
2.1
Table: 4.1 Dimensions of the cracks and their cycles

1500000
Number of cycles
1400000
1300000
1200000
1100000
Calculated
Ref. [4]
1000000
900000
800000
5
10
15
20
25
Crack depth (mm)
Fig. 3.13: Growth of the crack in depth

291
Number of cycles
1500000
1400000
1300000
1200000
1100000
Calculated
1000000
Ref. [4]
900000
800000
0
50
100
150
200
250
Surface length measured from the initiation point (mm)
Fig. 3.14: Growth of the crack along the surface
Growths of the crack in depth and along surface are illustrated in Fig. 3.13 and Fig. 3.14.
The charts show the rapid surface growth of the crack that agrees with the observed crack
growth. The agreement shown in these figures between the calculated and experimented
cycles at the deepest and surface point of the crack is very good. However, it is not
reasonable when the crack grows more deeply than 20mm because of not modelling precisely
the crack front on the crack surface.
Conclusions
The following conclusions are drawn:

1. Numerical results proved the efficiency of the mesh with the aid of the box having a
crack. This means that the method that decomposes the box into sub-domains is good
because it generates the fine mesh in the vicinity of the crack front, coarse mesh far
out in the remaining regions. Moreover, the transfinite meshes on the outside surfaces
of the box help the users to easily connect the meshed box with the rest of the meshed
structure.
2. The macro commands reduce time-consuming for generating mesh around a crack
and also overcome the complexity of the re-meshing cracked area after each extension
of the crack during propagation period.
3. The agreement between numerical results and the referenced ones proved the
reliability of our program to compute lifetime under fatigue based on Paris-Erdogan
equation with a two-parameter model.
However, it is lightly complicated to use the macro commands and the fatigue computation is
not automatically carried out because users can only work with SAMCEF through Bacon
commands.
292
References
[1]
[2]
[3]
[4]
[5]
[6]
A. Carpinteri, R. Brighenti, Circumferential surface flaws in pipes under cyclic axial loading,
Engineering Fracture Mechanics, Vol. 60 No.4, 383-396, 1998.
M. A. Hussain, L. F. Coffin, K. A. Zaleski, Three dimensional singular element, Computer &
Structures, Vol. 13, 595-599, 1981.
Bodeux Jean-Bernard, Modlisation et simulation de la croissance dendommagement et de fissuration
des structures mtalliques et composites soumises la fatigue et en prsence des contraintes
rsiduelles, Rapport final, Projet COST 512, LTAS-Mcanique de la Rupture, Universit de Lige.
D. Ritchie, C. W. M. Voermans, Stress intensity factors in an offshore tubular joint test specimen,
Koninklijke/Shell-Laboratorium, P.O. Box 3003, 1003 AA Amsterdam, The Netherlands.
SAMTECH, Samcef Fracture mechanics 6.1, User Manual M16-V6.1-96.
M. Klesnil, P. Lukas, Influence of strength and stress history on growth and stabilization of fatigue
cracks, Engineering Fracture Mechanics, Vol. 4, p. 141, 1972.
293
Influence of Moment-Shear Interaction to T-stub Effect

in Bolted Column-Beam Joints
Nguyen-Thanh Son*, Nguyen-Van Yen+ and Bui-Cong Thanh+
*Department of Project Management,
Indochina Capital, HCMC, Vietnam
+Faculty of Construction Technology,
University of Technology of HCMC, Vietnam
Abstract
The influence of the moment-shear interaction in formulations derived for computing the strength
of T-stub effects is studied in this paper. The proposed model uses the yield criteria of Drucker
(1956) for determining the strength of the T-stub flange. The approximate solutions of transcendent
equations describing the problem are presented. The obtained numerical results are verified by
comparison with experimental tests conducted by Swanson & Leon (1998) to evaluate the effective
level of the moment-shear interaction to T-stub effect.
Keywords: bolted joints, beam-column, t-stub effect, strength model, moment-shear interaction.
Introduction
The T-stub effect is an important factor that needs to be considered in bolted steel connections
(Gaylord et al., 1992; Owens et al., 1994; Bursi & Jaspart, 1998; Faella et al., 2000; Coelho et al.,
2002). Several analytical methods were introduced to simulate and derive the strength of T-stub
effect (AISC, 1989; AISC, 2002; BSI, 1990; Eurocode 3, 1992). However, all those methods that
compute the design resistance corresponding to Type 1 (the full yield of T-stub flange) and Type 2
(the yield of flange and bolt failure) mechanisms ignored the moment-shear interaction. Recently,
Fealla et al. (2000) introduced an approximate approach based on Hencky-Von Mises criteria (see
Nguyen, 1996) to determine this interaction influence on the strength of T-stub effect and the
authors assumed that the external fibres withstand the bending moment while internal ones are
subjected to shear stress only. According to the static theorem of plastic collapse, this assumption
leads to safe results.
This paper will propose a new approach based on the yield criteria of Drucker (1956) to evaluate
the influence of the moment-shear interaction on the design resistance of T-stub flange. It will be
seen that, depending on the geometry and the mechanism type, the moment-shear interaction
influence on the design resistance of T-split action can be neglected or not.
Moment-Shear Relation
In accordance with the Drucker (1956) yield criteria, yield limit occurs for an element subjected
to moment and shear when the moment-shear relationship satisfies the following condition:
Mf
M f0
Vf
=1
+
V
f0
(2.1)
294
In which:
Mf : the plastic bending moment of T-stub flange corresponding to
shear Vf.
tf
: the thickness of T-stub flange.
bf
: the width of T-stub flange.
fyf : the yield stress of T-stub flange.
M0 : partial safety factor of base material ( M 0 = 1.1 ).
Mf0 : the pure plastic bending moment of T-stub flange.
b f t 2f f yf
M f0 =
(TCXDVN 356, Eurocode 3)
4 M0
Vf0 : the pure shear of T-stub flange.
f yf
(TCXDVN 356, Eurocode 3)
Vf 0 = bf t f
3 M 0
(2.1a)
(2.1b)
From this criterion, the design resistance of T-stub flange will be derived for Type-1 and Type-2
mechanisms that take into account the moment-shear interaction.
Combining Eq. (2.1a) with (2.1b), we obtain:
3
M f0 =
V f 0t f
4
(2.2)
Type-1 Mechanism (Full Flange Yielding)
The prying force model is described in Fig. 1. From this figure, the shear of T-stub flange is
calculated by the equilibrium force equation:
Vf =
ntb
BRd Q =
F1. Rd
2
(2.3)
From Fig. 1, the equilibrium equation of forces can be given by:

Tm = (1 + ) M 1 + ntb M b
(2.4)
Where:
Mb
: the plastic moment of bolt itself (Wheeler et al., 2003).

Mb =
d b2 f ub
32 M 0
M b b /( p d hb ) M b b
=
M aa / p
M aa
p=
M b b
1
=
1 d hb / p M a a
2 LT stub
ntb
(2.4a)
(2.4b)
(2.4c)
295
Figure 1: The prying model of mechanism Type-1

LT-stub
dhb
db
Q
Ma-a
Mb-b
M1
ntb
aw
:
:
:
:
:
:
:
:
:
the length of flange.

the diameter of bolt hole.
the diameter of bolt.
the prying force.
the moment at web edge section.
the moment at bolt action section.
the moment at hinge number 1.
total number of bolts.
height of weld connecting to web and flange.
d hb
= (flange area at bolt hole section)/ (flange area at web edge).
p
g t
m= t w r
, roll section
2
= 1
m=
gt tw
aw 2
2
n = emin
1.25m
, welded built-up section
(2.4d)
(2.4e)
(2.4f)
296
Replacing Eq. (2.3) into Eq. (2.4), the plastic moment of the T-stub flange can be computed:
n
(1 + ) M f + tb M b
2
Vf =
m
ntb
Vf m
M
2 b
(2.5)
Mf =
1+
Replacing Eq. (2.2), (2.3) and (2.5) into Eq. (2.1), the yield criterion can be written as follows:
4
n
V f V f m tb 2 M b
+
V
1+
f0
4
=1
3V f 0 t f
4
ntb
M
Vf
Vf
m
4
4
2 b
= 1+
V
3 (1 + ) V f 0 t f
3 (1 + ) t f V f 0
f0
(2.6)
Because the accurate roots of the 4th rank-transcendent equation (2.6) still havent been found
out, this paper proposes an approximate solution of this equation:
4(1 + C3 )
1
(2.7)
V f = C1C2 m 1 +
2
C
m
2
where:
C1 = 0.548V f 0
(2.7a)
4
3 (1 + )
m
m=
tf
C2 =
(2.7b)
(2.7c)
ntb
C3 = C 2
M
2 b
t fVf 0
(2.7d)
The proposed root has average accurate percentage of +4.77% with a standard deviation of
0.037. As a result, it can be said that the root of Eq. (2.7) is acceptable.
Therefore, from the equilibrium condition (F1.Rd.V = 2Vf) of Eq. (2.3), the design resistance of
the T-stub flange is obtained:
4(1 + C3 )
F1. Rd ,v = 2C1C2 m 1 +
1
2
C
m
(2.8)
To evaluate the effected level of the moment-shear interaction to the design resistance of the Tstub flange, the following ratio is used:
v1 =
F1.Rd ,v
F1. Rd ,s
(2.9)
297
where F1.Rd,s is the design resistance of the flange related to the mechanism of type-1 computed by
neglecting the moment-shear interaction.
By means of Eq. (2.8), the influence of the moment-shear interaction on the design resistance of
mechanism type-1 is illustrated in Fig. 2. We observe that:
1 m / t f < 1.5 : the reduction of the design resistance of the type-1 mechanism due to shear
influence is less than 20%.
m / t f 1.5 : the reduction of the design resistance of the type-1 mechanism due to shear
Therefore, it can be concluded that the moment-shear interaction can be ignored when m / t f is
large enough ( m / t f 1.5 ) and an accurate value of the design resistance of flange is not required.
Type-2 Mechanism (flange yielding with bolt failure)
Similarly, the plastic moment at hinge 1 of this mechanism can be computed in accordance with
the proposed prying model illustrated in Fig. 3:
ntb
n
BRd m Q(m + n) tb M b
2
2
ntb
n
BRd m M f tb M b
2
Q= 2
m+n
Mf =
(2.10)
Consequently, from the equilibrium condition (F2.Rd = 2Vf), the following relationship is
obtained:
ntb
(BRd n + M b ) + M f
Vf = 2
(2.11)
m+n
Therefore, at the ultimate limit state, the plastic moment related to shear is given by:
n
M f = V f (m + n) tb (BRd n + M b )
2
(2.12)
Combining Eq. (2.2) and (2.12), we have:

Mf
M f . Rd
Mf
M f . Rd
=4
ntb
(BRd n + M b )
2
3V f .Rd t f
V f ( m + n)
V
4 m
(1 + ) f ntb BRd n ntb M b
V f .Rd 2 M f .Rd 2M f .Rd
3 tf
Set:
Rd =
2(1 + ) M f + ntb M b
m(ntb BRd )
(2.13)
(2.14)
298
ntb BRd n (1 + )
n M
=
+ tb b
Rd
2M f
2 Rd M f
(2.15)
Replacing Eq. (2.15) into (2.13), we have:
Mf
M f .Rd
V
4 m
(1 + ) f (1 + ) + ntb M b ntb M b
V f . Rd Rd
2 Rd M f 2M f .Rd
3 tf
(2.16)
Replacing Eq. (2.16) into Eq. (2.1), the yield criterion with moment-shear plastic domain can be
written in the following relationship:
4
V
+ 4 m (1 + ) f = 1 + (1 + ) + ntb M b + ntb M b
V f .Rd
2 Rd M f 2 M f .Rd
3 tf
Rd
(2.17)
F1.Rd (KN)
Comparison F1.Rd,s<> F1.Rd,v

6,000
5,000
4,000
F1.rd,s
F1.rd,v
3,000
2,000
1,000
2.3
2.56
2.06
2.06
2.06
1.9
1.9
1.6
1.6
1.56
1.5
1.56
1.1
1.1
1.07
1.07
1.06
1.06
1.06
1.06
m/tf
Moment-shear interaction effect

1.20
1.00
0.80
0.60
Effected percentage
0.40
0.20
2.56
2.30
2.06
2.06
2.06
1.90
1.90
1.60
1.60
1.56
1.56
1.50
1.10
1.10
1.07
1.07
1.06
1.06
1.06
0.00
1.06
F1.Rd,v/F1.Rd,s
Vf
V
f .Rd
m/tf
Figure 2: Moment-shear interaction influence on mechanism type-1

299
In Figure 2, F1.Rd,s and F1.Rd,v are the resistance of T-stub flange for Type-1 mechanism without
and with moment-shear interaction, respectively.
Figure 3: The prying model of mechanism Type-2

Similarly to Eq. (2.6), Eq. (2.17) is also a 4th rank-transcendent equation. An approximate root is
given by:
V f = C 4 C5 m 1 +
1 + C6
(2.18)
2
C
m
5
In which:
C4 = 0.6051V f 0 = 0.6051V f .Rd
C5 =
4
(1 + )
3
(2.18a)
(2.18b)
(1 + )
n M
n M
C6 =
+ tb b + tb b
2 Rd M f 2 M f .Rd
Rd
(2.18c)
The proposed root has an average accurate percentage of +1.16% with a standard deviation of
0.011. As a result, it can be said that the root of Eq. (2.17) is acceptable.
Therefore, from the equilibrium condition (F2.Rd.V = 2Vf), the design resistance of the T-stub
flange is obtained:
4(1 + C6 )
F2. Rd ,v = 2C 4 C5 m 1 +
(2.19)
1
2
C
m
To evaluate the effected level of the moment-shear interaction to the design resistance of a Tstub flange, the following ratio is used:
F
v 2 = 2.Rd ,v
(2.20)
F2.Rd ,s
where F2.Rd,s is the design resistance of the flange related to the type-2 mechanism computed by
neglecting the moment-shear interaction.
300
By means of Eq. (2.19), the influence of the moment-shear interaction on the design resistance
of the type-2 mechanism is illustrated in Fig. 4. It can be observed that:
1 m / t f < 1.5 : the reduction of the design resistance of the type-2 mechanism due to shear
m / t f 1.5 : the reduction of the design resistance of the type-2 mechanism due to shear
In addition, the values of m / t f 2.0 have been often selected in practical design. So, the
reduction of the flange design resistance is always less than 4%. As a result, the influence of
moment-shear interaction on the design resistance of the type-2 mechanism can always be
neglected.
F2.Rd (Kip)
Comparison F2.Rd,s<> F2.Rd,v

4,000
3,500
3,000
2,500
F2.rd,s
F2.rd,v
2,000
1,500
1,000
500
2.3
2.56
2.06
2.06
2.06
1.9
1.9
1.6
1.6
1.56
1.5
1.56
1.1
1.1
1.07
1.07
1.06
1.06
1.06
1.06
m/tf
F2.Rd,v/F2.Rd,s
Influence of moment-shear interaction

1.02
1.00
0.98
0.96
Effected percentage
0.94
0.92
0.90
2.56
2.30
2.06
2.06
2.06
1.90
1.90
1.60
1.60
1.56
1.56
1.50
1.10
1.10
1.07
1.07
1.06
1.06
1.06
1.06
0.88
m/tf
Figure 4: Moment-shear interaction influence on mechanism type-2

In Figure 2, F2.Rd,s and F2.Rd,v are the resistance of T-stub flange for Type-2 mechanism without and
with moment-shear interaction, respectively.
301
Table 1: Moment-shear interaction influence on the design resistance of the flange

Test ID Experiment
(1)
Capacity
(kN)
(2)
AISC-LRFD (2002)
without shear interaction

with shear
v
Capacity
% Diff
Capacity
% Diff
Capacity
% Diff
Capacity
% Diff
(kN)
(kN)
(kN)
(kN)
(3)
(4)=(3)/(2)
(5)
(6)=(5)/(2)
(7)
(8)=(7)/(2)
(9)
(10)=(9)/(2) (11)=(9)/(7)
TA-01
TA-02
TA-03
TA-04
TA-05
TA-07
2,099.8
2,044.7
2,020.9
1,742.3
2,120.4
1,944.2
2,361.11
1,814.47
1,399.73
1,139.32
2,361.11
1,399.73
1.12
0.89
0.69
0.65
1.11
0.72
2,184.76
1,841.38
1,591.28
1,317.27
2,184.76
1,591.28
1.04
0.90
0.79
0.76
1.03
0.82
2,436.37
2,053.11
1,680.48
1,352.59
2,436.37
1,680.48
1.16
1.00
0.83
0.78
1.15
0.86
2,298.62
1,994.03
1,627.97
1,361.97
2,298.62
1,627.97
1.09
0.98
0.81
0.78
1.08
0.84
0.94
0.97
0.97
1.01
0.94
0.97
TA-10
TA-11
TA-12
TA-15
1,910.7
1,903.6
1,900.9
1,905.3
1,867.33
1,421.40
1,147.40
1,421.40
0.98
0.75
0.60
0.75
2,094.99
1,617.50
1,317.27
1,617.50
1.10
0.85
0.69
0.85
2,242.47
1,698.84
1,367.36
1,647.66
1.17
0.89
0.72
0.86
2,030.82
1,648.75
1,378.65
1,604.05
1.06
0.87
0.73
0.84
0.91
0.97
1.01
0.97
TA-17
TA-18
TA-19
TA-20
1,413.0
1,306.8
1,153.2
1,051.5
1,291.55
1,168.84
1,028.95
889.06
0.91
0.89
0.89
0.85
1,252.47
1,055.62
912.24
803.15
0.89
0.81
0.79
0.76
1,372.67
1,167.29
1,002.88
879.07
0.97
0.89
0.87
0.84
1,344.14
1,149.72
1,000.24
884.18
0.95
0.88
0.87
0.84
0.98
0.98
1.00
1.01
TB-05
TB-06
TB-07
TB-08
2,252.5
2,225.7
2,226.1
2,212.2
2,616.78
1,841.44
2,599.42
1,841.44
1.16
0.83
1.17
0.83
2,668.18
2,169.91
2,407.35
2,035.47
1.18
0.97
1.08
0.92
2,947.40
2,341.01
2,650.64
2,241.62
1.31
1.05
1.19
1.01
2,743.16
2,123.94
2,498.58
2,082.12
1.22
0.95
1.12
0.94
0.93
0.91
0.94
0.93
TB-09
TB-10
1,411.6
1,241.4
1,272.58
1,097.88
0.90
0.88
1,245.60
1,049.04
0.88
0.85
1,372.67
1,167.15
0.97
0.94
1,351.51
1,152.29
0.96
0.93
0.98
0.99
TC-01
TC-02
TC-03
TC-04
TC-05
TC-07
2,619.5
2,353.8
2,097.5
1,868.2
2,434.4
1,999.4
2,631.90
2,389.34
2,138.09
1,886.84
2,351.10
1,950.02
1.00
1.02
1.02
1.01
0.97
0.98
2,664.79
2,286.14
2,001.70
1,713.62
2,351.10
1,866.39
1.02
0.97
0.95
0.92
0.97
0.93
2,716.67
2,512.55
2,187.30
1,936.60
2,351.10
2,025.59
1.04
1.07
1.04
1.04
0.97
1.01
2,716.67
2,474.38
2,179.19
1,945.06
2,351.10
2,023.62
1.04
1.05
1.04
1.04
0.97
1.01
1.00
0.98
1.00
1.00
1.00
1.00
TC-09
TC-10
TC-11
TC-12
TC-13
TC-15
2,970.2
2,919.6
2,601.5
2,296.4
2,836.7
2,313.5
3,206.84
2,913.02
2,591.16
2,269.31
2,834.01
2,301.46
1.08
1.00
1.00
0.99
1.00
0.99
3,076.09
2,669.33
2,340.96
1,928.50
2,785.54
2,123.05
1.04
0.91
0.90
0.84
0.98
0.92
3,397.96
2,929.99
2,574.84
2,296.48
2,935.30
2,318.13
1.14
1.00
0.99
1.00
1.03
1.00
3,275.34
2,874.46
2,555.68
2,298.46
2,935.30
2,310.19
1.10
0.98
0.98
1.00
1.03
1.00
0.96
0.98
0.99
1.00
1.00
1.00
0.99 Average
0.13 Stad. Dev
0.97
0.11
0.98
0.03
Average
Stad. Dev
The proposed approach
Eurocode 3 (1992)
0.93 Average
0.14 Stad. Dev
0.92 Average
0.11 Stad. Dev
Conclusions
The study presented in this paper showed that a solution procedure for computing the momentshear interaction influence on the design resistance of T-stub flange is now available. The yield
criterion of Drucker was used to evaluate the interaction and the approximate roots were proposed
to solve the 4th rank transcendent equations indicated that:
For the type-1 mechanism, the reduction of the T-stub flange design resistance due to the
moment-shear interaction is less than 10% compared with the theoretical design resistance if the
values m / t f are not small ( m / t f 1.5 ).
For the type-2 mechanism, the reduction of the T-stub flange design resistance due to the
moment-shear interaction is less than 7% compared with the theoretical design resistance for all
values m / t f .
Based on the verification between the obtained numerical results and experimental tests
conducted by Swanson & Leon, 1998 (Table 1) and the comparison with AISC (2002) and
Eurocode 3 (1992), it can be observed that:
302
There is a very good agreement between the obtained numerical and experimental results.
The proposed results are more accurate than the ones computed by AISC, 2002 and Eurocode 3,
1992 (-3.17% against -7.40% and -8.41%, respectively).
When the influence of the moment-shear interaction is considered on the design resistance, the
reduction of the T-stub flange design resistance is more than 2% with a standard deviation of
0.03.
Therefore, it can be concluded that the moment-shear interaction influence on the design
resistance of T-stub flange is available. However, it is too small to be ignored.
Finally, the author would like to acknowledge Prof. Leon and Dr. Swanson (Georgia of
Technology Institute) who provided experimental results.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
AISC, Manual of Steel Construction Allowable Stress Design, 9th ed., American Institute of Steel
Construction, Chicago, Illinois, 1989.
AISC, Manual of Steel Construction Load and Resistance Factor Design, American Institute of Steel
Construction, Chicago, Illinois, 2002.
B xy dng, TCXDVN 356, Kt Cu Thp Tiu Chun Thit K, Nh xut bn Xy Dng, H Ni, 2005.
BSI, BS 5950: Structural Use of Steelwork in Building Part 1, British Standards Institution, London, 1990.
Bursi, O.S., and Jaspart, J.P, Calibration of a Finite Element Model for Isolated Bolted End-Plate Steel
Connections, Journal of Constructional Steel Research, Elsevier Science Ltd , Vol. 44, No. 3, pp. 225-262,
1998.
Coelho, A.M.G., and Bijlaard, F., Silva, L.S., On the Behaviour of Bolted end Plate Connections Modelled by
Welded T-stubs, Proceedings of Third European Conference on Steel Structures, Euro Steel, Vol. II, Ed.
Lamas.A., and Silva. L.S, Coimbra, Portugal, pp. 907-918, Sep-2002.
Drucker, D.C. (1956), The effect of Shear on the Plastic Bending of Beams, Journal of Applied Mechanics ,
ASME, Vol. 23, No.4, pp. 509-514, 1956.
ENV 1993, Eurocode 3: Design of Steel Structures: Part 1.1 General Rules and Rules for Buildings, Comit
European de Normalisation (CEN), Brussels, Belgium, B-105 Brussels, 348 pp, 1992.
Faella, C., Piluso, V., and Rizzano, G., Structural Steel Semirigid Connections, JCRC Press, 2000
Gaylord, E.H., Gaylord, C.N., and Stallmeyer, J.E., Design of Steel Structures, McGraw-Hill International, 3rd
edition, Singapore, pp. 813, 1992.
Nguyen-Dang Hung., Advanced Metal Mechanics, the lecture of EMMC program, HCMC, Vietnam, October,
1996.
Owens, G.W., Knowels, P.R., and Dowling, P.J., Steel Designers Manual, 5th ed, The Steel Construction
Institute, Blackwell Scientific Publications, London, 1266 pp,1994.
Swanson. J.A., and Leon. R.T., SAC Steel Project T-stub, Clip Angles, and Stiffened Seat Connections,
Subtask 7.03, The FEMA Program, Georgia Institute of Technology, 1998.
Wheeler. A., Clarke. M., and Hancock. G.J., Design Model for Bolted Moment End Plate Connections Joining
Rectangular Hollow Sections Using Eight Bolts, Dept. of Civil Engineering, Uni. of Sydney, Report No. R827,
2003.
303
Determination of the Fatigue Characteristics

of Cracks by an Inverse Analysis
Hang-Minh Dinh * and Nguyen-Dang Hung+
EMMC, Ho Chi Minh City University of Technology, Vietnam
+
*
Abstract
A new way is proposed to determine the material constants of fatigue based on Paris law and
Circumferentially Cracked Round Bar specimens (CRB). And the accurate analytical formulas for
stress intensity factor of CRB specimens are necessary to this determination. This is carried out by
finite element calculations and literature review. Finally, a numerical application is performed for
the material 22NiMoCr32.
Keywords: stress intensity factor, finite element analysis, fracture mechanics, inverse analysis,
fatigue, cracked round bar specimen.
Introduction
Determination of the fracture properties of material, such as toughness and the fatigue constants, are
extremely necessary for evaluation of the lifetime of structures. Nevertheless, to perform
experiments for the determinations is a great challenge in both techniques and finance.
Scibetta[6] proposed the use of the small circumferentially Cracked Round Bar for the
measurement of fracture toughness. There are some advantages on using this type of specimen, such
as: the presumed weaker size requirement, the low cost to machine the specimen, the rotating
bending fatigue allowing for easy pre-cracking of specimens, the use of a standard tensile test
fixture and the axisymmetry of the specimen that avoids time consuming three-dimensional finite
element calculations.
Furthermore, it is found that from the results of pre-cracking the CRB specimens, a relation
between the fatigue constants C and m, based on the Paris law, is built. Since then, these constants
will be determined.
Analysis
2.1 Preparation
Stress intensity factor for CRB
In order to apply the Paris law for determination of the fatigue constants, the knowledge of the
formula to obtain the maximum stress intensity factor of a CRB under bending is required. For this
reason, the available in the literature are now investigated.
304
Maximum stress intensity factor (contact of the crack lips is not taken into account)
Benthem and Koiter[2] used an approximate analytical method to propose a formula of the
maximum stress intensity factor of a bar under a bending moment:
b
b
K = G b1
R
R
with the maximal stress =
(1)
4M
b 3
(2)
2
3
4
5
1 b 3 b
5 b
35 b
b 3
b
and G = 1 +
+ + +
+ 0.531
2 R 8R
16 R 128 R
R 8
R
(3)
where, 2b is the ligament diameter

2R is the bar diameter (see Figure 1)
a
2R
2b
2c
a
l
L
P
Figure 1: Geometry of cantilever beam configuration

Harris[3] proposed another relation:
K=
so,
b
G =
R
b(R b )
0.8b + 7.12(R b )
1
b
b
0.8 + 7.121
R
R
(4)
(5)
The attempts in three dimensional finite element analysis are performed to verify the above
formulas. By the aid of SAMCEF V8.1, the calculation becomes easier. A typical meshing is
illustrated in Figure 2 (approximately 4800 3D elements), in which the singular isoparametric
elements (Barsoum[1] elements) are used for modeling the crack tip. Due to symmetric condition
only half the structure needs to be modeled.
305
Figure 2: Meshing and results

After the linear static analysis, a direct calculation of energy release rates GI, GII, GIII is
performed by SAMCEF according to the so-called equivalent domain integral technique
(proposed by Nikishkov[4]). For linear elastic problems, the stress intensity factors KI, KII, KIII can
easily be obtained from the energy release rates with the following formulas:
K I , II =
1
2
8E'
+1
G I G II G III G I + G II G III
(6)
K III = 2E' G III
E
2(1 + )
with
E' =
and
= 3 4 (plane strain)
3
(plane stress)
1+
E and are Young modulus and Poisson factor.
The dimensionless stress intensity factor G is established from the finite element solution as
follows:
2
b
b
b
b
b 3
G = + 0.05356 + 1.06688 1.91918 + 1.57057
R
R
R
R
R 8
(7)
The three equations (3), (5) and (7) are compared in figure 3. The maximum difference between
(3) and (5) is 3.0 % for a/R = 0.231.
306
G BEN&KOI
G FE
G HAR
a/R
Figure 3: Stress intensity factor comparisons.

G is the dimensionless intensity factor defined by eq. (3), (5) and (7).
In the cantilever configuration, the bending moment is not uniform and a shear loading exists, so
the mode II is present. This mode induces an angle deflection (Figure 4). Nevertheless, from the
results of FEM, it is obtained that the stress intensity factor KII is much smaller than the KI (KII <
0.1KI). Moreover, this effect, which was observed in all broken specimens by Scibetta[6], is
generally negligible (< 0.1mm). Therefore, the effect of the mode II is not taken into account in
evaluating the fatigue of the pre-cracking process.
due to shear
due to bending
Figure 4: Crack growth is deviated due to shear loading

Maximum stress intensity factor (contact of the crack lips is taken into account)
During the rotary bending fatigue process, the crack lips are in contact at the compressive side of
the bar (Figure 5). The contact increases the rigidity of the structure and decreases the stress
intensity factor. The notch depth is thus an important parameter to consider as it changes the contact
area, the position of the neutral axis and the moment of inertia.
Analytical solution
A model proposed by Sawaki[5] based on the beam theory, is described below.

The position of the neutral axis is found by solving the equation:
307

1 n c
y 1 (n c + y ) dy =
2
where,
bc =
bc + n c
y b c2 (n c y ) dy
2
(8)
b
n
and n c =
c
c
The modified moment of inertia is:
I m = 4c
1 n c
y 1 (n c + y ) dy
2
(9)
The modified maximal stress for the bar with this moment of inertia is:
m =
b+n
M
Im
(10)
Figure 5: New location of the neutral axis

The dimensionless function S is defined as the ratio between the maximal stress with contact and
the maximal stress without contact:
S(b c ) =
m
=
b 3c (b c + n c )
1 n c
16
1 (n c + y ) dy
(11)
Scibetta[6] proposed an explicit form of S as follows:

b
for > 0.4 ,
c
2
3
b
b
b
b
S = 0.0184298 + 0.408542 + 1.05795 0.448753
c
c
c
c
(12)
Stress intensity factor of a bar under a bending moment (contact taken into account)
b
b b
K = S G b1
R
c R
(13)
308
Finite element solution
To imitate the contact between the crack lips, the .JER command in SAMCEF is used. This
command lets us generate the required data relating to the relative contact with or without friction
between two groups of nodes or simply between two nodes.
Data generated by the .JER command are nodes according to the following principle: starting
from each node of the first group of nodes, one looks for a node of the second group of nodes in a
preset direction. If this is found, it is assumed that a contact might be possible between both nodes.
Then, a condition of relative contact is generated (gap) for a dummy DOF (node and MPC) which
represents in fact the relative displacement of the second node with respect to the first node
according to said direction.
The meshing (about 4000 3D elements), and the results: displacement, stress tensors, contact are
described in Figure 6.
Figure 6:
From Figure 6, it is obvious that a compressive zone is induced due to the contact between two
crack lips. And the SAMCEF results denote that the maximum stress intensity factor in this case is
smaller than the one in the case without contact. These finite element results correspond to the
above prediction.
The dimensionless function is obtained approximately from finite element solution SAMCEF:
2
b
b
b
b
S = 0.0289 0.42354 + 1.06735 0.46775
c
c
c
c
(14)
309
1.00
S SCI
0.90
S FE
0.80
0.70
0.60
0.50
0.40
0.30
b/c
0.20
0.40
0.50
0.60
0.70
0.80
0.90
1.00
Figure 7: Comparison of the function S.

The difference between two functions S is negligible (< 1%). It is obtained from Fig. 7 that the S
function is always less or equal to 1, S is equal to 1 in the case of no existence of the contact
(b/c=1), or in other words the contact effects tend to decrease the stress intensity factor. These
corrections can become large if the ligament diameter is small compared to the notch diameter (b/c
< 0.8).
Variable amplitude fatigue

The process, which was performed in order to make the CRB specimens, is a variable amplitude
fatigue process. In this process, the displacement of load point is maintained in a constant value
while the load gradually decreases. It is the reason why the variable amplitude fatigue is brought up.
In view of the complexity of the mechanisms underlying variable amplitude fatigue, it is not
surprising that only semi-empirical approaches are capable of providing reasonable estimates of
service life in engineering components subjected to complex loading histories. One of them is the
characteristic approach.
The characteristic approach was first proposed by Paris (1960) for the random loading. The basic
hypothesis here is that, because of similarity, the random variation of the crack tip fields are
describable in terms of the root-mean-square value of the stress intensity factor range, Krms. The
variable amplitude crack growth rates are then given by the Paris-type relationship:
da
m
= C(K rms )
dN
(15)
where C and m are material constants,

and
n
K rms =
K
i =1
2
i
(16)
Ki is the stress intensity factor range in the ith cycle in a sequence consisting of n stress cycles.
Note that Krms=K for the constant amplitude fatigue. Therefore, the characteristic approach
310
suggests that the average crack growth rate in spectrum fatigue can be predicted from constant
amplitude fatigue data.
2.2 Details
2.2.1 Pre-processing
In order to evaluate the fracture toughness, Scibetta[6] used the rotating bending fatigue technique
for the pre-cracking the CRB specimens. This technique is based on the application of a bending
moment to a rotating notched round bar, generating an oscillating load leading to fatigue crack
initiation and stable crack propagation. The rotating bending fatigue technique has two options
(Figure 8):
9 The cantilever configuration

9 The four-point beam configuration
Figure 8: Cantilever beam four-point bending configurations

There are some optical techniques to monitor the crack growth for the CRB. These techniques
are required to stop the pre-cracking process at the right moment. The main techniques for crack
depth evaluation are:
The compliance method
The AC or DC potential drop method
The measurement of the resonant frequency
Among them, the compliance method is very popular as it is a simple method with standard
instrumentation.
Load drop during pre-cracking
The compliance is the ratio between the load point displacement (maintained constant values) and
the load. Applying this definition to the initial state (no crack growth) and the final state gives:
load drop = 1
Pfinal
C(c )
=1
Pinitial
C(b )
(17)
where Pinitial is the initial applied force, Pfinal the measured load after crack growth, C(c) and C(b)
are the compliances.
311
It is obtained from equation (17) that the load drops as the compliance increases. To quantify the
load drop, Scibetta[6] proposed an analytical expression for the compliance in the cantilever beam
configuration and it is verified by the finite element calculation. This expression is given as follows:
3.525R 2
1 2 2 b
1
+
+
24
Rl H
4L2
L3
R
with Catt is the system attachment compliance

b, R, l, L are the geometry properties of CRB (Figure 1)
E is the Young modulus, the Poisson factor
C(b ) = C att +
H ( ) =
0.046875
4L3
3ER 4
(18)
0.06303639 0.01006494 2 + 0.03450366 3 + 0.009301849 4
+ 0.004570464 5 + 0.002724667 6 + 0.0001693451 7 + 0.004956346 8
(19)
From equations (17) and (18), it can be seen that it is possible to get a CRB specimen with a
desired crack length if the load drop is controlled during the pre-cracking process, or in other words
the applied load onto the specimen is monitored.
Consequently, the procedure to pre-crack a notched bar is:
Position the specimen into the collet chuck. The specimen alignment should be better than
0.005mm in order to avoid eccentricity of the remaining ligament
Apply a given load through the ball bearing system
Monitor the load drop as a function of time. When a given load drop is achieved, the precracking machine is stopped.
The last phase is the most time one (from 30 minutes to 4 hours). To avoid the need to monitor
the load drop by the operator, a computer was equipped with a data acquisition card. The load drop
is recorded and when the desired load drop is reached the computer stops the pre-cracking machine.
2.2.2 Determination of the fatigue constants

The stress intensity factor range in the ith cycle can be written by using eq. (13) (contact between the
crack lips is taken into account):
b
b b
K i = K i ,max = S G i b1
R
c R
(20)
4M
b 3
b
b
S is obtained from eq. (14), and G from eq. (7)
c
R
And from eq. (16), we have:
with i =
K rms
b b
= S G
c R
i =1
2
i
b1
R
312

n
b b 4
= S G 3
c R b
i =1
2
i
b1
R
(21)
Therefore, the relation between C and m is built from eq. (15):

C=
1
N
aN
da
(K )
a0
(22)
rms
The exact solution of the above relation is impossible because m is a real. So a numerical
approximation to a definite integral using rectangles is computed. The height of each rectangle
(box) is determined by the value of the function at the midpoint of each interval.
For examples, to compute the fatigue constants C and m of the material 22NiMoCr32 based on
the experimental results of precracking CRB specimens.
Specimen
Id.
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
D
(mm)
10.019
10.017
10.026
20.004
20.013
19.987
20.005
b
(mm)
2.178
2.227
2.220
5.513
5.252
4.908
5.364
D
(mm)
7.0
7.0
7.0
14.0
14.0
14.0
14.0
N
(cycle)
79386
81464
75290
114263
123033
139144
104058
M
(kNmm)
11.387
11.220
11.279
65.678
64.512
66.780
69.237
Table 1: Test results on 22NiMoCr32 material
D=2R is specimen diameter, 2b ligament diameter, and d=2c notch diameter.

N is number of fatigue cycles, M bending moment.
From the data in table 1, an approximation of eq. (22) is computed and its graph is drawn as
follows with the aid of MAPLE 6.
C
X5C1
X5B2
X5C2
X5B3
X5C4
X5B4
X5B8
Figure 9: Relation between C and m with m = 1.55

313
C
X5B2
X5B3
X5C1
X5B4
X5C4
X5C2
X5B8
Figure 10: Relation between C and m with m = 1.52

With each (Ci,mi), the number of fatigue cycles is re-computed by using eq. (15):
aN
1
Ni =
Ci
da
(K )
a0
(23)
mi
rms
where Krms is calculated by equation (20).

Then, the above computed Ni is compared to the experiment N. The results are presented in table 2.
C 10-5
Specimen
Id.
Nexperiment
(cycle)
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
79386
81464
75290
114263
123033
139144
104058
79386
81464
75290
114263
123033
139144
104058
mm kN
cycle mm 2
mm
4.55
1.65
4.61
1.67
Npredicted Difference
(cycle)
%
80041
80348
79871
106935
123897
131599
105025
80021
80365
79878
107089
123760
131598
105033
0.83
-1.37
6.09
-6.41
0.70
-5.42
0.42
0.80
-1.35
6.09
-6.28
0.59
-5.42
0.43
rms
3.98
3.95
-5
C 10
Specimen
Id.
Nexperiment
(cycle)
X5C1
X5C2
X5C4
79386
81464
75290
mm kN
cycle mm 2
4.67
mm
m
1.69
(cycle)
%
80016
80397
79899
0.79
-1.31
6.12
rms
3.92
314
X5B2
X5B3
X5B4
X5B8
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
114263
123033
139144
104058
79386
81464
75290
114263
123033
139144
104058
79386
81464
75290
114263
123033
139144
104058
79386
81464
75290
114263
123033
139144
104058
79386
81464
75290
114263
123033
139144
104058
4.74
1.71
4.79
1.73
4.87
1.75
5.01
1.79
107262
123943
131620
105062
79857
80273
79765
107226
123885
131387
104886
79882
80335
79816
107438
124113
131458
104953
79757
80245
79717
107446
124106
131280
104821
79559
80118
79570
107530
124170
131008
104623
-6.13
0.74
-5.41
0.46
0.59
-1.46
5.94
6.16
0.69
-5.57
0.29
0.62
-1.38
6.01
-5.97
0.88
-5.52
0.36
0.47
-1.50
5.88
-5.97
0.87
-5.65
0.23
0.22
-1.65
5.68
-5.89
0.92
-5.84
0.04
3.91
3.88
3.88
3.87
-5
C 10
Specimen
Id.
Nexperiment
(cycle)
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
X5C1
X5C2
X5C4
X5B2
X5B3
X5B4
X5B8
79386
81464
75290
114263
123033
139144
104058
79386
81464
75290
114263
123033
139144
104058
mm kN
cycle mm 2
5.14
5.21
mm
m
1.83
1.85
(cycle)
%
79582
80214
79644
107909
124577
130599
104712
79538
80206
79625
108022
124092
130051
104682
0.25
-1.53
5.78
-5.56
1.25
-6.14
0.13
0.19
-1.54
5.76
-5.46
1.35
-6.54
0.10
rms
3.89
3.96
Table 2: Comparison results of the number of fatigue cycles.

315
With the ranges of C and m as follows:
mm kN
C = 4.9110 6%
cycle mm 2
-5
mm
m = 1.75 4.6%
The maximum difference in the number of fatigue cycles between experiment and prediction is
approximate to 6%.
The reasons why the ranges of C and m are large can be:
9 The sensitive of the eq. (15) (the Paris law) is very large.
9 The errors in the experimental process.
9 The number of specimens used in the above calculation is few.
In order to decrease the dimension of these ranges, more specimens are necessary to be
performed with high reliability. Nevertheless, the difference 6% in the number of fatigue cycles
between experiment and prediction is acceptable.
Conclusions
This study contributes a way to evaluate the fatigue characteristics based on the CRB specimens
whilst Scibetta[6] only used them as a means of fracture toughness. To perform the evaluation of
the fatigue characteristics, the data of the pre-cracking process are used with the application of the
Paris law in which the knowledge of the formula of the maximum stress intensity factor is
necessary; therefore, this formula has been established by the analytical method as well as the finite
element method.
The advantage of this way is to salvage the data of the pre-cracking the CRB instead of carrying
out other new experiments for the evaluation.
Nevertheless, the disadvantage here is the values of fatigue characteristics are given in an
interval instead of the definite ones. It can be explained that the high sensitive of the Paris law is
one of the main causes. Moreover, the small number of specimens also causes this disadvantage.
Acknowledgement
The authors are grateful to Dr. Marc Scibetta for the providing of the experimental data.
References
[1]
[2]
[3]
Barsoum, R.S., On the use of isoparametric finite elements in linear fracture mechanics, International Journal
for Numerical Methods in Engineering, Vol.10, pp. 25-37, 1976.
Benthem, J.P. & Koiter, W.T., Asymptotic approximations to crack problem, Methods of analysis and
solutions of crack problems (Edited by G.C. Sih), chapter 3, pp. 131-178, Noordhoft International Publishing,
Groningen, 1973.
Harris, D.O., Stress intensity factors for hollow circumferentially notched round bars, Journal Basic
Engineering, 89, pp. 49-54, 1967.
316

[4]
[5]
[6]
Nikishkov, G.P. & Atluri, S.N., An equivalent domain integral method for computing crack-tip integral
parameters in non-elastic, thermal mechanical fracture, Engineering Fracture Mechanics, Vol. 26, pp. 851867, 1987.
Sawaki Y., Aoyama N., Abe Y., Oshima T., & Kawasaki T., Determination of fatigue fracture toughness by
conventional rotary bending specimens, Joint Conference on Experimental Mechanic, pp. 555-560, OahuMaaui, Hawaii, May 22-23 1982.
Scibetta, M., Contribution to the evaluation of the circumferentially-cracked round bar for fracture toughness
determination of reactor pressure vessel steels, Doctoral thesis, University of Liege, 1999.
317
Limit Analysis of 2D Structures using Gliding Line Mechanism

Generated by Rigid Finite Elements
Phan-Hong Quang* and Nguyen-Dang Hung+
EMMC-Ho Chi Minh City University of Technology, Vietnam
+
*
Abstract
A computational approach to the plane strain problem of limit analysis is developed and tested.
By assumption of the existence of gliding lines and the finite element discretisation, the structures
are discretized into rigid perfect-plastic elements so that the connection allows the glide between
each other with a kinematical admissible rate of displacement. The efficiency and accuracy of the
method is demonstrated by solving some problems in plane strain. Results are compared with those
given by other methods.
Keywords: Limit analysis, plasticity, finite element, indentation, and extrusion.
Introduction
This paper presents a simple technique of generating a system of lines of discontinuity rate,
basing on cinematically approach of limit analysis. The structures are discretized into rigid perfectplastic elements so that the connection allows the glide between each other with a kinematical
admissible rate of displacement. The determination of the internal of energy dissipation of all the
gliding lines is obtained without difficulty. An upper bound solution is then obtained by virtual
power equation. The upper bound theorem of limit analysis under the rigid perfectly plastic
behavior is applied to obtain the limit load. In order to have an approximation of the exact solution,
we have to minimize the upper bound P+ depending on the geometric parameters defining the
gliding lines in the actual mesh of rigid finite elements. It appears in this approach that the
coordinates of the nodes are these geometrical parameters. A process of optimization is
implemented so that the form of the mesh leads to the good one very close to the limit mechanism.
Finally, numerical analysis for some sample practical problems is realized and these results are
compared with those given by other methods.
Energy dissipation on the line of discontinuity [1], [2]
Rigid perfectly plastic Lvy-von Mises material is concerned and the plastified region is
confined inside a certain glide line with width equal to d. The rigid finite elements generate a
cinematically admissible velocity field, that is the licit collapsed mechanism of limit analysis.
318

x
(1) (1)
(1) Vx ,Vy
Vx
dx
Vy
dy
(2) (2)
,Vy
(2) Vx
Vn
Vs
vx
v y
v s
vn
= vx( 2) vx(1)
= v (y2 ) v (y1)
= v x n y + v y n x
= v x n x + v y n y
vx, vy : velocity
components.
nx, ny : the direction
cosines of the unit
normal n of the
discontinuity surface.
Fig. 1: Line of discontinuity

If the von Mises yield condition is adopted, the power of dissipation for Von Mises material is
[1]:
D = k 2&ijp&ijp
For plane strain this is reduced to the form:
2
3
D = kv
2&x2 + 2& y2 2&x& y + & xy2
(1)
3
2
Now, if the strain rates are replaced by the displacement variations, we have:
D=kv 1 4vn2+vs2
(2)
d 3
As we suppose the rigid elements glide along their sides, no discontinuities of material are
allowed that means vn = 0.
Finally, we can obtain the following expressions for D
D=kv vs
(3)
d
Rigid finite element
In this work, triangular elements are used to construct a rigid finite element mesh. Their
connection allows the glide between each element so that a field of kinematical admissible rate of
displacements is generated. The displacement field and the strain field of rigid elements are simply
in the following:
u = ay + b
v = ax + c
x = u = 0
u
=0
y =
y
1 u v
xy = + = a + ( a ) = 0
2 y x
i
v
u
j
Fig. 2: Triangular element
Therefore, D = 0 everywhere but not on the slip lines. Hence, the triangular element constitutes a
rigid block.
We proceed now to determine the energy dissipation on the gliding lines between adjacent
elements. Considering two adjacent element i and j with pq is a common side (fig. 3)
319

q
v = ai y + bi
Element i i
v y = ai x + ci
(j)
i
x
V
n y
(i)
Vy
(j)
Vx
s
(i)
v xj = a j y + b j
Element j j
v y = a j x + c j
Vx
Fig. 3: Rigid elements and gliding line pq
By geometry, we have
vn = v x n x + v y n y
= (vxj vxi )nx + (v yj v iy )n y
[
= (a
= (a
] [
a )( yn xn ) + (b b )n + (c
a )s + (b b )n + (c c )n
= (a j y + b j ) (ai y + bi ) n x + (-a j x + c j ) (ai x + ci ) n y
Similarly,
j
j
ci )n y
vs = - (a j ai )( yn y + xnx ) (b j bi )n y + (c j ci )nx
Condition vn = 0 leads to two following consequences:

a j ai = 0.
v n = 0
(b j bi )n x + (c j ci )n y = 0.
(4)
(5)
(6)
From (5), (6) and noting that n2x + n2y = 1, we obtain:
v s =
(b
bi ) + (c j ci ) = const
2
Finally if d is the width of a plastified line, we deduce:
D=kv 1 vs =const
d
The variation of the total dissipation along all slip lines is obtained by the integration over the
plastified domain. This plastic domain with width d will reduce to zero to identify to a slip line. The
details of this deduction may be put down in the following [1]:
Wp =eDdnds=ekvvs dnds
d
= ek v v s ds
Wp =ekvvs Li
(7)
Li : the length of the interface

e : the thickness of the structure.
The computation of the energy dissipation provides the following results:
The energy dissipation does not depend on the rotation parameter a.
For a gliding line, the energy dissipation is constant.
Consider the second condition of (6) recalled here for commodity,
(b j bi )nx + (c j ci )n y = 0
where (bi, ci) and (bj, cj) constitute the coordinates of two points which define the vector Wij .
This condition means that vector pq must be parallel to the vector Wij and on the other hand we
have:
320
vs = Wij = (bi bj )2+(ci c j )2
1/ 2
(8)
This geometric situation is illustrated on fig. 4

q
(b ,jc )
bj
ij
(b ,ic )
bi
cj
s
ci
O(0,0)
Fig. 4: Gliding line and the two adjacent elements

It appears that Wij and n are two orthogonal vectors and this geometric relationships is very
important in the correct evaluation of the energy of dissipation.
LAPSS program
The implementation of LAPSS algorithm is based on the above mentioned analytical
development. This program is written in Maple and Matlab language.
In summary, LAPSS tries to solve two following problems:
1/ For a triangular-element mesh with determined coordinates of nodes, LAPSS finds out directly
a corresponding upper bound.
2/ For the same mesh, but coordinates of nodes are variables, LAPSS tries to find a better upper
bound by moving the coordinates in such a way that the total dissipation leads to a minimum.
The minimization procedure is based on some OPTIMIZATION functions of described in
Matlab.
Results obtained by LAPSS are shown in filename.m file. It displays the limit mechanism just
before and just after the attained limit state.
Examples
P
5.1 Example 1: Punch indentation in

plane strain
In fact, this punch indentation problem is well
known; the exact solution for the limit load is [4]
P/k = 2a(2 + 2)
2a
Fig. 5: Punch indentation problem
321
Case = /2
Results and comparison
Number Results of Exact
of elements
LAPSS solution
5
5.400
9
5.193
11
5.173
5.142
19
5.153
21
5.147
2a
Error
(%)
4.785
0.994
0.607
0.227
0.106
Table 1: Results for punch indentation problem, = /2

0.2
5,450
04
-0.2
-0.4
6
-0.6
-0.8
-2
5,400
7
-1.5
5
-1
-0.5
0.5
results of LAPSS
slip-line solution
P/k
5,350
5,300
5,250
0.2
04
-0.2
0.2
-0.4
5,200
12
-0.8
-2
-1.5
04
13
-0.6
11
10
-1
8
-0.5
23
5
7
-0.2
22
0.5
-0.4
21
20
-0.6
-0.8
-2
14
13
15
19 18
17 16
-1.5
-1
12 11 10 9 8
-0.5
0.5
5,150
5,100
4
10
11
12
13
14
15
16
17
18
19
20
21
22
nel (number of elements)
Fig. 6: Comparison of LAPSS results and slip-line

OPTIMAL GRID*** gamma = 90* Nel = 19* Nnode = 21* P/k = 5.1533
OPTIMAL GRID***gamma = 90* Nel = 9* Nnode = 11* P/k = 5.1949

0.2
0.2
04
04
-0.2
-0.2
-0.4
-0.4
8
-0.6
-0.8
-2
11
10
-1.5
-1
-0.5
0.5
-0.5
-1
10
-0.8
-2
18 17 16 15
-1.5
-1
04
20
18
Fig. 7a: = /2, nel = 9
0.5
0.5
13 12
-1.5
-2
10 9
15 14
17
-0.5
11 10
9
-0.5
16
-1
-1
12
19
21
-1.5
14
-0.5
5
-1.5
-2
-2
8
11
19
INITIAL GRID*** gamma = 90* Nel = 19* Nnode = 21* Po = 30.8504
INITIAL GRID***gamma = 90* Nel = 9* Nnode = 11* Po = 22.8628

04
13
20
2
5
-0.6
3
21
-1.5
-1
11
-0.5
0.5
Fig. 7b: = /2, nel = 19
322
Case = /4
Results and comparison
Number Results of Exact
of elements LAPSS solution
5
3.604
9
3.578
11
3.575
3.571
19
3.573
21
3.573
45o
Error
(%)
0.921
0.184
0.112
0.062
0.045
Table 2: Results for punch indentation problem, = /4

The tables 1 and 2 illustrate the results for respectively the two cases: = /2, = /4
It appears (Fig. 8) that the values of upper bound converge rapidly to the exact value, (the errors:
from 0.95% with nel = 5 to 0.04% with nel = 21). Output results corresponding to number of
elements nel = 9 or nel = 19 are shown in the figures 7a and 7b or 8a and 8b, respectively for two
geometry cases. It can be seen that optimized mesh is close to exact mechanism pointed out by the
analytical slip line theory.
P/k
0.2
0
3,610
3,605
3,600
3,595
3,590
3,585
3,580
3,575
3,570
3,565
-0.2
-0.4
6
-0.6
7
4
-0.8
-2
-1.5
-1
5
-0.5
1
0
0.5
results of LAPSS
slip-line solution
0.2
0
-0.2
-0.4
8
-0.6
11 10
4
-0.8
-2
-1.5
-1
9
-0.5
0.2
6 5
0.5
1
-0.2
-0.4
-0.6
4 23
-0.8
-2
-1.5
22
14
15 1312
1110
1716
9 87
21201918
-1
-0.5
0.5
1
1
10 11 12 13 14 15 16 17 18 19 20 21 22
nel (number of elements)
Fig. 8: Comparison of LAPSS results and slip-line

OPTIMAL GRID***gamma = 45* Nel = 5* Nnode = 7* P/k = 3.6042
OPTIMAL GRID***gamma = 45 Nel = 19* Nnode = 21* P/k = 3.5732

0.2
0.2
0
-0.2
-0.2
-0.4
-0.4
13
14 12
1110
1615
9 87
191817
-0.6
7
-0.8
-2
-1.5
-1
-0.6
-0.5
0.5
-0.8
-2
INITIAL GRID***gamma = 45* Nel = 5* Nnode = 7* Po/k = 6.9439

0
21 20
-1.5
-1
-0.5
0.5
INITIAL GRID***gamma = 45 Nel = 19* Nnode = 21* Po = 23.3716
-0.5
4
-1
-0.5
1
7
19
4 21 20
-1.5
18
16
-1
13 12
17
-1.5
-1
-0.5
Fig. 9a: = /4, nel = 5
0.5
-1.5
-2
-1.5
-1
1
6 5
7
11
-2
-2
10 9
15 14
-0.5
0.5
Fig. 9b: = /4, nel = 19
323
5.2
Example 2: Extrusion through square dies

D IE
B IL L E T
RAM
2h
2H
2H: initial thickness

2h: final thickness
h
: fractional reduction
r = 1
H
bc
e=
: eccentricity of the die,
b+c
e (0,1)
Fig. 10: Extrusion problem

The second example is an extrusion problem illustrated in fig. 10 [3].
This problem has many separate cases and some of them will be examined separatly. For each
case, we will compare the solution computed presently by limit analysis (LAPSS program) to that of
the slip-line method. Readers can find the detailed solutions of the slip-line method in [3]. Besides,
the empirical solution will be also included for comparison.
5.2.1 Case of symmetrical extrusion
Because of symmetry (e = 0), only a half of container is considered as shown in fig. 11

DIE
RAM
h
BILLET
Fig. 11: Symmetrical extrusion
h
0.5 Small reduction
H
h
r = 1 > 0.5 Large reduction
H
r = 1
Case = 900
a). Frictionless extrusion

The results are plotted in fig. 12 for comparison with the slip-line solution.
12.0
results of LAPSS
10.0
slip-line solution
P/k
8.0
6.0
4.0
2.0
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
r (Fractional reduction)
Fig. 12: Comparison of LAPSS results and slip-line solution

The fig. 12 represents the values of the limit load for different values of r in the range 0 < r <1.
This figure permits us to see the effect of the parameter r on the magnitude of the upper bound.
The optimized mesh of the failure mechanism is very close to exact one with an initial arbitrarily
mesh.
324
OPTIMAL GRID***r = 0.428* Nel = 8* Nnode = 10* P/k = 2.4394
0.5
0.5
1
10
12
10
11
-0.5
9
8
-0.5
7
6
-1
-0.1
0.1
0.2
0.3
0.4
0.5
r = 0.428
-1
-0.1
0.6
INITIAL GRID***r = 0.428* Nel = 8* Nnode = 10* Po = 11.3915
0.1
0.2
0.3
0.4
0.5
0.6

01
01
12
-0.2
-0.2 11
10
-0.4 2
-0.6
-0.8
-1
0
0.1
0.2
0.3
9
8
-0.6
7
10
-0.4 2
9
8
0.4
0.5
0.6
-0.8
-1 r =0.4283
0
0.1
0.7
5 6
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Fig. 13b: = 90 , r = 0.428, grid 2,

nel = 10
Fig. 13a: = 90 , r = 0.428, grid 1,

nel = 8
b. Perfectly rough container wall
P/k
The results of this case are plotted in fig. 14 for more clear comparison
9.0
8.0
7.0
6.0
5.0
4.0
3.0
2.0
1.0
0.0
0.0
results of LAPSS
0.1
0.2
slip-line solution
0.3
0.4
0.5
emprical solution
0.6
0.7
0.8
0.9
1.0
r ( Fractional reduction)
Fig. 14: Comparison of LAPSS results, slip-line solution and empirical solution
0.5
0.5
10
12
1
11
2
-0.5
-0.5
10
9
8
7
6
5
-1
-0.1
r = 0.406
0
0.1
0.2
0.3
0.4
0.5
0.6
r = 0.406
0
0.1
0.2
0.3
0.4
0.5
01
01
12
-0.2 11
-0.2
10
-0.4 2
10
-0.4 2
9
8
-0.6
0.2
r =0.4063
-1
0.3
0.4
0.5
0.6
Fig. 15a: = 90 , r = 0.406, nel = 8

0
0.7
9
8
-0.6
-0.8
-0.8
0.6
-1 r =0.4063
0
0.1
-1
-0.1
0.1
5 6
4
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Fig. 15b: = 90 , r = 0.406, nel = 10

0
325
Case = 450
Frictionless extrusion

0
1
10
1
12
-0.2
11
8
-0.4 2
-0.6
10
-0.5
-0.8
4
r =3 0.4
-1
-0.4
-0.3
-0.2
-0.1
0.1
0.2

0
r =3 0.4
-1
-0.4
0.3
-0.3
-0.2
-0.1
0.1
0.2
0.3
1
12
-0.2
11
10
-0.4 2
-0.6
-0.8
-1
-0.4
r =0.4
-0.3
-0.2
-0.1
10
9
8
-0.5
6
5
0.1
9
8
0.2
0.3
-1
-0.4
0.4
Fig. 15a: = 90 , r = 0.4, nel = 8

0
r =0.4
-0.3
-0.2
-0.1
0.1
5 6
0.2
0.3
0.4
Fig. 15b: = 90 , r = 0.4, nel = 10

0
5.2.2 Case of unsymmetrical extrusion

Case = 900
a) Frictionless extrusion
4.0
3.5
3.0
P/k
2.5
2.0
1.5
results of LAPSS
slip-line solution
1.0
0.5
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
e (eccentricity of the die)

OPTIMAL GRID***r1 = 0.51r2 = 0.09* Nel = 4* Nnode = 7* P/k = 3.4637
1
0
0.5
-0.2
0
15
-0.4
14
13
-0.5
-0.6
11
-0.8
6
-1
-0.05
3
4
0
0.05
0.1
5
0.15
0.2
0.25
0.3
0.35
INITIAL GRID***r1 = 0.51r2 = 0.09* Nel = 4* Nnode = 7* Po = 5.5795
-1
-0.05
-0.4
0.05
5
-0.8
0.06
0.08
0.1
0.12
0.14
0.16
0.18
Fig. 17a: = 90 , r = 0.6, e = 0.7, nel = 4

0
0.2
14
0.25
0.3
0.35
13
10
11
9
8
-0.8
3
-1 4
0.04
0.15
12
2
-0.6
-0.6
0.02
0.1
-0.4
109
67 8
15
-0.2
3
-1 4
0
3
4

01
01
-0.2
12
7 6
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Fig. 17b: = 90 , r = 0.6, e = 0.7, nel = 12

0
326
b) Perfectly rough container wall

5.0
P/k
4.0
3.0
2.0
results of LAPSS
1.0
0.0
0.0
0.1
0.2
0.3
0.4
slip-line solution
0.5
0.6
0.7
0.8
0.9
e (Eccentricity of the die)

0.5
01
-0.4
11
10
-0.5 2
3
-1 4
0
0.05
0.1
7
0.15
0.25
0.3
0.35
12
5
-0.8
8
0.2
13
2
-0.6
9
5
01
15
-0.2
14
0.4
3
-1 4
0
6
0.05
7
0.1
11
10
9
0.15
0.2
0.25
01
10
15
-0.2
14
0.4
13
-0.4
-0.6
6
0.05
0.1
0.15
0.2
0.25
10
-0.8
0.3
0.35
Fig. 19a: = 90 , r = 0.6, e = 0.6, nel

=8
0
12
2
-0.6
-0.8
3
-1 4
0
0.35
01
11
-0.2
-0.4
0.3
3
-1 4
0
11
7 6
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Fig. 19b: = 90 , r = 0.6, e = 0.6, nel = 12

0
The difference figures are auto explained and illustrate the surprising fast convergence of the
procedure.
Through examples 1 and 2, we can see that the advantage of LAPSS (limit analysis) is that it can be
easily applied for all problems with wide range of the fraction reduction and the eccentricity, while
it is not easy to do that by the slip-line method because it is very complex to establish a slip-line
field for each value of the fraction reduction or the eccentricity.
Conclusions
Through the upper bound theorem with the assumption of the existence of the gliding lines and
the finite element discretization, this work proposes an automatic procedure to generate the
collapsed mechanism using slip line assumption. The method determines also directly the plastic
collapse loads for plane strain structures. In this short paper, we are restricted to present some
numerical tests of some practical problems where analytic results are well known. These tests have
clearly demonstrated the efficiency of the method. We may thus expect that this method will be
particularly useful in computation of more complex situation, since an automatically meshing will
be involve in our method. We hope in the near future presenting an extension of this method to the
cases of plane stress structures or three-dimensional structures.
327
References
[1]
[2]
[3]
[4]
[5]
[6]
Nguyen-Dang Hung, Cours calcul avanc en mcanique des solides et des structures, University of Lige,
1999.
W. F. Chen and D.J. Han, Plasticity for structural engineers, Springer- Verlag New York Inc, 1988.
J. Chakrabarty, Theory of plasticity, McGraw-Hill, Inc, 1988.
L. M. Kachanov, Foundations of the theory of plasticity, North-Holland Publishing Company - Amsterdam London (1971).
Phan-Hong Quang, Limit analysis of plane strain structure by assumption of the existence of the gliding lines
and the finite element discretization, Graduation thesis, EMMC (European Master of Mechanism of
Construction), 2001.
Philip E. Gill Walter Murray and Margaret H. Wright, Practical optimization, Academic Press INC. (London)
LMT, 1981.
328
An Application of Homogenizations Method

for the Behaviour Elastic of Polymer Concrete
Nguyen-Huy Gia, E. Ghorbel and S. Ortola
L2MGC Laboratoire de Mcanique et Matriaux du Gnie Civil
Universit de Cergy-Pontoise, FRANCE
Abstract
Polymer concrete consists of a polymer resin with an aggregate blend mixed. The addition of glassfiber more than doubles the flexure strength. We propose in this paper to use the homogenizations
method to describe the behaviour of such materials when subjected to flexion tests. The silica sand
is modelling by sphere whereas the glass fibers are considered as cylindrical particles. The
homogenization is performed into two steps. At first self-consistent method is used to the
computation of the effective properties of the composite composed with the polymer and the sand.
The obtained equivalent and homogenous material is considered as an effective matrix at which
short fibers are added. Then, the elastic behaviour of the reinforced polymer concrete is obtained
using Mori Tanaka method. Calculations with Castems software are done using the elastic
properties obtained throughout the homogenizations method. Three point flexion tests are
simulated in elasticity regime. Good correlation between experiments and simulated results is
obtained when we compare to the force/deflection curves.
Keywords: Composites materials; Homogenization; Effective properties; Self-consistent model;
Mori-Tanaka approximation.
Introduction
This paper is concerned with the determination of behaviours law of composites with elastic
phases. From a general point of view, a heterogeneous material is at least composed with two
different materials: a matrix and reinforcement. The added reinforcement makes it possible to
obtain a better resistance to the requests, an improvement of ductility, the attenuation of the heating
effects, etc
To determine the equivalent behaviour of such materials, the homogenization methods used in this
study are the self-consistent scheme Aboudi [1] and the Mori-Tanaka theory Mori et al. [10], which
are presented in the first section. Second we introduce the influence of the orientations of fibers in
the matrix on the equivalent behaviour of the heterogeneous material. Its numerical simulations are
presented in last section, in which we also investigate two types of applications of composite
materials: the reinforcement is harder than the matrix (Behaviour of Polymer Concrete) and the
reinforcement is weaker than the matrix (Behaviour of Lightweight Aggregate Concrete).
Method of homogenization
We consider, in this section, a composite consisting of a matrix phase denoted by the index 0 and
an inclusion phase denoted by the index r. The inclusion phase consists of ellipsoidal particles that
are of the same shape but that can be of different sizes. These ellipsoidal inclusions can be aligned
or randomly oriented in the matrix phase.
329
The tensor C (tensor of order 4) of the effective modulus of the composite satisfies:
= CE
(1)
where and E are the macroscopic constraint and the macroscopic deformation of the
composite. The problem will be solved through the analysis of the constraints and the deformations
in a material known as of comparison, here the matrix:
= C0 E0
(2)
The imposed constraint produces in the matrix inside the composite an average deformation
0 and an average constraint 0 different from the macroscopic values:
0 = + ~
= C 0 ( E 0 + ~ )
(3)
where ~ and ~
are respectively the averages, in the matrix, of the strains and the stresses of
perturbation which are generated by the presence of many heterogeneities. The constraint in the r th
phase is written like:
r = + ~
+ rpt = C r ( E 0 + ~ + rpt ) = C 0 ( E 0 + ~ + rpt *r )
(4)
are respectively the strain and the stress of perturbation compared to the average
where ~ and ~
state in the matrix. By using the theory of the equivalent inclusion of Eshelby, we have:
rpt = S Esh *r
(5)
where *r is the deformation of transformation, S Esh is the tensor of Eshelby.

We have the following relations:
r =n
r =n
E = frr
= frr
r =0
(6)
r =0
with: f r is the volume fraction of r th phase. The strain and stress in the inclusions are related to
the macroscopic strain and stress of the composite by:
r = Ar
r = Br
(7)
where Ar and Br are respectively the tensor of strains localization and the tensor of stress
concentration. Using Equations (6) and (7) yields:
r =n
r =n
r =n
r =0
r =0
r =0
r =n
r =n
r =n
r =0
r =0
r =0
E = f r r = f r S r r = ( f r S r B r )
(8)
= f r r = f r C r r = ( f r C r Ar )E
(9)
with:
E = S*
= C* E
(10)
330
Using equations (8), (9) and (10), we obtain the stiffness tensor ( C * ) and the compliance tensor
( S * ) of the composite materials:
r =n
r =n
C * = f r C r Ar
S * = f r S r Br
r =0
(11)
r =0
or
r =n
r =n
C * = C 0 + f r ( C r C 0 ) Ar
S * = S 0 + f r ( S r S 0 )B r
r =1
(12)
r =1
Relations (12) show that the effective behavior of the composite is computed as soon as the
tensor of localization Ar or of concentration Br are known. This determination depends on the
hypothesis made on the VER. We present in the next sections the both approximations we have
choosen.
2.1 The self-consistent method

The self-consistent method allows estimating the tensors of localization of an inclusion supposed to
be plunged in an infinite matrix having the homogenized characteristics which one seeks.
Then, the tensor of strain localization Ar is:
1
ArSC = [ I + S EshC* ( Cr C* ] 1
(13)
and, from relation (12), the tensor of the effective modulus is written Bourgeois [4]:
r =n
C * = C 0 + f r ( C r C 0 )[ I + S Esh C * ( C r C * ] 1
1
(14)
r =1
This method will be used in the third part, to take into account, the sand of polymer concrete and
the aggregates of lightweight concretes.
2.2 The Mori-Tanakas method

The work of Mori et al. [10] originally concerned with calculating the average internal stress in a
matrix of a material containing inclusions with transformation strain.
By introducing (5) into (4), *r is expressed by:
*r = Qr ( E 0 + ~ )
(15)
with:
Qr = [( C 0 C r )S Esh C 0 ] 1 ( C r C 0 )
(16)
The equilibrium condition of constraint in the reinforcements and the matrix must satisfied:
331

r =n
r =n
r =1
r =1
= f 0 0 + f r r = f 0 C 0 ( E 0 + ~ ) + f r C 0 ( E 0 + ~ + rpt *r )
(17)
Equation (17) can simplify by using (2) and (5):

r =n
r =n
r =1
r =1
~ = f ( pt * ) = f ( S Esh I ) *
r
r r r r
(18)
where I represents the identity tensor of the fourth order.

Replacing *r by (15), we have:
r =n
r =n
r =1
r =1
~ = [ I +
f r ( S Esh I )Qr ] 1 [ f r ( S Esh I )Qr ]E 0
(19)
In addition, the deformation in the composite is written like the average of the deformations of the
matrix and the reinforcements:
r =n
r =n
r =1
r =1
E = f 0 ( E 0 + ~ ) + f r ( E 0 + ~ + rpt ) = E 0 + f r *r
(20)
By using equations (15) and (19), we obtain a relation between E and E 0 :

r =n
r =n
E = [ I + ( f r Q r )[ I + f r ( S Esh I )Q r ] 1 E 0
r =1
r =1
(21)
Finally, with the definitions (1) and (2), we have:

r =n
r =n
C * = C 0 [ I + ( f r Q r )[ I + f r ( S Esh I )Q r ] 1
r =1
r =1
(22)
2.3 Random orientation of fibers

The aim of this section is to introduce in this above model, the influence of several orientations of
the inclusions in the effective behavior of the composite. Also, we introduce the three Euler
orientation angles ( , , ) .
Let the coordinate axes (1,2,3) rotate into the axes (1,2,3) using the matrix a( , , ) ,
which can be presented in the form of a product of three successive rotations (fig. 1) : a rotation
matrix a( ,0,0 ) corresponding to the angle around the axis 3, a rotation matrix a( 0, ,0 )
associated to the angle around the axis 1, and a rotation matrix a( 0,0, ) corresponding to the
angle once again around the axis 3 see Odegard et al. [12]
332

3=3'
rotation around axis 3

2'
3=3'
2'''
2''
2''
2' 2
rotation around axis 3''
2'
3=3'
3''
rotation around axis 1'
1'
3''=3'''
1'=1''
1'=1''
1'''
Figure 1: Effective fiber orientation

cos sin 0
a( ,0,0 ) = sin cos 0
0
0
1
0
0
1
a( 0 , ,0 ) = 0 cos sin
0 sin cos
cos sin 0
a( 0 ,0 , ) = sin cos 0
0
0
1
Then the following matrix a( , , ) : (1, 2, 3) = a( , , ) (1, 2, 3) can be expressed as

a product of the three matrices corresponding to the Euler rotations with the angles , , :
a( , , ) = a( ,0,0 ) a( 0, ,0 ) a( 0,0, )
Using the following notations: c1 = cos , c 2 = cos , c 3 = cos , s1 = sin , s 2 = sin ,
s 3 = sin , a = sin 2 , b = sin 2 , c = sin 2 , d = cos 2 , e = cos 2 , f = cos 2 . The
transformation matrix from axes (1, 2, 3) to axes (1, 2, 3) can be written as:
c1 c 3 s1 c 2 s 3
a( , , ) = s1 c 3 + c1 c 2 s 3
s 2 s3
c1 s 3 s1 c 2 c 3
s1 s 3 + c1 c 2 c 3
s1 c 3
s1 s 2
c1 s 2
c 2
'
The components of the fourth-order elasticity tensor Cijkl
with respect to the rotation frame are
obtained by:
333

'
C ijkl
= a pi a qj a rk a sl C pqrs
(23)
with the relevant transformation for the stress and the strain tensors
'ij = a ik a jl kl
'ij = a ki alj kl
(24)
In the case of n fibers families of different orientations, the axes of the micro-scale do not any more
correspond to the macro-scale axes so that we introduce the matrix of passage Pr with dimension
(6x6). The Mori-Tanakas formula (22) becomes:
r =n
r =n
C = C 0 [ I + ( f r Pr Qr Prt )[ I + f r Pr ( S Esh I )Qr Prt ] 1

r =1
r =1
(25)
where the components of the transforming matrix Pr are given below.

1
1
2 2
2 2 2
2 2
2 2 2
c1 c3 + s1 c 2 s3 2 acc 2 s1 c3 + c1 c2 s3 + 2 acc 2
1
1
c12 s32 + c 22 c32 s12 + acc 2 s12 s32 + c12 c 22 c32 acc 2
2
2
s12 s 22
c12 s 22
Pr =
1
1
2
( as 2 s3 bc12 c3 )
( as 2 s3 + bs1 c3 )
2
2
1
1 ( as 2 c3 bs12 s3 )
( as 3 c3 + bc12 s3 )
2
2
1
1
2 2
2
2 2
2
afc
c
s
c
c
afc
[
(
)]
[
2
1
2
1
2 + c ( c1 c 2 s1 )]
2
2
s 22 s32
cs1 s 2 + bc1 s32
cc1 s1 bs1 s32
s12 c32
bc1c32 cs1 s 2
cc1 s 2 bs1c32
c22
1
bc 3
2
1
bs 3
2
1 2
cs 2
2
bc1
bs1
s1c2 s3 + ec1c3
c1c 2 s3 es1c3
ec1 s3 s1c2 c3
c1c2 c3 es1 s3
1
fs1 s 2 bcc1
2
1
fc1 s 2 bcs1
2
cdc 2 + a ( s32 c22 s32 )
2
as 2
1
ds 2 s3 + abc 3
ds 2 c3 + abs 3
1
c2 ( dc 32 + s12 s32 ) ac ( 2 + c 22 )
4
a (c32 c22 s32 ) + cdc 2
This equation (25) is identical to previous equation (22) when this matrix Pr is equal to the identity
matrix I, as to say, when the three angles satisfy = = =0.
Numerical results and discussions
We illustrate, in this section, the above presented homogenization methods with different types of
applications: the usual case of stiffening inclusions ( 3.3) and the case of easing inclusions ( 3.1).
The section ( 3.2) concerns the influence of the fibers orientations to the effective modulus of the
composites.
3.1 The reinforcement is weaker than the matrix (Behaviour of Lightweight

Aggregate Concrete)
An example of this particular type of composites characterized with a ratio E0 / Er upper than one,
can be illustrated with the lightweight aggregates concretes (LWAC) used for structural
applications.
In order to determine the influence of the aggregates content on the LWACs strength,
experimental study has been carried through four volumic fractions of aggregates: 0.125, 0.25, 0375
and 0.45 of the total volume of concrete see Beaucour et al. [2]. Compressive strength tests have
been realized on 16x32mm cylindrical specimens. One of the tested experimental series is here
used, for which the properties of the mortar matrix and the lightweight aggregates are respectively:
E 0 =23630 MPa, 0 =0.2, E 1 =5679 MPa and 1 =0.15.
Several homogenizations models have been here computed for this second application: the
classic models of Reuss and Voigt approximations which give the lower and upper bounds of the
equivalent behavior, the Hashin-Shtrikman-Walpole bounds, the diluted solution, the self334
consistents model and the Mori-Tanakas theory. All the corresponding numerical results have
been compared with the experimental results on figure 2.
All the experimental values of the equivalent Young modulus are distributed between the upper
bound of Voigt model and the Mori-Tanaka approximation, which merges, in case of two phases
only, with the lower bound of Hashin-Shtrikman-Walpole. In the case of small volume fraction (f2 <
0.2), theses both methods compute the same equivalent behavior as the diluted approximation the
self-consistent method. The numerical simulations show that, in case of weaker inclusions, the most
appropriate model to predict suitably the experimental results is the Mori-Tanakas theory.
YOUNG'S MODULUS
27
Voigt
HSW-max
25
Reuss
Self-Consistent
Mori-Tanaka
Dilute
HSW-min
Experimental
23
E*(GPa)
21
19
17
15
13
11
9
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
volume fraction of the reinforcement
Figure 2: Different homogenizations method for LWACs behaviour
3.2 The orientation of fibers in the matrix

Now we consider the various cases of rotation of the fibers in a matrix by applying the MoriTanakas method (31). This application concerns glass fibers (with the volumic fraction of 15%) in
an epoxy matrix, of whom elastic moduli are presented in table 1.
Table 1: Elastic moduli of the components
Young modulus (GPa)
Poissons ratio
Epoxy R=0.8
3.8
0.31
Glass fibers E
73
0.15
The corresponding elastic stiffness tensor of the epoxy matrix is:

C0 =
5.2672
2.3664
2.3664
0
0
0
2.3664
5.2672
2.3664
0
0
0
2.3664
2.3664
5.2672
0
0
0
0
0
0
1.4504
0
0
0
0
0
0
1.4504
0
0
0
0
0
0
1.4504
and the components of the elastic stiffness tensor of the glass fibers are:
C1 =
77.0807
13.6025
13.6025
0
0
0
13.6025
77.0807
13.6025
0
0
0
13.6025
13.6025
77.0807
0
0
0
0
0
0
31.7391
0
0
0
0
0
0
31.7391
0
0
0
0
0
0
31.7391
335
In the case = = = 0 , the matrix of passage Pr is the fourth order unit tensor and the elastic
stiffness tensor of the composite (31) is:
Ccomposite =
5.3704
2.4053
2.3856
0
0
0
2.4053
5.3704
2.3856
0
0
0
2.3856
2.3856
6.3040
0
0
0
0
0
0
1.4906
0
0
0
0
0
0
1.4906
0
0
0
0
0
0
1.4826
We will now compute the equivalent behavior with different combinations of the rotations of the
fibers. In the following, these computations reduce to the plane case = 0 (figure 3).
3
fibres
2
1
Figure 3: Fiber orientation in the plane case

First of all, with several orientations around one axis (axis 1) and then, with different rotations of
around the two axes (axis 1 & 3). Case of fibers orientations around axis 1 only: ( = 0, = / 2 ).
Then, the equivalent behaviour differs from the previous one:
Ccomposite =
5.3704
2.3856
2.4053
-0.0000
0
0
2.3856
6.3040
2.3856
0.0000
0
0
2.4053
2.3856
5.3704
-0.0000
0
0
-0.0000
0.0000
-0.0000
1.4906
0
0
0
0
0
0
1.4826
0.0000
0
0
0
0
0.0000
1.4906
Maximal difference between diagonals terms: 0.9336 and shearing terms: 0.008
Case of rotation around axis 1 and axis 3: ( 0, 0 ).

We consider two cases of combinations of the families of fibers presented in the following table 2
and table 3, which also precise the corresponding volume fractions of inclusions.
Table 2: Case of four families of oriented fibers:
= / 3
= /6
= / 3
0.0025
0.0030
= /6
0.0060
0.0035
The resulting equivalent behaviour has the following components:

Ccomposite =
5.3704
2.3947
2.3960
0.0023
-0.0000
0.0000
2.3947
5.6941
2.5634
-0.2587
0.0000
-0.0000
2.3960
2.5634
5.6319
0.1509
0.0000
-0.0000
0.0006
-0.0647
0.0377
1.6680
0.0000
-0.0000
0.0000
0.0000
0.0000
0.0000
1.4863
-0.0005
-0.0000
0.0000
0.0000
-0.0000
-0.0005
1.4869
336
Table 3: Case of nine families of oriented fibers:

= / 6
= /6
= /2
= / 6
0.0016
0.0015
0.0017
= /6
0.0015
0.0016
0.0017
= /2
0.0019
0.0018
0.0017
The computed effective behaviour C * becomes:

Ccomposite = Fiber orientation in the plane case
5.5061
2.4436
2.4494
-0.0000
0.0003
0.0013
2.4436
5.5144
2.4458
-0.0000
0.0009
0.0041
2.4494
2.4458
5.7087
-0.0000
0.0041
-0.0001
-0.0000
-0.0000
-0.0000
1.5439
0.0000
0.0005
0.0001
0.0002
0.0010
0.0000
1.5474
-0.0000
0.0003
0.0010
-0.0000
0.0005
-0.0000
1.5354
If we go on with more complicated combinations, we will find an isotropic or quasi-isotropic

behavior of this heterogeneous material. So we find that the random orientation of no-spherical
reinforcements (here, it is the cylindrical reinforcement) can give an isotropic composite material
(or almost isotropic). It depends on the number of oriented fibers families and on their
combinations.
3.3 The reinforcement is harder than the matrix (Behaviour of Polymer

Concrete)
This application concerns the three-point-flexure experiment on a polymer concrete beam published
by Griffiths et al. [8] (fig.4). In order to compute the equivalent behavior of this composite, we have
modeled the silica sand by spherical reinforcements and the glass fibers by cylindrical inclusions.
Figure 4: Specimens & configuration used for three-point-flexure tests

The equivalent behavior of the beam was calculated within two steps: first, the self-consistent
model takes into account the silica sand (spherical reinforcements, 0.1-1mm, 78%) in the polyester
resin (20.5% of matrix); and second, the glass fibers (cylindrical reinforcements, 1.5%) are at least
introduced with the Mori-Tanaka theory (fig.5).
fiber
sand C
1
C2
+
C0*
C0
step 1
C0*
step 2
C*
Figure 5: Two steps to calculate a polymer concrete

The elastic behaviors retained for the three phases of the composite beam are recapitulated in
table 4.
337
Table 4: Elastic moduli of the components

Young modulus (GPa)
4
73.08
73
Polyester resin
Silica sand
Glass fibers E
Poissons ratio
0.3
0.172
0.15
The properties used for the silica sand refer to values published in the handbook of Robert
S.Carmichael [14]. The two steps of homogenization allow to obtain the equivalent coefficients
(Young modulus and Poissons ratio) of the elastic behavior of the composite beam.
Following step 1, we have:
C0* =
48.9199
8.7818
8.7818
8.7818
48.9199
8.7818
8.7818
8.7818
48.9199
0
0
0
0
0
0
0
0
0
0
0
0
20.0691
0
0
0
0
0
0
20.0691
0
0
0
0
0
0
20.0691
Following step 2, we have:

C* =
49.2306
8.8433
8.8366
8.8433
49.2306
8.8366
8.8366
8.8366
49.3374
0
0
0
0
0
0
0
0
0
0
0
0
20.2050
0
0
0
0
0
0
20.2050
0
0
0
0
0
0
20.1936
These last values have been then used in the finite element code Castem to compute the three
point deflection test and the numerical results of this simulation have been compared with the
experimental results published.
2500
2000
Force (N)
CASTEM
1500
1000
500
TEST
0
0
0.5
1.5
2.5
Deflection (mm)
Figure 6: Curve of force/deflection obtained by the test of Griffiths-2000 and by the numerical
simulation with CASTEM
Figure 6 shows that, for the elastic behaviour of the composite beam, the obtained results are
very closed to the experimental values. To take into account the nonlinear response observed in
experiment results, damage and visco-plasticity of the behaviour of the matrix must be introduced
in the models.
Conclusion and future work
In this paper, the elastic behavior of composite used in civil engineering is investigated using
homogeneous method. The considered materials are composed with at least three phases: the matrix
338
(here polymeric material), the sand (modelized as spherical) and reinforcements (short fibers). In
the elastic behavior, the effect of fibers on the polymer concrete is not predominant. This has been
outlined by previous experimental results. However it will be excepted that they affect the nonlinear
behavior and the damage as well as fracture mechanisms.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
Aboudi, J., A Unified Micromechanical Approach, Mechanics of Composite Materials, Elsevier, 1991.
Beaucour, A.-L., Fiorio, B., Ortola, S., Prediction of the Elastic Modulus of Structural Lightweight
Aggregates Concrete, CANMET Roumanie, 2003.
Benveniste, Y., A new approach of the application of Mori-Tanakas theory in composite materials,
Mechanics of Materials No.6, pp.147-157, 1987.
Bourgeois, N., Caractrisation et modlisation micromcanique du comportement et de lendommagement
dun composite matrice mtallique : Al/SiCp, Thse de lEcole Centrale de Paris, juin, 1994.
Decolon, C., Structure Composites Calcul des plaques et des poutres multicouches, Hermes, 2000.
Fiorio, B., Beaucour, A.-L., Ortola, S., Optimisation of the Mechanical Behavior of Lightweight Aggregate
Concrete by the use of High Performances Cementitious Matrixes, Concrete Structures: the Challenge of
Creativity, Symposium - April 26-28 - Avignon, France, 2004.
Guang, G., Etude micromcanique du comportement de structures composites a renforts discontinus, Thse
de lEcole Nationale Suprieure dArts et Mtiers, fvrier, 1997.
Griffiths, R., Ball, A., An assessement of the properties and degradation behaviour of glass-fibre-reinforced
polyester polymer concrete, Composites Science and Technology, No. 60, pp. 2747-2753, 2000.
Le Pen, E., Approche micromcanique du comportement en fatigue dun matriau composite matrice
aluminium renforce par fibre dalumine, Thse de lEcole Nationale Suprieure dArts et Mtiers, juillet,
1999.
Mori, T., Tanaka, K., Average stress in matrix and average elastic energy of materials with misfitting
inclusions, Acta Metallurgica Vol.21, pp.571-574, 1973.
Mura, T., Micromechanics of Defects in Solids, Martinus Nijhoff, 1987.
Odegard, G.M., Gates, T.S., Wise, K.E., Park, C., Siochi, E.J., Constitutive modeling of nanotube-reinforced
polymer composites, Composites Science and Technology, No.63, pp.1671-1687, 2003.
Qui, Y.P., Weng, G.J., The influence of inclusion shape on the overall elastoplastic behavior of a two-phase
isotropic composite, Int. J. Solids Structures, No.27, pp.1537-1550, 1991.
Robert S.Carmichael, Handbook of Physical Properties of Rocks, Volume III, CRC Press, pp. 92-94, 1984.
Tandon, G.P., Weng G.J., Average stress in matrix and effective moduli of randomly oriented composites,
Composites Science and Technology, No.27, pp.111-132, 1986.
Weng, G.J., Some elastic properties of reinforced solids, with special reference to isotropic ones containing
spherical inclusions, Internat. J. Engng. Sci. No.22, pp.845, 1984.
339
Dual Limit Analysis of Plate Bending
Le-Van Canh*, and Nguyen-Xuan Hung+ and Nguyen-Dang Hung

*
Department of Engineering, College of Construction No2, Vietnam
+
Division of Computational Mechanics, Department of Mathematics and Informatics, University of
Natural Sciences-VNU-HCM, Vietnam
Abstract
This paper presents a procedure for computing the lower bound and the upper bound of limit load of
bending Krichhoff plate. The lower bound is realized by using a new equilibrium element. The
upper bound is determined by using the conforming HCT element. By applying the procedure of
Richardsons extrapolation for the load factors, we can obtain a reliable approximation of the exact
limit load factor for problems of which true solution is not available. Finally, numerical examples
show good agreement with the literature.
Keywords: plate bending, equilibrium element, conforming element, dual method, limit analysis.
Introduction
Limit analysis had been played a significant role in many practical engineering such as the design of
mechanical structures and the analysis of soil mechanics. Based on the rigid-perfectly elastoplastic
model of material, the theory of limit analysis allows evaluate the loading capacity or the maximum
load intensity that the structure is able to support under proportional loading. Beyond this limit, the
structure will collapse due to global plastic flow [3, 5].
General limit analysis may be realized by obtaining estimate lower bound or upper bound of the
exact solution. The upper bound approach uses a kinematically admissible velocity field. The lower
bound approach must satisfy a statically plastically admissible stress that means a suitable satisfying
equilibrium conditions and yield criterion [3, 7]. The optimization techniques then are used from a
form of programming problems. The main objective of this paper is to establish a dual limit analysis
for the resolution of plate bending problems based on equilibrium and displacement finite element
models. Then, the stage of optimization technique for determining limit factors is solved by using
one toolbox of Matlab program. By comparing limit load factor of both models, we estimate the
reliability of the computation and error of the approximate solution.
The outline of this paper goes as follows. In section 2, the main results of the dual limit analysis
are given. In section 3, a summary of plastic theory of plate bending is described under Krichhoffs
assumption, so on the yield condition and the dissipation function. Section 4 presents the
conforming HCT element and the modified equilibrium element derived from Morley element with
implementing a second degree field of moment. Applying this equilibrium element to limit analysis
of plate bending is developed here for the first time. In section 5, three numerical examples are
illustrated to show the efficiency and the convergence of the methods. Finally, section 6 indicates
some conclusions and perspectives.
340
Dual limit analysis
Consider a perfectly rigid-plastic body of the volume of R 3 with boundary = u g .

Generally, at a part of the boundary u the displacement boundary conditions are prescribed, and at
another part g the stress boundary conditions are assumed. The external load subjects to the
surface and the volume which are denoted respectively by g and f . Let u& be a plastic velocity
fields or flow field that belongs to a space Y of kinematically admissible velocity fields. Let be a
stress field belonging to an appropriate space of symmetric stress tensor X. If the bilinear form
a ( , u& ) and the linear form F (u& ) represent the well-known internal and external work rates,
respectively, then the equilibrium equation is described in the form of virtual work rate as follows
a( , u& ) = F (u& ) , u& Y and u& = 0 on u
(2.1)
where the internal work rate is given by

a ( , u& ) = T &d
(2.2)
and the power of the external forces is

F (u& ) = f .u&d + g .u&d
(2.3)
with & = (u& ) denotes the strain rate. In addition to the equilibrium equation the stress tensor must
satisfy the yield condition. This stress field belongs to a convex set, B, obtaining from the von
Mises field condition
B = { X | sij sij 2k 2 }
(2.4)
where sij = ij m ij denotes stress deviator tensor, k is a parameter that depends on material
properties.
Now, let us consider the following problem of limit analysis: find the collapse multiplier
exact such that
exact = sup{ | B : a( , u& ) = F (u& ), u& Y }

= sup inf a( , u& )
B F ( u& ) =1
(2.5)
(2.6)
Equation (2.5) is known as the static limit load theorem (lower bound).
Reversing the order of maximum and minimum in (2.5) and (2.6) leads to the dual problem: find
the collapse multiplier exact such that
exact = inf { + | u& Y : a( , u& ) = + F (u& ), B}

= inf W& I (u& )
F ( u& ) =1
(2.7)
(2.8)
where
341
W& I = sup a ( , u& )
(2.9)
is total energy dissipation rate combined with the flow field u& .
Equation (2.7) is known as the kinematic limit load theorem (upper bound)
The problems (2.5) and (2.7) are traditionally known as the static and kinematic theorems of
limit analysis, respectively. The limit load of both approaches converges to the exact solution.
Herein, there exists a saddle point ( * , u& * ) such that both the maximum of all lower bounds and
the minimum of all upper bounds + coincide and are equal to the exact value exact [3, 13].
3 Plate bending equations, plastic conditions and power of

dissipation
The main equations for the analysis of thin plates are now briefly described on the plate of uniform
thickness, t, area A , kinematical boundary w wn and static boundary M M n , where the
subscript n stands for outside normal.
Equilibrium: Collecting the bending moments in the vectors M T = [ M xx M yy M xy ] , the
equilibrium equations can be written as () T M + p = 0 where p is the transverse load, , defined

the usual differential operators such that ()T = [ 2 / x 2 2 / y 2 2 2 / xy ] .
Compatibility: Collecting the curvatures in the vectors, T = () T w , obtain from the
compatibility equations of the rotations vector and the curvatures.
Constitutive relations: Assuming a plane stress state, the constitutive relations are given by
M = H where H represents the stiffness.
Boundary conditions: One states the general boundary condition frequently encountered in
Kirchhoff plates as selected combinations of the conjugate quantities: w = w or K n = K n and wn = wn
or M n = M n where K n , M n are the Kirchhoff shear load and normal moment, respectively.
Yield conditions: The von Mises yield criterion in the space of moment components can write
as follows
M xx2 M xxM yy+ M yy2+3M xy2 M P2
(3.1)
where M P = P t 2 / 4 is the yield criterion of moment.

The total power of dissipation (the total energy dissipation rate):
2M P
W& I =
3
[& & ]
1/ 2
dA
(3.2)
where
2 w& 2 w& 2 w& 2 w& T 2 w& 2 w& 2 w& 2 w&
, & = 2
2
2
2
2
2
x y x xy
x y y xy
& =
(3.3)
The external power:

W& E = pw& dA
(3.4)
342
Finite element for plate bending
4.1 The conforming HCT element

To obtain a compatible triangular element, by using only a cubic expansion, the triangle is divided
into three sub-triangles (Figure 1a). (The point 0 is the centroid of the original triangle)
(c)
Figure 1: Triangular element and sub-triangle division. The point 0 is the centroid
of original triangle.
These include the transverse displacements of each corner, wi , the rotations of each corner,
x = w / x i , y = w / y i , i = 1,2,3
i
(4.1)
x i and y i , as well as the rotation at three mid-side nodes 4 , 5 and 6 about axes parallel to the
corresponding sides.
i = w / n |i = wn |i , i = 4,5,6
(4.2)
The displacement expansion w ( x , y) is now assumed over each sub-triangle

w ( k ) ( x, y ) = N ( k ) ( x, y )q ( k ) , k = 1,2,3
(4.3)
where q ( k ) is the vector comprising the 10 displacement components of sub-element k.

The set of 10 cubic Lagrangian interpolation polynomials for a triangle can be described in area
coordinates = (1 , 2 , 3 ) as follows [9]
w = P
where
(4.4)
P = 13 23 33 12 2 12 3 22 3 221 321 32 2 1 2 3
T = [1 2 3 4 5 6 7 8 9 10 ]
(4.5)
(4.6)
The displacements in sub-element k may be expressed as
q
w ( k ) ( ) = N ( k ) ( )q = N e( k ) N 0( k ) R , k = 1,2,3
q E
(4.7)
343
where N ( k ) ( ) has additional zero components corresponding to the nodal displacements of the
complete assembly, not associated with sub-triangle, k. The partitions N e( k ) and N 0( k ) represent the
interpolation functions for the external displacements q R , and internal nodal displacements q E ,
respectively.
From condition of continuity, wn , at nodes{7, 8, 9},see Figure 3.1c, we establish in matrix form
the compatibility requirements as follows
q E = Cq R
(4.8)
where C is the matrix of elimination determined from condition of continuity of wn at internal

nodes.
Substituting the equation (4.8) into equation (4.7), we obtain
w ( k ) = [ N e( k ) + N 0( k ) C ]q R , k = 1,2,3
(4.9)
Now, the power of dissipation for ith subelement according (3.4) is:
[&
2M P
W& I( i ) =
3
& (i ) ] dA (i )
1/ 2
(i )
(4.10)
(i = 1, 2, 3: according above section, the number of sub-element equal to 3)

The total power of dissipation through each element is calculated in the form as follows
2M P
W& Ie =
3
w [&
& (ji ) ]
ng
i =1 j =1
(i )
j
1/ 2
(4.11)
where
- ng is the total number of integration point in each element A(i)
- w j is the weighting factor of the Gauss point j of Gauss-Legendre integration technique
Thus, we have the power of dissipation of system by assembling over all elements of the structure:
2M P
W& I =
3
w [&
ne
& (ji ) ]
ng
e =1 i =1 j =1
(i )
j
1/ 2
(4.12)
where ne is the total number of elements in domain A

The same way, we also obtain the external power of system:
ne
ng
W& E = w j pw& (ji )
(4.13)
e =1 i =1 j =1
4.2 The equilibrium Morley element

Classically, the moment distribution is assumed that it is constant within an element. One writes the
moments in the form
(4.14)
M = N ( x, y )
where N = [ ij ], i, j = 1,2,3 is a constant matrix is a constant matrix, = [ 1
3 ]T are
unknowns.
344
Consider the triangular element as Figure 2, load corners are calculated by:
Z i = M nt (i+ ) M nt (i ) i = 1, 2, 3
(4.15)
where
- M nt = M xx CS + M xy (C 2 S 2 ) + M yy CS twisting moment.
- C and S are the direction cosines of the outward normal to the boundary.
Figure 2: The generalized loads

Therefore, three load corners are rewritten in the form:
Z 1 = (C 3 S 3 C1 S1 ) M xx + (C1 S1 C 3 S 3 ) M yy + (C12 S12 C 32 + S 32 ) M xy
Z 2 = (C1 S1 C 2 S 2 ) M xx + (C 2 S 2 C1 S1 ) M yy + (C 22 S 22 C12 + S12 ) M xy
(4.16)
Z 3 = (C 2 S 2 C 3 S 3 ) M xx + (C 3 S 3 C 2 S 2 ) M yy + (C 32 S 32 C 22 + S 22 ) M xy
Total normal moment on each side ij:
lij
M n ds = (C i2 M xx + S i2 M yy + 2C i S i M xy )l ij
(4.17)
where
- C i = ( y j y i ) / l ij , S i = ( xi x j ) / lij , ij =12, 23, 31 and lij is the length of side ij.
- the normal moment: M n = C 2 M xx + S 2 M yy + 2CSM xy
However, with this moment fields, it is not possible to obtain exact equilibrium when applying a
constant pressure. For this, a special mode has to be complemented. It is constructed as follows. Let
side 1-2 be the X axis and Y axis be perpendicular to it, passes through node 1 and orientate in such
a manner that Y3 is positive. Let c1(X,Y), c2(X,Y) and c3(X,Y) be the three area coordinates and
c i = 0 be the equation of the side which i opposite to node i.
It is a second degree field as follows:
M comp = T
(4.18)
where M comp is equilibrium with constant pressure p and amplitude refers to pressure p .
345
Figure 3: The between global system (Oxy) and local system (OXY) generalized loads
For this equilibrium, a special mode T of moment field is of the form
X X2
X (X3 X2)
1 X
1
Txx = 3 c1 + 3
c2 3
c3 + ( e ) ( X 2 X 22 c2 X 32 c3 )
3 Y3
Y3
Y3
X2
2A
1 Y
1
Tyy = 3 c3 + ( e ) (Y 2 Y32 c3 )
(4.19)
3 X2
2A
2X3 X2
1
1
Txy = 12 c1 + 12 c2
c3 + ( e ) ( XY X 3Y3c3 )
3
2X2
2A
The complementary field in (4.19) has to be added to the basic field (constant field) in order to
give the general field. The result of this new moment field has been developed in the nearest article
by J.F.Debongnie [4, 10].
Therefore, the new moment field can be expressed as follows
M xx 1
Txx

M = M yy = 2 + T yy
M xy 3
Txy
(4.20)
Let the three corner loads of triangular element be added by
Z1 = Z 2 = Z 3 =
P
1
=
3
3
(4.21)
and one implements load g e = where P = pA ( e ) , A(e) is the area of triangular element e.
Thus, these assembled loads may be written in the matrix form
346
Z1
Z 2 C3 S3 C1S1 C1S1 C3 S3 C12 S12 C32 + S32 1/ 3

2
2
2
2
Z 3 C1S1 C2 S2 C2 S2 C1S1 C2 S2 C1 + S1 1/ 3 1
C2 S2 C3 S3 C3 S3 C2 S2 C32 S32 C22 + S22 1/ 3

g M n ds
2
2
S12l12
2C1 S1 l12
0
g = 12
= C1 l12
3
e M ds
C22l23
S22l23
2C2 S2 l23
0
n

23

2
S32l31
2C3 S3 l31
0
C3 l31
M n ds
0
0
0
1
31

P
(4.22)
= C
According to formulation (4.19), special mode T depends on coordinate of a point in area

ci(X,Y). So that we can replace local criterion for average criterion (Professor Nguyen-Dang Hung
& Knig propose an approximate method [3]):
f ( ) =
1
f ( )dA
A A
(4.23)
We can rewrite the yield condition:

f = 12 + 22 1 2 + 3 32 + 2
+
1
A
( T
A
2
xx
dA + T yy2 dA + 3 Txy2 dA
A
1
1 (2Txx T yy )dA + 2 (2T yy Txx )dA
A
A A
1 2
TxxT yy dA 6 3 Txy dA M P2 0
A
A
A
The same reason, the local moment can be replaced by the average moment:
Txx dA
1
A
M = 2 + T yy dA
A A
3
T dA
xy
A
(4.24)
(4.25)
Numerical examples
5.1 Clamped square plate subjected to a uniform load

Consider a square plate, edge length L = 10 m, thickness t = 0.1 m, yield stress p = 250 MPa. It is
clamped on all edges with a uniform distributed load, with p/p0 = (The coefficient is called the
multiplication of load). For symmetric geometry and boundary condition, we only solve the
problem with model as a quarter (upper-right) of the plate, see Figure 4. Mesh of M x M elements
over one quarter were used, with M = 2, 3, 4, 5, 6, 8 and 12. Figure 5 illustrates the convergence of
limit load factor.
347
Comments
In equilibrium approach (Morley element), the lower bound of load factor converges from
below, which indicates the elements are more flexible the exact solution. When mesh is refined the
lower bound of load factor is convergence to the exact solution.
In displacement approach (HCT element), the upper bound of load factor converges from
above, which indicates the element are more stiff then the exact solution. When mesh is refined the
upper bound of load factor is convergence to the exact solution.
Figure 4: clamped plate model with coarse mesh

Table 1: The numerical solution of the lower bound and upper bound of load factor
Number
of refines
1
2
3
4
5
6
7
Element
number
8
18
32
50
72
128
288
Lower bound
solution
0.17857
0.20855
0.22670
0.23768
0.24454
0.25262
0.26031
(*) The analytic upper bound of [5]: + =
Upper bound
solution
0.36866
0.34781
0.33464
0.32561
0.31668
0.30710
0.29720
Reference
Upper bound
0.30000 (*)
48M P
L2
To estimate the exact limit load factor of the structure, a series of meshes uniformly from
the uniform initial one have been created. Both displacements elements and equilibrium elements
have been used. Note that, by analogy, the Fraeijs de Veubeke displacement plate bending elements
can be used in place of the real equilibrium elements [1]. By applying the procedure of
Richardsons extrapolation for the displacement model, the extrapolation of limit load factor is
equal to 0.27811. Similarly, for the equilibrium model, the extrapolation of limit load factor is equal
to 0.27239. The mean value of these two extrapolated limit load factor is 0.27525, so that we can
take this value as a good approximation of the exact limit load factor [12].
348
Figure 5: The convergence of load factor for clamped square plate
5.2 Clamped ellipse plate subjected to a uniform load

Consider an ellipse plate, see Figure 6, length of half of the real axis a = 6 m, length of the image
axis b = 5.4 m, thickness t = 0.1 m, yield stress p = 250 MPa. It is clamped around the boundary
and uniform load, with p/p0 = (The coefficient is called the multiplication of load). As problem
is symmetric, it is considered only quarter of plate which has the form as follow: points on
boundary deflection w = 0, mean rotation = 0. Elements has its side on the axis the mean rotation
= 0.
Table 2: Numerical solutions of the lower bound and upper bound of load factor
Reference
Lower bound
Upper bound
Upper bound
solution
solution
0.16399
0.29420
0.19018
0.29024
0.20509
0.28467
0.21403
0.27907
0.2328 (*)
0.22014
0.27458
0.22707
0.26340
0.23034
0.25760
2
2
6M P (a + b ) + 6 M P (a 2 + b 2 )
=
(*) The analytic upper bound of [5]: + =
a 2b2
a 2b2
Number
of refines
1
2
3
4
5
6
7
Element
number
8
18
32
50
72
128
288
Figure 6: clamped ellipse plate model

349
Figure 7: An illustration of refinement meshes

Figure 8 illustrates the convergence of the lower and upper bounds and the average values. The
mean value of upper bound 0.23280 and lower bound 0.23034 is equal to 0.23157, so that we can
take this value as a good approximation of the exact limit load factor.
Figure 8: The convergence of load factor for clamped ellipse plate
5.3 L-shaped plate subjected to a uniform load

Consider a L-shaped plate with a uniform pressure and clamped on a part of its boundary, cf. Figure
9. Data of problem is the same of the first problem. The meshes will generally be composed of 3node or 6-node triangles with two different levels of refinement. Figure 10 shows that a uniform
mesh is generated.
350
Figure 9: L-shaped plate subjected to a uniform pressure
Figure 10: An illustration of refinements using uniform mesh

Table 3: Numerical solutions of the lower bound and upper bound of load factor
Number of
refines
1
2
3
4
5
6
Element
number
16
36
64
100
144
256
Lower bound
solution
0.07513
0.08317
0.08604
0.08730
0.08827
0.08975
Upper bound
solution
0.16056
0.14454
0.13724
0.13335
0.13218
0.13152
The convergence of the lower and upper bounds are shown by Figure 11.
Comments
The convergence rate of lower bound and upper bound method in this case are slow. To
have the better results, the refined mesh nearing the angle may be used.
By applying the procedure of Richardsons extrapolation for the displacement model, the
extrapolation of limit load factor is equal to 0.13129. Similarly, for the equilibrium model, the
extrapolation of limit load factor is equal to 0.11463. The mean value of these two extrapolated
limit load factor is 0.12296.
351
Figure 11: The convergence of load factor for L- shaped plate
Conclusions
The report presents the dual limit analysis for bending plate and shows how to estimate the lower
bound and upper bound of load factor. Through applying the theory of lower bound and upper
bound, the discretisation in finite elements and their applications, we achieved the following results:
Using the HCT element and Morley element, we can determine directly limit load with good
convergence of bending plate problem. Although having a little difference from the theory of
moment distribution, the results in chapter five are reliable and admissible.
Result of solution depends on software (local solution because of choosing bad start point, error
software) and technical meshing. The average criterion replaces local criterion, so that the yield
condition not be close. Furthermore, replacing the local moment field by the average moment field
brings about the distribution of moment has a little difference from theoretical distribution.
By using the Richardsons extrapolation for the displacement elements and equilibrium
elements, we can propose the good approximation of the exact solution of the problems which do
not have analytical solutions.
Although the lower and upper bound numerical formulations always supply rigorous bounds on
the limit loads, the quality of these bounds, like any finite element method, depends on the quality
of the finite element meshes. A discrete problem by finite element method can lead to weak bounds
and significant uncertainty in the exact collapse load. The above results only derive the evolution of
the computed limit load factor of the plate using a uniform refinement of the mesh. This in turn
leads to a larger factor of safety in design and an increased cost of construction such as the singular
problems. For complex practical problems, a sound knowledge of plasticity theory is needed to
generate finite element meshes, which result in cost-effective and accurate solutions. We thus
should combine with adaptive meshing that these limit load factor estimates prove to be effective
parameters in mesh optimization for plate analysis. This suggests a clear need for research into
adaptive mesh refinement procedures for lower and upper bound limit analysis techniques [14]. The
result of this investigation will be shown in another paper.
352
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
B.Fraeijs de Veubeke B, G.Sander, An equilibrium model for plate bending, Int. J. Soilds Struct., 4, 447 468, 1968.
G.Sander. Applications de la methode des elements finis a la flexion des plaques, Coll. Pub. Fac. Sc. Appli.
Univ. of Lige, N015, 1969.
Nguyen-Dang Hung, Cours avanc de mecanique des solides et des structures, Univ. of Lige.
J.F. Debongnie, Applying pressures on plate equilibrium elements, Univ. of Lige, may 2004.
M.A. Save, C.E. Massonnet, Plastic analysis and design of plates, shells and disks, North Holland publishing
Company. Amsterdam.
Yan Ai Min, Contributions to the direct limit state analysis of plastified and cracked structures, These de
doctorat 1999.Univ. of Lige.
Vu-Duc Khoi, Dual limit and shakedown analysis of structure, These de doctorat 2002.Univ. of Lige.
Yan Ai Min Vu-Duc Khoi, Nguyen-Dang Hung, Kinematical Formulation on Limit and Shakedown. Analysis
of Structures, Publication Series of the John von Neumann Institute for Computing (NIC). NIC series. To
appear.
Nguyen-Xuan Hung, Ladevze-type compatibility error assessment for plate bending. Master thesis of EUEMMC, HoChiMinh City University of Technology, 12/2003.
Nguyen-Xuan Hung, J.F.Debongnie, The equilibrium element finite model and error estimation for plate
bending, International Congress Engineering Mechanics Today, HoChiMinh City, August, 2004.
Le-Van Canh, Dual limit analysis of plate bending, Master thesis of EMMC, HoChiMinh City University of
Technology, November, 2004.
P.Beckers, H.G.Zhong, E.Maunder, Numerical comparison of several a posteriori error estimators for 2D
stress analysis, European Journal of Finite Elements, 1993.
E. Christiansen, On the collapse solution in limit analysis, Arch.Rational Mech. Anal.91, 119 -135, 1986.
H. Ciria, J.Peraire, Computation of upper and lower bounds in limit analysis using second-order cone
programming and mesh adaptivity, The 9th ASCE Specialty Conference on Probabilistic Mechanics and
Structural Reliability, 2004.
K. Krabbenhoft, L. Damkilde, Lower bound limit analysis of slabs with nonlinear yield criteria, Computers
& Structures, 80, 2043-2057, 2002.
353
Limit Analysis of Cracked Structures

by Primal-Dual Optimization
Vu-Duc Khoi
University of Kaiserslautern
P.O. Box 3049, D-67653 Kaiserslautern, Germany
Abstract
In this paper, the primal-dual theory of nonlinear optimization is applied to compute the collapse
load of cracked structures. With the use of displacement-based finite elements, the problem of limit
analysis is formulated by using both kinematic (velocity) and static (stress) variables. Numerical
examples show that the use of duality leads to rapid convergence and accurate results.
Keywords: limit analysis, duality, nonlinear optimization, crack.
Introduction
The theory and numerical methods of limit analysis have been the subject of study for decades.
Despite that fact that there exist analytical tools to deal with the problems of limit analysis, they are
limited to simple cases. Numerical methods, from very simple examples of two dimensions to very
complicated applications of three dimensions, have shown their great competence. Based on
mathematical programming and finite element technique, the direct method is generally more
effective than step-by-step method in estimating the limit load factor of structures. This method
appears to be more and more a powerful tool thanks to the rapid evolution of computer technology
in the past decades. Among pioneers in the developments of direct method are Biron and Hodge [2],
Hodge and Belytschko [9], Neal [16], Maier [13], Nguyen-Dang Hung et al. ([17],[18]), Casciaro
and Cascini [3], Morelle [15], Jospin [10], Yan [23]
Traditionally, the problem of limit analysis can be solved by either static or kinematic approach.
The static approach is based on the static theorem of limit analysis and uses stresses as main
variables. This approach leads to a maximization problem. The kinematic approach is based on the
kinematic theorem with displacement rates as main variables and leads to a minimization issue. The
use of static or kinematic variables separately often leads to convergence difficulties when largescale problems are considered. The convergence problem appears when von Mises yield condition
is used in the static approach because a large system of nonlinear inequalities needs to be solved. In
the kinematic approach, this problem may stem from the use of penalty methods to enforce the
incompressibility condition of plastic flows. This convergence problem can be overcome by
employing a primal-dual optimization method developed in the framework of the European project
LISA [19]. In this method, the advantages of kinematic and static approach are combined and their
disadvantages are reduced by the simultaneous use of velocities and stresses ([20], [21] and [22]).
In the next sections, a brief description of primal-dual method is presented for limit analysis and
numerical examples concerning cracked structures are introduced to demonstrate the effectiveness
of the method.
354
Primal-dual method in limit analysis
We consider a structure made of elastic perfectly plastic material and subjected to external loading
F, which consists of general body force f and surface traction t . We assume that all loads are
applied in a proportional way:
F = F0
(2.1)
where F0 = (f 0 , t 0 ) denotes the nominal load and is the load multiplier. The value l
corresponding to the plastic collapse state of the structure is called the safety factor of structure or
the limit load multiplier. We assume that the deformation and displacement are sufficiently small so
that changes in geometry can be neglected in the equilibrium equations and strain-displacement
relations are in linear form. According to the kinematic theorem of limit analysis, the upper bound
of the limit load multiplier can be formulated as:
+ = min V
(&ijp )dV
E
ij
&ijp dV
u& i = 0
s.t : p 1 u& i u& j
&ij = 2 x + x
i
j
(2.2)
on Vu
in V
where + denotes the upper bound of the limit load multiplier, V the volume of the structure under
consideration, Vu the boundary of V where fixations are applied, D p the plastic dissipation
function, &ijp and ijE the plastic deformation rate and the fictitious elastic stress, u& i and u& j
velocities along xi and x j . If kinematically admissible elements and von Mises yield criterion are
used, this upper bound can be discretized in a simple form as:
NG
+ = min 2k v e Ti e i + 2
i =1
e i B i q = 0 i = 1, NG
s.t : D v e i
= 0 i = 1, NG
NG
e Ti t i 1 = 0
i =1
(2.3)
where we define the normalized strain rate vector e i = wi D1 / 2 & ip , the normalized fictitious elastic
stress vector t = D 1 / 2 E and the normalized deformation matrix: B = w D1 / 2 B . In these
i
1 / 2 1
definitions D1 / 2 and D 1 / 2 are symmetric matrices such that: D 1 / 2 = D
and D = D1 / 2 D1 / 2 .
In a three-dimensional model D and D v have the size 66 such that:

1
D = Diag 1 1 1
2
1
2
1
2
(2.4)
Dvij = 1 i & j = 1,3, Dvij = 0 other i, j
355
& ip and iE denote the vector of deformation rate and vector of the fictitious elastic stress at Gauss
point i, q is the nodal displacement vector, Bi is the deformation matrix, NG denotes the total
number of Gauss points of the whole structure with integration weight wi at Gauss point i, is a
regularization parameter of small value and k v = 0 3 , where 0 is the yielding limit of
material.
On the other hand, based on the static theorem of limit analysis, the lower bound of the limit load
multiplier can be written as:
= max
j ij = 0
in V
s.t : n j ij = 0
on V
E
f ( ij + ij ) 0
(2.5)
where denotes the lower bound of the limit load multiplier , ij denotes the residual stress
field, V denotes the boundary of V where external tractions are applied, n j denotes the normal
vector of V and f denotes the yield function. If kinematically admissible elements are used, this
formulation can be discretized as [20]:
= max
i ,
NG T
B i i = 0
s.t : i =1
f (t + ) 0
i
i
(2.6)
where i is the normalized residual stress vector.

If the kinematic formulation (2.3) is used, we have a kinematic procedure and the obtained solution
will be an upper bound of the limit load multiplier under certain conditions. On the contrary if the
formulation (2.6) is used, we have a static procedure and the obtained solution will be an
approximation of the limit load multiplier without any bounding characteristic [20]. In both cases
experience shows that convergence difficulty often appears when a large number of elements are
required to discretize the structure leading to a large-scale problem. The optimization process in
those cases can be arbitrarily slow or breaks down very early and sometimes leads to under- or
overestimation of the structures loading capacity. A remedy for this situation is proposed in [20]
and [22], where a combined system of (2.3) and (2.6) is constructed such that the non-linearity of
the problem is reduced and the two solution + and can be obtained simultaneously while only
a small amount of additional computation effort is required. The combination of the two systems
(2.3) and (2.6) can be written as follows:
356
2k e e T e (t + + ) = 0
v i
i i
i
i
i
D v e i = 0
e i B i q = 0
NG T
B i i = 0
i =1
NG
e Ti t i 1 = 0
i =1
t i + i + i 2k v
(2.7)
In the above system von Mises yield criterion is used. The vectors i , i and the scalar actually
play the role of Lagrange multipliers when the dual problem of (2.3) is considered. It is easy to
prove that at the end of the optimization process, when all equality conditions in (2.7) are satisfied,
we have [22]: i = D v (t i + i ) / 3 . Therefore the last inequality of (2.7) will become exactly the
von Mises yielding condition for the normalized stress vector (t i + i + i ) .
Direct use of Newton method to solve the system (2.7) will lead to high computational cost.
Alternatively a primal-dual procedure can be employed [22], which solves at the same time both
problems (2.3) and (2.6). This primal-dual procedure focuses on kinematic solution and uses static
solution as feedback information for the kinematic one. While the kinematic variables are computed
directly to ensure that the corresponding objective function decreases after each iteration, the static
variables are computed indirectly through kinematic variables. Only yield condition is enforced
explicitly. By this approach, the compatibility and incompressibility of the kinematic field
corresponds also to the (quasi-) equilibrium of the static field. Although this process still solves the
static formulation (2.6) and gives precise solution at the end, it does not need special optimization
method to deal with the nonlinear system of inequalities as required in the purely static approach.
The simultaneous use of both stress and strain as variables in the optimization process reduces the
singularity of the Jacobian matrix, which is needed when Newton method is involved in solving
(2.7), and therefore increases the convergence rate, overcome early breakdown that kinematic or
static approach may encounter. Together with various numerical tests that were carried out in the
LISA project ([19]-[24]), the numerical applications presented in the following section will confirm
this property of the primal-dual method.
Numerical examples
3.1
Double-edge cracked plate subjected to tension
We study here the plastic collapse problem of a double-edge cracked plate under tension p, as
shown in Figure 1. Both plane stress and plane strain conditions are considered.
Plane stress condition
In plane stress condition, the solution is trivial if the Tresca yield criterion is used. The yield zone
has the form of a straight neck running across the minimum section 2c. Let 2b denote the wide of
357
the plate and a the length of the crack. If we define the limit load factor by the ratio of the limit
stress c in minimum section (2c) to the uniaxial yield stress 0 , then the exact solution can be
written as:
c
=1
0
(3.1)
2b
A
a
2c
Figure 1: Double-edge cracked plate: FEM model and mesh

Now let us consider the Von Mises yield criterion. According to Hill [8], the yield zone also takes
the form of a straight neck across the minimum section 2c. However, for deeply cracked plate we
need to replace 0 by the plane strain limit of material:
c
2
=
0
3
(3.2)
Unfortunately, the solution (3.2) is not valid for short cracked plates. Following Ewing and Spurr
[7], the limit load factor in this case can be estimated as:
= 1 + 1.08
a
b
a :
a
0.143
b
(3.3)
Numerical investigations carried out by Yan [23] showed that the solution (3.3) is overestimated.
According to Yan, the transferring value of a/b is about 0.261:
358
c
2
=
0
3
a
= c = 1 + 0.593
0
b
a :
a
> 0.261
b
(3.4)
a
a : 0.261
b
Plane strain condition
In plane strain condition, Ewing and Hill [5] showed that, for a cracked bar under tension, the limit
load factor is:

= 1 +
2
a :
b
= 1 + ln( + ) a :
2 2c
a
0.884
b
(3.5)
a
< 0.884
b
where = 1 if the Tresca yield criterion is used and = 2 / 3 in case of Von Mises.
1.16
1.14
Limit Load Factor
1.12
1.1
Numerical solution
Yan's solution
Hill's solution
1.08
1.06
1.04
1.02
1
0.1
0.2
0.3
0.4
0.5
a/b
0.6
0.7
0.8
0.9
Figure 2: Collapse limit load factor of double-edge cracked plate

(Plane stress state)
Also dealing with strain condition, Kumar et al. [11] obtained the following approximate solution
by using elastic-plastic finite element computations:
1
2
b
c
= (0.72 + 1.82)
(3.6)
In our numerical studies, the cracked plate is modeled by 824 plane 8-node quadrilateral elements
as shown in Figure1. Among these elements, there are 8 singular elements surrounding the crack tip
as described in the work of Yan and Nguyen [25]. Numerical results for different ratio a / b are
depicted in Figure 2 and Figure 3. A good agreement with reference solutions can be observed in
both plane stress and plane strain condition.
359

3.2
3
2.8
Limit Load Factor
2.6
Numerical solution
Ewing's solution
Kumar's solution
2.4
2.2
2
1.8
1.6
1.4
1.2
1
0.1
0.2
0.3
0.4
0.5
a/b
0.6
0.7
0.8
0.9
Figure 3: Collapse limit load factor of double-edge cracked plate

(Plane strain state)
2
Iteration
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
1.8
Limit Load Factor
1.6
1.4
1.2
Upper bound
Lower bound
1
0.8
0.6
0.4
8
Iteration
12
Upper bound
1.86281
1.70722
1.66896
1.65672
1.64834
1.64266
1.63699
1.63378
1.63201
1.63119
1.63094
1.63085
1.63082
1.63082
1.63082
1.63082
Lower bound
0.51687
0.88552
1.24499
1.46136
1.51038
1.56813
1.58206
1.58485
1.59396
1.59588
1.59816
1.60083
1.60108
1.60116
1.60116
1.60116
16
Figure 4: Limit analysis of double-edge cracked plate

The evolutions of upper bound and lower bound for a / b = 0.5 with plane strain condition (Figure
4) show that both upper and lower bounds converge quite well to analytic solution: the dual
algorithm leads to the upper bound + = 1.63082 and the lower bound = 1.60116 . Maximum
error of the present solutions is of 2 % compared with the corresponding analytical solution of
Ewing: E = 1.62289 (formula 3.5) and Kumars approximation: K = 1.63000 (formula 3.6).
3.2
Single-edge cracked plate subjected to tension
A single-edge cracked plate under tension is considered in Figure 5. The plate is subjected to
uniformly distributed load p at the two ends.
360
a
b
p
c
Figure 5: Single-edge cracked plate under tension

1
0.9
0.8
Numerical solution
Ewing's analytical solution
Ewing's experimental test
Limit Load Factor
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
a/b
0.6
0.7
0.8
0.9
Figure 6: Collapse limit load factor of single-edge cracked plate

(Plane stress state)
1.2
Numerical solution
Ewing's analytical solution: Upper bound
Ewing's analytical solution: Lower bound
Limit Load Factor
0.8
0.6
0.4
0.2
0.1
0.2
0.3
0.4
0.5
a/b
0.6
0.7
0.8
0.9
Figure 7: Collapse limit load factor of single-edge cracked plate

(Plane strain state)
361
Plane stress condition
In plane stress condition, an analytical solution was proposed by Ewing and Richards [6] by using a
slip-line method for deeply-cracked plate. The limit load factor defined by = p / 0 can be
computed as:
1 2
= (x +
) + (1 x) 2
2
1/ 2
(x
1
2
(3.7)
where x = a / b . If Trescas yield criterion is used then = 1 and the formula is valid for all values
of x. If von Mises criterion is used then = 2 / 3 in case of deeply-cracked plate. When
x 0.146 the limit load factor should be approximated as:
= 1 x x2
(3.8)
Plane strain condition
For a deeply-cracked bar in plane strain condition, the limit load factor can be computed by [6]:
= 1.702
{ (0.206 x)
+ 0.5876(1 x) 2 + (0.206 x)
x :
x > 0.545
(3.9)
In case of short cracks such that x < 0.545 , the limit load factor can be approximated by using the
two bounds given by Ewing (see Miller [14]):
(1 x 1.232 x 2 + x 3 )
(3.10)
[1 x 1.232 x 2 + x 3 + 22 x 3 (0.545 x) 2 ]
In our studies, the cracked plate is modeled by 824 plane 8-node quadrilateral elements as already
shown in Figure 1. Numerical results for different ratios a / b are depicted in Figure 6 and 7. Very
good agreements with reference solutions can be observed in both plane stress and plane strain
conditions. Note that in these figures, the numerical solutions are the average of the upper bound
and lower bound obtained by the dual algorithm.
0.6
Iteration
1
2
3
4
5
6
7
8
9
10
11
12
0.55
Upper bound
Lower bound
0.5
Limit Load Factor
0.45
0.4
0.35
0.3
0.25
0.2
Upper bound
0.60021
0.43976
0.40350
0.38574
0.37824
0.37484
0.37342
0.37303
0.37297
0.37296
0.37296
0.37296
Lower bound
0.13505
0.23306
0.28712
0.31438
0.33956
0.36045
0.36533
0.36810
0.36997
0.37002
0.37003
0.37003
0.15
0.1
0
6
8
Iteration
10
12
Figure 8: Limit analysis of double-edge cracked plate

362
The evolutions of upper bound and lower bound are plotted in Figure 8 for a / b = 0.5 with plane
strain condition. It is observed that the two bounds converge quite well to solution: the dual
algorithm leads to the upper bound + = 0.37296 and the lower bound = 0.37003 . Maximum
error of the present solutions is of 2 % compared with the corresponding analytical solutions of
Ewing: upper bound E+ = 0.37247 , lower bound E = 0.36604 .
3.3
Centre-cracked plate subjected to tension
We consider a centre-cracked plate under tension as shown in Figure 9. In plane stress condition,
the limit load factor = p / 0 can be calculated as (Miller [14]):
= 1
a
b
(3.11)
where a is the crack length, b is the plate width. The solution is valid for both Von Mises and
Tresca yield criteria.
In plane strain condition, if the Tresca yield criterion is used then the formula (3.11) is still valid.
On the other hand, if the Von Mises yield criterion is used, the limit load factor can be computed as:
2 a
1
3 b
(3.12)
c
p
a
c
Figure 9: Centre-cracked plate under tension

1.2
Numerical solution - Plane strain
Analytical solution - Plane strain
Numerical solution - Plane stress
Analytical solution - Plane stress
Limit Load Factor
0.8
0.6
0.4
0.2
0.1
0.2
0.3
0.4
0.5
a/b
0.6
0.7
0.8
0.9
Figure 10: Collapse limit load factor of centre-cracked plate

363
Numerical investigations are carried out with the FEM mesh described in Figure 1. Numerical
results are plotted in Figure 10. Maximum errors of the upper bound and lower bound are less than
3 % compared with analytical solutions (3.11) and (3.12).
The evolutions of upper bound and lower bound are plotted in Figure 11 for the case a / b = 0.5
with plane strain condition. The upper bound is + = 0.58639 and lower bound is = 0.57558 .
Note that analytical solution in this case gives = 0.57735 . Numerical error is thus less than 2%.
0.9
Iteration
1
2
3
4
5
6
7
8
9
10
11
12
13
0.8
Upper bound
Lower bound
Limit Load Factor
0.7
0.6
0.5
0.4
0.3
0.2
6
8
Iteration
10
Upper bound
0.88450
0.69686
0.62877
0.60192
0.58955
0.58944
0.58943
0.58884
0.58874
0.58813
0.58701
0.58639
0.58639
Lower bound
0.23940
0.37873
0.50266
0.55282
0.57211
0.57503
0.57551
0.57557
0.57558
0.57558
0.57558
0.57558
0.57558
12
Figure 11: Limit analysis of centre-cracked plate
3.4
Cylinder with circumferential crack subjected to axial tension
Consider a cylinder with a fully circumferential internal crack. The cylinder is subjected to an
uniform axial tension as shown in Figure 12. The problem was studied by Ainsworth 0, Kumar et
al.[11] and Yan [23].
p
A-A
Ri
Ro
A
t
Figure 12: Cracked cylinder under axial tension

364
By
= p / p net
defining
as
the
limit
load
factor
of
the
cracked
cylinder
where
p net = o R (Ri + a ) , Yan [23] proposed the following solution based on numerical tests:
2
2
o
1 + 0.76 t
=
1.19
a : a / t 0.25
(3.13)
a : a / t > 0.25
A
Ri
Figure 13: FEM mesh of cylinder with circumferential crack under tension
The limit load factor can also be estimated by a lower bound approximation as (Kumar et al.[11]):
= 2/ 3
(3.14)
1.25
1.2
Limit Load Factor
1.15
1.1
Numerical solution
Yan's solution
Kumar's solution
Ainsworth's solution
1.05
0.95
0.9
0.1
0.2
0.3
0.4
0.5
a/t
0.6
0.7
0.8
0.9
Figure 14: Collapse limit load factor of cracked cylinder under axial tension
For thin-walled cylinder of mean radius Rm =
1
Ro + Ri , Ainsworth et al. derived another lower
2
bound of the limit load (see Laham [12]):

1/ 2
a
3 a 2
FL = 2Rm (t a ) o
+ 1
a :
(
)
2
4
t
a
t
FL =
2
3
o [2Rm (t a )]
a :
t
1+ 3
(3.15)
t
1+ 3
365
where FL is the maximum axial force which we can apply on the cylinder. By using the
formulation (3.15), the limit load factor defined above can be written as:
1/ 2
a
3 a 2
+ 1
=
a :
2(t a ) 4 t a
a :
t
1+ 3
(3.16)
t
1+ 3
Numerical computations are carried out with the following parameters: Ro = 110 mm ,
Ri = 100 mm . The cylinder is modeled by 824 8-node axisymmetric elements (Figure 13) with 8
singular elements surrounding the crack tip as described in the work of Yan and Nguyen [25].
Numerical results for different crack ratios a / t are depicted in Figure 14. It is observed that for
deeply cracked cylinder the present numerical solution is close to the solution of Kumar or
Ainsworth, while for shallow crack the numerical result is roughly the approximation of Yan.
1.8
Iteration
1
2
3
4
5
6
7
8
9
10
11
12
Upper bound
Lower bound
1.6
Limit Load Factor
1.4
1.2
0.8
Upper bound
1.70116
1.43918
1.31013
1.22466
1.17279
1.15871
1.15547
1.15470
1.15427
1.15408
1.15408
1.15408
Lower bound
0.39189
0.66206
1.00633
1.08360
1.12829
1.13892
1.14669
1.14986
1.15022
1.15040
1.15043
1.15044
0.6
0.4
0
6
8
Iteration
10
12
Figure 15: Limit analysis of cracked cylinder under axial tension

The evolutions of upper bound and lower bound are plotted in Figure 15 for the case a / t = 0.5 . The
upper bound is + = 1.15408 and the lower bound is = 1.15044 . Numerical error is thus less
than 0.5% compared with analytical solutions of Kumar and Ainsworth = 1.15470 .
3.5
Cylinder with longitudinal crack subjected to internal pressure
In this example we study a long cylinder with longitudinal crack at internal wall (Figure 16.a). The
cylinder is subjected to internal pressure. The plane strain condition is studied here. Chell [4]
proposed the following analytical solution (see Miller [14]):
366
pl = o
Ri
ta
Ri + a
(3.17)
Ro
a)
b)
Figure 16: Cylinder with longitudinal crack under internal pressure

where t is the thickness of the cylinder, a is the crack length, = 1 if the Tresca yield criterion is
used, = 2 / 3 in case of Von Mises, pl is the internal pressure at plastic collapse.
Based on a solution of Hill for thick-walled uncracked cylinder:
R
p o = o ln o
Ri
(3.18)
Miller [14] suggested that this solution may be used to give an approximate solution for the cracked
cylinder by putting Ri Ri + a :
Ro
pl = o ln
Ri + a
(3.19)
By using formulas (3.17), (3.18) and (3.19) the limit load factor = pl / p o can be written as:
367

1.1
1
Numerical solution
Chell's solution
Miller's solution
Yan's solution
0.9
Limit Load Factor
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
a/t
0.6
0.7
0.8
0.9
Figure 17: Collapse limit load factor of cracked cylinder under internal pressure
Iter.
1
2
3
4
5
6
7
8
9
10
...
15
16
17
18
19
20
0.9
Upper bound
Lower bound
Limit Load Factor
0.8
0.7
0.6
0.5
0.4
0.3
8
12
Iteration
16
Upper bound
1.00756
0.78562
0.66022
0.60065
0.57294
0.56663
0.56461
0.56376
0.56339
0.56318
...
0.56294
0.56293
0.56293
0.56293
0.56293
0.56293
Lower bound
0.31978
0.44542
0.48800
0.52185
0.53530
0.54663
0.55090
0.55392
0.55829
0.55944
...
0.56221
0.56240
0.56266
0.56270
0.56271
0.56271
20
Figure 18: Limit analysis of cracked cylinder under internal pressure

- Chells solution:
R
ln o
Ri
ta
Ri + a
(3.20)
- Millers solution:
R
= ln o ln o
Ri + a
Ri
(3.21)
368
Based on numerical computations, another approximation was given by Yan [23]:
pl
a
a
= 1 0.7716 0.2267( ) 2
po
t
t
(3.22)
In our calculations, only one half of the cylinder is modeled due to its symmetry. FEM mesh
contains 1864 plane 8-node quadrilateral elements with 8 singular elements surrounding the crack
tip as described in the work of Yan and Nguyen [25]. Figure 16.b shows one part of the mesh near
the crack front. Numerical computations are carried out with the following parameters:
Ro = 60 mm , Ri = 50 mm . Numerical results for different crack ratios a / t are depicted in Figure
17. It is observed that the present numerical solution is higher than Millers solution and close to
Yans approximation.
The evolutions of upper bound and lower bound for a / t = 0.5 are plotted in Figure 18. The upper
bound obtained by dual algorithm is + = 0.56293 and the lower bound is = 0.56271 compared
with the solutions M = 0.47724 of Miller, C = 0.49862 of Chell and Y = 0.55753 of Yan.
Conclusions
Limit analysis provides a safety assessment and design of structures. It may be improved by using
the primal-dual method. By combining both static and kinematic approach, the primal-dual method
is able to approximate the upper bound of the loading capacity of cracked structures and at the same
time gives a quasi lower bound. Although this is not a true lower bound of the limit load multiplier,
it provides a good tool to determine the exactness of the obtained upper bound. Besides, the use of
static variables helps accelerate the convergence rate and therefore helps reduce the computing cost.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
Ainsworth A. The assessment of defects in structures of strain hardening material, Engineering Fracture
Mechanics, Vol. 19, No. 4, pp. 633-642, 1984
Biron A., Hodge P. G. Limit analysis of rotationally symmetric shells under central boss loadings by a numerical
method, Journal of Applied Mechanics, Volume 34, pp. 644-650, 1967
Casciaro R., Cascini L. A mixed formulation and mixed finite elements for limit analysis, Int. J. Num. Meth. in
Eng., Vol. 18 , pp. 211-243, 1982
Chell G. G. Elastic-plastic fracture mechanics in Developments in Fracture Mechanics-1 (edited by Chell G.
G.), Applied Science Publishers, London, pp. 67-105, 1979
Ewing D. J. F., Hill R. The plastic constraint of V-notched tension bars, J. Mech. Phys. Solids, Vol. 15, pp. 115124, 1967
Ewing D. J. F., Richards C. E. The yield-point loading of singly-notched pin-loaded tensile strips, J. Mech.
Phys. Solids, Vol 22, pp. 27-36, 1974
Ewing D. J. F., Spurr R. J. The yield-point loads of symmetrically-notched metal strips, J. Mech. Phys. Solids,
Vol. 22, pp. 37-46, 1974
Hill R. On discontinuous plastic states, with special reference to localized necking in thin sheets, J. Mech.
Phys. Solids, Vol. 1, pp. 19-30, 1952
Hodge P. G., Belytschko T. Numerical Methods for the Limit Analysis of Plates, Journal of Applied Mechanics,
Volume 35, pp. 796-801, 1968
Jospin R. J. Etats limites des tuyauteries par la mthode des lments finis et la programmation mathmatique,
Thse de doctorat, Universit de Lige, 1992
Kumar V., German M. D., Shih C. F., An engineering approach for elastic-plastic fracture analysis. NP-1931,
EPRI 1237-1, Electric Power Research Institute, USA, 1981
Laham S. Al Stress intensity factor and limit load handbook, EPD/GEN/REP/0316/98, Issue 1, Nuclear Electric
Ltd, 1998
369

[13] Maier G. A matrix structural theory of picewise-linear plasticity with interacting yield planes.,Meccanica 7, pp.
51-66, 1970
[14] Miller A. G. Review of limit loading of structures Containing Defects, Int. J. Pres. Ves. & Piping, 32, pp.197327, 1988
[15] Morelle P. Analyse duale de l'adaptation plastique des structures par la mthode des lments finis et la
programmation mathmatique, Thse de Doctorat, Universit de Lige, Belgique, 1989
[16] Neal, B. G., Limit load of a cantilever in plane stress. In: Engineering Plasticity (edited by Heyman J. and
Leckie F. A.), pp. 473-488, 1968
[17] Nguyen-Dang Hung Direct limit analysis via rigid-plastic element computer methods, Computer Methods in
Applied Mechanics and Engineering, Vol. 8, pp. 81-116, 1976
[18] Nguyen-Dang Hung, Trapletti M., Ransart D. Bornes quasi- infrieures et bornes suprieures de la pression
[19]
[20]
[21]
[22]
[23]
[24]
[25]
de ruine des coques de rvolution par la mthode des lments finis et par la programmation non-linaire, Int. J.
Nonlinear Mechanics, Vol.13, pp. 79-102, 1978
Staat M., Heitzer M. Numerical methods for limit and shakedown analysis, NIC Series Vol. 15, John von
Neumann Institute for Computing, Jlich, Germany, 2003
Vu-Duc Khoi Dual limit and shakedown analysis of structures, Doctoral Thesis. Collection des Publications,
Universit de Lige, Belgique, 2002
Vu-Duc Khoi, Yan Ai Min., Nguyen-Dang Hung A dual form for discretized kinematic formulation in
shakedown analysis, International Journal of Solids and Structures, 41, pp. 267-277, 2004
Vu-Duc Khoi., Yan Ai Min, Nguyen-Dang Hung A primal-dual algorithm for shakedown analysis of structures,
Computer Methods in Applied Mechanics and Engineering. Article in Press, available online at:
http://dx.doi.org/10.1016/j.cma.2004.03.011, 2004
Yan Ai Min Contributions to the direct limit state analysis of plastified and cracked structures, Doctoral thesis,
University of Liege, 1997
Yan Ai Min., Vu-Duc Khoi., Nguyen-Dang Hung Kinematical formulation of limit and shakedown analysis,
In: Staat M., Heitzer M. (Eds.) Numerical Methods for Limit and Shakedown Analysis. NIC Series Vol. 15, John
von Neumann Institute for Computing, Jlich, Germany, 2003
Yan Ai Min, Nguyen-Dang Hung Limit analysis of cracked structures by mathematical programming and finite
element technique, Computational Mechanics, Springer-Verlag, pp. 319-333, Vol 23, 1999
370
Initiation of Interface Crack at Free Edge

between Thin Films with Weak Stress Singularity
Abstract
In this study, delamination tests using a sandwich type specimen are conducted for eight
combinations of materials, thin films on silicon substrate, in order to develop a method for
quantitative evaluation and comparison of crack initiation strength from the free edge. Stress
distribution along interface at the crack initiation is analyzed by boundary element method. Since
the order of stress singularities, , in the materials are less than 0.07, the stress field near the
interface edge is almost constant in atomic level. Then, the critical strength for the interface
cracking is quantitatively represented by the concentrated stress near the edge, ya. Based on this
critical stress, the effects of the several factors such as existence of oxidized interlayer, species of
thin film materials and deposition process of thin film on the interface strength are examined as
well.
Keywords: thin film; crack initiation; interface edge; interface strength; stress singularity;
delamination; criterion
Introduction
In microelectronic devices, an interface edge, where a stress concentrates due to a deformation

mismatch, is a favorable initiation site of a crack. Since the micro-delamination brings about fatal
malfunctions of devices in service, the precise evaluation of the crack initiation criterion at the free
edge between thin films is a key task to improve the reliability of devices.
The interface toughness, which is the critical stress intensity (or energy release rate) for the crack
propagation, has been intensely investigated by many researchers [1-9]. Although it gives a
quantitative measure of interface strength, the crack initiation is more crucial in a small device [1013]. We proposed, then, a method for evaluating the resistance to crack initiation at a free edge
between thin films [14-17], and revealed that the crack initiation is characterized by the intensity of
the singular stress field. This is an analogous criterion to the fracture toughness which is the critical
strength of crack propagation in the ordinary fracture mechanics concept.
The singular stress field near an interface edge in a bi-material, ij, is expressed by the following
equation [18].
ij =
K ij
r
(1)
Here, r is the distance from the edge, Kij is the intensity of the singular stress field and is the order
of the singularity. The Kij has the dimension of MPam-, where depends on the elastic constants of
materials and the edge shape. Hence, quantitative comparison of critical Kij among different
371
materials is impossible even if they have the same edge shape; for example an orthogonal interface
edge (see Fig. 1). However, many of the pairs of metallic and/or ceramics films used in electronic
devices have no or low singularity with < 0.1; namely weak stress singularity. As shown in later
sections (in Figs. 7, 8 and table 4), the stress field near the interface edge is almost uniform in nanometer level. This may allow us the direct comparison of crack initiation strength on the basis of the
concentrated stress near the edge.
In this study, crack initiation tests are conducted for eight combinations of materials: thin films
formed on silicon substrates, in order to compare the strength among them. Then, the effects of the
several factors such as species of thin film materials, existence of oxidized interlayers and
deposition processes of thin films are also investigated.
Fig. 1 Orthogonal interface edge in bi-material
2.1 Materials
Materials tested are listed in Table 1. A silicon (Si) wafer with a (100) surface is employed as the
substrate of all materials. Adhesion strength is examined for several metal films; gold (Au),
aluminium (Al) and copper (Cu) films of 200 nm deposited on the substrates by the evaporation
method under the pressure of 1.9 x 10-3 Pa. These are denoted by e-Au/Si, e-Al/Si and e-Cu/Si,
respectively. The dependence of interface strength on the deposition method such as the evaporation
in vacuum and the sputtering in Argon gas is investigated by the specimens of a Cu film on the
substrate e-Cu/Si and s-Cu/Si. In some materials (e-Au/SiO2 and e-Al/SiO2), the surface of the Si
substrate is thermally oxidized to form the silicon dioxide (SiO2) film with the thickness of 100 nm
before the deposition of metal. An Au (or Al) film deposited on the substrate with natural SiO2 (eAu/Si and e-Al/Si) is tested for comparison in order to investigate the effect of oxidized interlayer.
Sputtered multi-layer films on the Si substrate s-Cu/Ta and s-Cu/SiO2 are also prepared. All
specimens, which consist of popular materials in microelectronic devices, possess clear interfaces
without a reacted interlayer between the thin films.
Table 1 Tested materials
Abbreviation
Tested material (thickness)
Fabrication method of film
e-Al/Si
Al (200 nm) / Si (350 m)
Vacuum evaporation (Al)
e-Al/SiO2
Al (200 nm) / SiO2 (100 nm) / Si (500 m )
Thermal oxidization (SiO2)

Vacuum evaporation (Al)
e-Au/Si
Au (200 nm) / Si (350 m)
Vacuum evaporation (Au)
e-Au/SiO2
Au (200 nm) / SiO2 (100 nm) / Si (500 m)
Thermal oxidization (SiO2)

Vacuum evaporation (Au)
372

e-Cu/Si
Cu (200 nm) / Si (550 m)
Vacuum evaporation (Cu)
s-Cu/Si
Cu (200 nm) / Si (550 m)
Sputtering (Cu)
s-Cu/Ta
Cu (200 nm) / Ta (50 nm) / Si (550 m)
Sputtering (Ta)
Sputtering (Cu)
s-Cu/SiO2
SiO2 (500 nm) / Cu (200 nm) / SiN (100 nm) /

Si (750 m)
Chemical vapor deposition

(SiN)
Sputtering (Cu)
Sputtering (SiO2)
2.2 Experimental method of delamination

The experimental setup is schematically illustrated in Fig. 2 [15, 16]. In the figure, the case of eAl/SiO2 is shown as an example. A test coupon is cut from the plate material, and a cantilever of
stainless steel is glued on it with standard epoxy. The size of specimens and cantilevers is listed in
Table 2. Load, P, is applied through the cantilever on the thin film, and delamination is expected to
be initiated at the left end of the interface. Since the cantilever as well as the substrate is relatively
stiff in comparison with the thin film, the sandwich structure restrains plastic deformation of the
thin film. The thin film outside the glue region is removed from the substrate before the test for
avoiding a film fracture. The validity and applicability of this method on the initiation of
delamination cracks were shown in the previous papers [15-17].
A micro-compression-testing-machine (Shimadzu, MCTE-500) is used to conduct the
delamination tests. The load, P, is applied at the end of the cantilever by an electro-magnetic
actuator under the constant loading rate of 0.02 N/s. The displacement at the loading point, uy, is
monitored continuously by a linear variable differential transformer during the tests. Two to seven
specimens for each material are prepared to check the repeatability. All the tests are conducted in
air at a room temperature of about 298 K.
Fig. 2 Schematic illustration of specimen and loading system; an

example for e-Al/SiO2 is shown
373
Table 2 Specimen size and critical load

Test No.
L1, mm L2, mm H1, mm H2, mm Critical load Pc, N
e-Al/Si-1
e-Al/Si-2
e-Al/Si-3
10.08
10.06
11.51
1.07
1.13
1.01
1.83
1.77
1.95
1.78
1.82
1.57
1.27
1.74
0.91
e-Al/Si-4
11.02
1.28
1.69
1.79
2.10
e-Al/Si-5
10.01
0.83
1.61
1.90
1.40
e-Al/Si-6
e-Al/Si-7
e-Al/SiO2-1
e-Al/SiO2-2
10.33
10.05
10.26
10.80
1.07
0.86
1.18
1.10
1.76
1.88
1.82
1.84
1.89
1.65
1.90
1.91
1.86
1.21
1.86
2.88
e-Al/SiO2-3
11.00
1.05
1.73
2.01
1.93
e-Al/SiO2-4
11.00
0.95
1.73
2.05
1.75
e-Al/SiO2-5
e-Au/Si-1
10.76
8.18
0.94
1.70
1.88
0.98
1.76
2.10
1.76
0.59
e-Au/Si-2
8.59
1.60
0.98
1.98
0.60
e-Au/Si-3
8.83
1.93
0.97
2.07
0.55
e-Au/Si-4
8.75
2.00
0.97
1.90
0.55
e-Au/Si-5
7.36
2.31
0.98
1.94
0.74
e-Au/SiO2-1
9.12
1.73
0.98
1.96
0.55
e-Au/SiO2-2
8.22
1.75
0.95
1.64
0.57
e-Au/SiO2-3
7.57
1.70
0.98
1.70
0.41
e-Au/SiO2-4
7.70
1.92
0.98
1.95
0.68
e-Au/SiO2-5
7.14
1.90
0.98
1.90
0.47
e-Cu/Si-1
10.00
2.13
1.80
1.95
0.76
e-Cu/Si-2
10.00
2.01
1.80
1.95
0.69
s-Cu/Si-1
10.00
1.54
1.80
1.96
1.92
s-Cu/Si-2
10.00
1.46
1.80
1.94
1.57
s-Cu/Ta-1
10.00
1.26
1.80
1.95
1.80
s-Cu/Ta-2
10.00
1.66
1.80
1.97
2.50
s-Cu/Ta-3
10.00
2.18
1.80
1.96
4.70
s-Cu/SiO2-1
s-Cu/SiO2-2
10.00
10.00
2.65
2.27
1.80
1.80
1.97
1.97
2.37
2.23
s-Cu/SiO2-3
10.00
2.76
1.80
1.95
2.79
Experimental results
Figure 3 shows typical relationships between the applied load, P, and the displacement at the
loading point, uy, in the eight materials. They are linear up to the point x in all cases. The
delamination crack is initiated at the interface edge at the x, and it propagates through the
interface just after the initiation without sub-critical crack growth. The critical loads at the
delamination, Pc, obtained by all the specimens are listed in Table 2. The fracture surfaces of both
the film side and the substrate side observed by optical microscopy are smooth and flat in all
specimens. The atom species on the fracture surfaces are examined by Auger electron spectroscopy
374
(AES). Figures 4 (a) and (b) show the spectra on the fracture surfaces of e-Au/SiO2 as an example.
On the film side, only Au peaks (69, 2024 and 2111 eV) are confirmed while Si and O peaks (92
and 1619 eV for Si and 510 eV for O) are recognized on the substrate side. This indicates perfect
interface cracking between the Au film and the SiO2 layer. The other AES analyses prove the
perfect interfacial cracking in all tests.
Fig. 3 Relationships between load and displacement
(a) Film side
(b) Substrate side

Fig. 4 Auger electron spectra on the fracture surface of e-Au/SiO2
375
Interface strength of delamination cracking
4.1 Stress analysis

The elastic stress distribution near the interface edge at the delamination load, Pc, is numerically
analyzed under the plane strain condition by boundary element method (BEM). In order to
reproduce precisely the stress field, the region near the interface edge where the stress concentrates
due to the deformation mismatch is carefully divided into fine elements. The element length at the
interface edge is set at 2 nm. Figure 5 illustrates the typical mesh-division, which is used for the
analysis of specimen e-Au/SiO2-1. The materials are assumed to be isotropic, and the elastic
constants used in the calculations are listed in Table 3.
Table 3 Elastic constants used in BEM analysis
Material
Si
SiO2
SiN
Ta
Au
Al
Cu
Epoxy
Stainless
steel
Youngs modulus
E, GPa
167
92
304
186
83
70
129
2.5
200
Poissons ratio
0.30
0.30
0.27
0.34
0.44
0.35
0.34
0.30
0.30
Figure 6 shows the distributions of a normal stress, y, and a shear stress, xy, along the interface
between Au and SiO2 layers in specimen e-Au/SiO2-5 at the delamination. The y highly
concentrates near the interface edge, and the intensity of xy is lower than that of y. This suggests
that the crack initiation is governed by the concentrated y; namely Mode I-dominated cracking.
Since y prevails over xy near the edge at the delamination in all tests, we discuss the critical
strength below on the basis of the normal stress. Figure 7 shows the magnified view (log-log plot)
of the stress distributions near the interface edge for the e-Au/SiO2 specimens. There is little
fluctuation on the stress near the interface edge among the specimens while they have large
difference in the region r > 500 nm. The coincident stress fields near the interface edge point out
that the concentrated stress governs the crack initiation. The line in the figure represents the
averaged-stress field near the interface edge at the crack initiation. The difference far from the edge
can be attributed to the difference of specimen size. The far field does not directly control the
initiation.
Similar tendency is observed for all of the materials. In the following section, hence, the
interface strength of each material for the crack initiation is compared on the basis of the averagedstress fields.
4.2 Stress singularity and concentrated stress criterion

Bogy [18] analyzed the stress field near the interface edge in dissimilar materials, and the order of
stress singularity in equation (1) is determined by a characteristic equation [19]. Bi-materials with
an orthogonal interface edge (see Fig. 1) are classified into three cases depending on [20]: bad
pair > 0, equal pair = 0 and good pair < 0. The stress singularity appears in the bad pair while
it does not in the good and the equal pairs. The magnitudes of in the present combinations of
materials are calculated using the characteristic equation and they are listed in Table 4. Here, we
defined the bi-materials with || < 0.001 as the equal pair although these are not exactly zero.
Materials, e-Al/Si, e-Al/SiO2, e-Au/Si, e-Cu/Si, s-Cu/Si and s-Cu/Ta are classified as the bad pair,
and the others are done as the equal pair. The stress singularity obtained by the numerical
calculation agrees very well with the analytical solution. The previous papers [15, 16] revealed that
376
the interface strength of crack initiation in the material of bad pair is prescribed by the critical stress
intensity, Kyc. The intensities are also listed in Table 4. Since Ky has the dimension of MPam-, the
strength, Kyc, is presented in different dimensions depending on the material combination. This
signifies that Kyc is not a good measure for the comparison of strength among the different
combinations of materials. Moreover, since the stress singularity does not exist for the material of
good or equal pairs, Kyc can not be an appropriate parameter to represent the interface strength of
crack initiation.
Fig. 5 Mesh-division for BEM analysis of specimen eA /SiO 1
Fig. 6 Stress distribution along the interface between Au and SiO2 at the delamination
Fig. 7 Magnified view of stress distributions along the interface between Au and SiO2 at
the delamination
377
On the other hand, deformation mismatches for the pairs of metallic and/or ceramics films
usually used in micro-devices are not so large in general, so that the magnitudes of in the bad
pairs are not great. As shown in Table 4, on the combinations used in this study are less than 0.07
(weak stress singularity). Figure 8 shows the critical stress field near the interface edge of e-Al/Si.
Although it has the singularity, the slope is very moderate, and the stress near the edge is almost
constant in the nano-meter (or atomic) scale. Then, the concentrated stress, yc, defined as the
following equation is evaluated.
yc
1
r1
r1
ydr
(2)
Here, r1 is the size of integration region. Figure 9 shows the relationship between yc and r1. yc are
almost flat in the region of r1 < 100 nm and Table 5 lists the values of yc at r1 = 1 nm and at r1 = 10
nm. The concentrated stress is insensitive to the region size. For example, even in e-Al/Si ( =
0.069) which has the strongest singularity in the tested materials, the difference between yc at r1 =
1 nm and yc at r1 = 10 nm is no more than 11%. Thus, yc is a general parameter which
characterizes the intensity of the stress field near the interface edge with weak or no stress
singularity. Since the crack initiation is governed by the stress field near the edge as shown in Fig.
7, we can use the concentrated stress instead of the stress singularity parameter Kij to quantitatively
evaluate the strength of the interface cracking. Since the dimension of yc is always MPa
independently of the magnitude of , the concentrated stress criterion can compare the crack
initiation strength among different combinations. The strength of materials can be summarized as
follows:
e-Al/Si > e-Al/SiO2 > s-Cu/Ta > s-Cu/Si > s-Cu/SiO2 > e-Au/Si > e-Cu/Si > e-Au/SiO2.
Of course, many of material combinations possess the strong stress singularity, > 0.1 (for
example, epoxy/Si, tungsten (W)/SiO2 and Au/silicone nitride (SiN)). These are out-of-scope in this
study, and the comparison method will be dissussed elsewhere.
Table 4 Orders of stress singularity and critical stress intensity factors
Material
Kyc, MPa m-
e-Al/Si (bad pair)
0.069
92.6
e-Al/SiO2 (bad pair)
0.011
138.0
e-Au/Si (bad pair)
0.067
8.3
e-Au/SiO2 (equal pair)
e-Cu/Si (bad pair)
0.009
15.5
s-Cu/Si (bad pair)
0.009
68.1
s-Cu/Ta (bad pair)
0.014
84.6
s-Cu/SiO2 (equal pair)
378
Fig. 8 Stress distribution near the interface edge of e-Al/Si at the delamination
Fig. 9 Relationships between concentrated stress and integration region at the delamination
4.3 Comparison among materials

4.3.1 Metal films on Si substrate
The adhesion of Al, Au and Cu films to Si substrate is quite different as shown in Table 5. The
interface strength of e-Al/Si is about ten times higher than those of e-Au/Si and e-Cu/Si. Though the
difference between e-Au/Si and e-Cu/Si is not so great in Fig. 9, yc in the Table 5 indicate that the
former (e-Au/Si) possesses 25 - 35% higher strength than the latter (e-Cu/Si).
Existence of the tantalum (Ta) layer increases the bonding strength of Cu film on a Si substrate.
The adhesion strength of s-Cu/Ta is 50% higher than that of s-Cu/Si. The sputtered silicon dioxide
film, which is used as insulator in electronic devices, on the Cu film has quite weaker interface than
the sputtered Cu film on the Si substrate.
4.3.2 Effect of SiO2 interlayer
Quantitative comparison based on Table 5 reveals that the metal films with the artificial SiO2 are
weaker than the ones without it. The interface strength of e-Al/SiO2 (or e-Au/SiO2) is about 20%
379
(30%) lower than that of e-Al/Si (e-Au/Si). Here, the effect of a natural SiO2 layer in e-Al/Si and eAu/Si on the stress distributions along the interface is examined by a preliminary BEM analysis
assuming that the thickness of SiO2 is about several nanometers. This guarantees that the effect is
negligible.
Table 5 Concentrated stresses
Material
yc, MPa (r1 = 1 nm)
yc, MPa (r1 = 10 nm)
e-Al/Si (bad pair)
240.0
213.0
e-Al/SiO2 (bad pair)
186.0
181.0
e-Au/Si (bad pair)
23.5
21.1
e-Au/SiO2 (equal pair)
17.0
16.6
e-Cu/Si (bad pair)
17.4
17.0
s-Cu/Si (bad pair)
74.2
72.4
s-Cu/Ta (bad pair)
112.0
109.0
s-Cu/SiO2 (equal pair)
32.3
32.3
4.3.3 Effect of deposition method

The interface strength of the sputtered Cu film and Si substrate (s-Cu/Si) is stronger than that of
the evaporated Cu film (e-Cu/Si). The yc of s-Cu/Si is about four times larger than that of e-Cu/Si
as shown in Table 5.
Conclusions
Delamination tests for eight materials consisting of thin films and a silicon substrate are
conducted in order to develop the quantitative comparison method of crack initiation strength from
the free edge among various pairs of the materials. The effects of several factors such as species of
thin film materials, oxidized interlayers and deposition processes on the interface strength are
investigated on the basis of concentrated stress near the edge. The results obtained are summarized
as follows.
(1) All the combinations of materials show brittle delamination where a crack is initiated at the
interface edge and propagates along the interface at once.
(2) For each material, the crack initiation is governed by the stress field near the interface edge.
(3) The orders of singularity, , in the materials are less than 0.07 (weak singularity), and the stress
field near the interface edge is almost constant in the atomic (nano-meter) scale. Thus, the
critical strength for the interface cracking is quantitatively represented by the concentrated
stress, yc, regardless of the magnitude of .
(4) Using the above critical stress near the interface edge, the strength of combinations tested is
summarized as follows; e-Al/Si > e-Al/SiO2 > s-Cu/Ta > s-Cu/Si > s-Cu/SiO2 > e-Au/Si > eCu/Si > e-Au/SiO2
380
(5) The interface strength of the evaporated Al film on the Si substrate (e-Al/Si) is about ten times
higher than those of the Au (e-Au/Si) and the Cu (e-Cu/Si) films.
(6) The adhesion strength between the sputtered Cu and Ta films (s-Cu/Ta) is 50% higher than that
of the Cu on Si substrate (s-Cu/Si). The sputtered SiO2 on the Cu film (s-Cu/SiO2) has less than
half the interface strength of the sputtered Cu film on the Si substrate (s-Cu/Si).
(7) The interface strength of Al (Au) on Si substrate with the artificial SiO2 interlayer (e-Al/SiO2 (eAu/SiO2)) is about 20% (30%) lower than that without the SiO2 interlayer (e-Al/Si (e-Au/Si)).
(8) The interface strength of the sputtered Cu film on Si substrate (s-Cu/Si) is about four times
stronger than that of the evaporated one (e-Cu/Si).
References
[1] J.W. Hutchinson., Z.Suo, Mixed mode cracking in layered materials, Advances in applied mechanics 1992; 29:
pp.63-191
[2] D.B. Marshall, A.G. Evans, Measurement of adherence of residually stressed thin films by indentation Mechanics
of interface delamination, J. Appl. Phys 1984; 56: pp. 2632-2638.
[3] Bagchi A., Lucas G.E., Suo Z. and Evans A.G., A new procedure for measuring the decohesion energy for thin
ductile film on substrates, Journal of Materials Research 1994; 9: pp. 17341741.
[4] Dauskardt R.H., Lane M., Ma Q. and Krishna N., Adhesion and debonding of multi-layer thinfilm structures,
Engineering Fracture Mechanics 1998;61: pp. 141-162.
[5] De Boer M.P., Kriese M.D. and Gerberich W.W., Investigation of a new fracture mechanics specimen for thin film
adhesion measurement, Journal of Materials Research 1997; 12: pp. 2673-2685.
[6] M.D. Kriese, W.W. Gerberich, Quantitative adhesion measures of multilayer films- Part I. Indentation mechanics,
Journal Materials Research 1999; 14(7) : pp. 30073018.
[7] Kamiya S., Kimura H., Yamanobe K., Saka M. and Abe H., A new systematic method of characterization for the
strength of thin films on substrates - evaluation of mechanical properties by means of film projection, Thin Solid
Films 2002; 414; pp. 91-98.
[8] Gao L., Nakasa K., Kato M. and Nishida H., Evaluation of interfacial fracture toughness of thermal barrier coating
under heat cycles, Key Engineering Materials 2003; 243: pp. 267-272.
[9] M. Omiya, H. Inoue, K. Kishimoto, Evaluation of Interfacial Strength by Multi-Stages Peel Test, Key Engineering
Materials 2004; 483: pp.261-263.
[10] E.D. Reedy Jr., T.R. Guess, Comparison of butt tensile strength data with interface corner stress intensity factor
prediction, Int. J. Solids Struct. 1993; 30: pp. 2929-2936.
[11] T. Hattori, S. Sakata, G. Murakami, A stress singularity parameter approach for evaluating the interfacial
reliability of plastic encapsulated LSI devices, Journal of Electronic Packaging 1989; 111: pp. 243-248
[12] Kinbara, A., Kusano, E., Kamiya, T., Kondo, I. and Takenaka, O., Evaluation of adhesion strength of Ti films on
Si (100) by the internal stress method, Thin Solid Films 1998; 317: pp. 165168.
[13] W. Xie, S.K. Sitaraman, An experimental technique to determine critical stress intensity factors for delamination
initiation, Engng. Fract. Mech. 2003; 70: pp. 1193 1201.
[14] H. Hirakata, T. Kitamura, Y. Yamamoto, Evaluation of interface strength of micro-dot on substrate by means of
AFM, Int. J. Solids Struct. 2004; 41: pp. 3243-3253
[15] T. Kitamura, H. Hirakata, T. Itsuji, Effect of residual stress on delamination from interface edge between nanofilms, Engng. Fract. Mech. 2003; 70; pp. 2089-2101.
[16] T. Kitamura, T. Shibutani, T. Ueno, Crack initiation at free edge of interface etween thin films in advanced LSI,
Engng. Fract. Mech. 2002; 69:pp. 1289-1299.
[17] F. Shang, T. Kitamura, H. Hirakata, I. Kanno, H. Kotera, K. Terada, Experimental and theoretical investigations
of delamination at free edge of interface between piezoelectric thin films on a substrate, Int. J. Solids Struct. 2005; 42:
pp. 1729 - 1741
[18] Bogy D.B., Edge bonded dissimilar orthogonal elastic wedges under normal and shear stress, Journal of Applied
Mechanics 1968;35: pp. 146154.
[19] Bogy D.B., Two edge bonded elastic wedges of different materials and wedge angles under surface tractions,
Journal of Applied Mechanics 1971; 38: pp. 377-386
[20] Yuuki R., Xu Q.J. and Kayama N., Stress singularity at the interface edge and the tip of an interface crack under
thermal stresses, Advances in Electronics Packaging, ASME EEPVol. 1993; 4-1: pp. 119-125.
381
Effect of Frequency on Fatigue Crack Growth along Interface

between Copper Film and Silicon Substrate
Abstract
The mechanical crack growth along the interface between a submicron film (Cu) and a
substrate (Si) under fatigue is experimentally investigated under two cyclic frequencies of 0.1
Hz and 1 Hz in a laboratory air (455% R.H.). A modified four-point bend specimen, which
has only one interface crack to facilitate the control of crack growth, is proposed for the tests.
The results reveal that the clear interface crack between Cu and Si grows under the cyclic
load. The crack growth rate, da/dN, is governed by the stress intensity factor range, Ki, and
the sigmoidal relationship consisting of three stages are observed in the da/dN - Ki curve;
the threshold, the stable growth and the critical growth. The region of subcritical crack
growth is narrow due to the high yield stress of the thin film and the constraint by the
substrate. The crack growth rate accelerates as the frequency decreases in the stable growth
region. This is due to the effect of stress corrosion by the humidity in air.
Keywords: fatigue; crack propagation; fracture mechanics; thin film, interface
Introduction
Microelectronics devices consist of small structured materials such as thin films. One of
the major failures observed in these devices is delamination because the stress concentrates at
an interface between different materials due to the mismatch of deformation. Many studies
have been conducted to evaluate the strength of an interface [1-11] between a thin film and a
substrate. However, most of the studies have focused on evaluation of the interface strength
under the monotonic loading [1-8], and few studies have investigated on subcritical crack
growth under the cyclic loading [9-11]. In particular, fatigue crack growth along the interface
between a submicron-thick film and a substrate has not been reported.
A crack propagates in bulk materials or along an interface due to cyclic loading, and the
fatigue crack growth rate is characterized by the stress intensity factor range or the energy
release rate range [12, 13]. Although the fatigue crack growth is essentially a time
independent phenomenon which is caused by irreversible dislocation motion, it is well known
that the crack growth is greatly influenced by the stress corrosion and shows the dependence
on the loading frequency [14-17]. However, the effect of loading frequency on the crack
growth along a thin film/substrate interface has not been investigated so far.
The fatigue crack growth behavior along the interface between a submicron metal film and
a substrate differs from that in bulk bi-materials, since the dislocation motion is restricted in
the thin layer of metal film and the plastic deformation of the film is constrained by the stiff
382
substrate. The purpose of this study is to elucidate the crack growth behavior between a
submicron copper (Cu) film and a bulk silicon (Si) substrate under fatigue loading. Moreover,
the effect of loading frequency on the fatigue crack growth is also investigated.
2.1 Material and specimen

Figure 1 shows a TEM micrograph of the multi-layered material tested. After a Cu film
with the thickness of 200 nm is sputtered on a (100) surface of a Si substrate, a silicon nitride
(Si3N4) film of thickness 500 nm is deposited on the Cu film by sputtering to restrain the
large plastic deformation of the Cu film during the test.
The modified four-point bend specimen [11] shown in Fig. 2 is adopted for the test.
Because the standard four-point bend specimen has two symmetrical cracks, the same
singular stress field theoretically takes place near both crack tips under an applied load.
However, in reality, it is difficult to control the equal growth of the cracks because a slight
unbalance in the loading brings about the asymmetrical singularities. The modified specimen,
which has only one interface crack, enables us to control the stable crack growth.
Si3N4
Cu
Si
200 nm
Fig. 1 TEM micrograph of a multi-layered material tested
Stainless steel (thickness: 590 m)

Epoxy (thickness: 10 ~ 16 m)
Si3N4(thickness: 0.5 m)
Cu (thickness: 0.2 m)
Si (thickness: 500 m)
Interface of interest
(Cu/Si)
11
11
18
P/2
P/2
17
a: Crack length
Specimen width:
4.2 ~ 5.1 mm
21
42
Fig. 2 Modified four-point bend specimen for multi-layered material Si3N4/Cu/Si
383
P/2
P/2
Epoxy
Si3N4
Cu
Si
Crack
y
Cu
Crack
Si
x
100 nm
Fig. 3 FEM model and mesh division near crack tip
Table 1 Elastic constants

Material
Stainless steel
Epoxy
Si3N4
Cu
Si
Youngs modulus E, MPa

200
2.5
304
130
130
Poissons ratio
0.3
0.3
0.27
0.34
0.28
A plate of stainless steel is polished with emery papers and diamond paste, and the plate is
cleaned by ultrasonic vibration in acetone and isopropyl alcohol. After the plate is glued
carefully on the wafer with the Si3N4/Cu films by standard epoxy, the wafer is cut into
rectangular coupons using a dicing machine. The shape and the size of specimens are given in
Fig. 2. The thickness of epoxy layer and specimen width is carefully examined in each
specimen, and they are used for the stress analysis.
The stress in the modified four-point bend specimen is numerically analyzed under the
plane strain condition by a commercial FEM code, ABAQUS 6.5. Elastic analyses are
conducted for a specimen with different crack length, where the elastic constants used are
listed in Table 1. Since carefully deposited thin films have almost the same elastic modulus as
that of their bulk [18-19], we use it for the Cu film. Figure 3 shows the boundary condition
and mesh division near the crack tip in an FEM model. The region near the crack tip is
carefully divided into fine elements with the smallest element size of 2 nm to precisely
analyze the stress field on the basis of continuum mechanics concept. Figure 4 shows the
dependence of the total stress intensity factor, Ki, and the phase angle, = tan-1(K2/K1), and
on the crack length, a. Here, the stress field near the tip of an interface crack is characterized
by the complex stress intensity factors, K1 and K2, which is in the form,
384
y + i xy =
K 1 + iK 2
(2r )1 / 2
r i
lk
(1)
and the total stress intensity factor, Ki, is defined as follows.
K i = K12 + K 22 = lim 2r y2 + xy2

r 0
(2)
1.2
100
1.0
80
0.8
60
0.6
40
0.4
Ki/Ki(a=2mm)
0.2
20
Inner loading point
6
10
8
12
Crack length a, mm
14
16
Phase angle = tan-1(K2/K1), deg.
Normalized stress intensity factor Ki /Ki( a=2mm)
Here, y and xy are the normal and the shear stresses, respectively. r is the distance from the
crack tip. is the oscillation index and lk is the characteristic length which can be set
arbitrarily [20]. In Fig. 4, Ki is normalized by the magnitude of Ki at a = 2 mm. The figure
clearly indicates that and Ki are almost constant under a constant applied load if the crack
tip is sufficiently inside the inner loading point. is about 47 degrees for the characteristic
length, lk, of 1 nm. Meanwhile, Ki decreases when the tip is in the vicinity of the inner
loading point.
0
18
Fig. 4 Dependence of phase angle and normalized stress intensity

factor on crack length in the modified four-point bend specimen
Since the interface between a thin film and a substrate is usually brittle, the introduction of
a pre-crack with a desired length is not an easy task. In each specimen, however, the stress
intensity factor prominently decreases as the crack tip approaches the inner loading point as
shown in Fig. 4 and thereby a well-controlled interfacial pre-crack can be introduced by
decreasing the length between the inner loading points in Fig. 2.
Residual stress in thin films affects the crack growth behavior in many specimen
configurations in general. However, in the four-point bending specimen, stress relaxation of
the thin film due to debonding is essentially constrained by the stiff substrates. Therefore, the
effect on the crack propagation is small [1].
385
2.2 Fatigue test

A cyclic load is applied to the specimen by a micro fatigue machine with an electromagnetic actuator (Shimadzu MMT-100N) as illustrated in Fig. 5. The load, P, and the
displacement, u, are monitored at the loading point during the tests by a load cell and a
differential transformer, respectively. The resolution of the load is 0.01 N which corresponds
to the stress intensity factor of Ki 0.002 MPam1/2. All the tests are performed in laboratory
environment at relative humidity (RH) of 45 5% at room temperature of 295 2 K.
Fatigue tests are conducted under constant stress amplitude of a sinusoidal waveform at
the frequencies of 1.0 Hz and 0.1 Hz with the load ratio, R = Pmax/Pmin, of 0.54. Here, Pmax
and Pmin are the maximum and minimum loads, respectively. Test conditions are listed in
Table 2. The crack length, a, is evaluated by the compliance method [11]. The relationship
between the compliance, , and a is calculated by the FEM for each specimen.
Fig. 5 Schematic illustration of a testing machine

The fracture toughness is also evaluated by applying a monotonic load on the specimen.
Table 2 Loading frequency and applied stress intensity factor range
Specimen
number
A-1
A-2
A-3
A-4
A-5
A-6
B-1
B-2
B-3
B-4
Loading
frequency, Hz
1.0
0.1
Ki, MPam0.5
0.20
0.22
0.26
0.26
0.27
0.37
0.19
0.21
0.25
0.31
386
14
0.0385
Max
12
10
0.0335
0.0309
da/dN = 1.91 x 10-5 m/cycle
0.0284
Min
0.0259
Specimen A-2
45 5% RH
1.0 Hz
Ki = 0.22 MPam0.5
0.0233
0.0208
200
400
600
800
1000
Crack length a, mm
Compliance , m/N
0.0360
2
0
1200
Number of cycles N, cycles

Fig. 6 Relationship between compliance (crack length) and number of cycles in A-2
Results and discussion
3.1 Fatigue crack growth behaviour

Figure 6 shows the relationship between the compliance, , and the number of cycles, N,
in specimen A-2 (Ki = 0.22 MPam0.5) under the cyclic load with the frequency of 1.0 Hz.
The converted crack length, a, is indicated on the right side as a reference. The crack begins
to propagate just after the test is started, and the crack growth rate, da/dN, is almost constant
up to approximately a = 10 mm (450 cycles). This is consistent with the region where the
applied Ki is constant. After that, the crack growth slows down and stops. This is because
the crack approaches the inner loading point where a = 12.4 mm and the magnitude of Ki
decreases rapidly as shown in Fig. 4. In the linear region, da/dN is evaluated from the slope
of the a N curve as about 1.91 x 10-5 m/cycle. The fluctuation is estimated as 1.23 x 10-5 <
da/dN < 2.32 x 10-5 m/cycle by the chained and the dashed lines shown in Fig. 6.
On the other hand, Fig. 7 shows the crack growth curve of specimen A-1 (Ki = 0.20
MPam0.5) under the cyclic load with the frequency of 1.0 Hz. The magnitude of is almost
constant even after 106 cycles and this indicates no crack growth. Thus, there is a threshold in
the crack growth, and the stress intensity factor range Ki = 0.20 MPam0.5 is under the
threshold. Though a threshold normally exists in the fatigue crack growth in bulk metals and
along interfaces in a bulk [9, 21, 22], it is rarely reported along an interface between a
submicron film and a substrate.
After the fatigue tests, the fracture surfaces of both the film and substrate sides are
examined by Auger electron spectroscopy (AES). As typically shown in Fig. 8, only Cu
peaks (56, 764, 836 and 916 eV) are confirmed on one side while Si (89 and 1614 eV) is
387
found on the other side for all specimens tested. Thus, the crack propagates perfectly along
the interface between Cu and Si.
14
Specimen A-1
45 5% RH
1.0 Hz
Ki = 0.20 MPam0.5
Compliance , m/N
0.0375
0.0349
12
10
0.0322
0.0296
0.0270
0.0243
0.0217
Crack length a, mm
0.0401
0
10
Number of cycles N, x 105 cycles
Intensity
Fig. 7 Relationship between compliance (crack length) and number of cycles in A-1
Cu Cu
Cu
0
Cu
500
1000
Energy of electron, eV
1500
2000
Intensity
(a) Cu film side
Si
Si
0
500
1000
Energy of electron, eV
1500
2000
(b) Si substrate side

Fig. 8 Spectra of Auger electron on fracture surfaces
388
3.2 Effect of frequency on fatigue crack growth

Figure 9 shows the relationships between da/dN and Ki at the loading frequency of 1.0
and 0.1 Hz. Each curve is expressed as a typical sigmoidal curve with three distinct regions;
the threshold, the stable growth and the critical growth. The stable crack growth region is
described by the Paris law,
da/dN = CKim
(3)
The magnitudes of m are 5.11 and 7.76 at the frequencies of 1.0 Hz and 0.1 Hz, and those
of C are 0.0567 and 21.9, respectively. The crack growth greatly accelerates with a decrease
of the loading frequency. This indicates that some environmental effect due to humidity in air
at the crack tip plays an important role on the crack growth [23, 24]. Although clarifying the
mechano-chemical influence on interface cracking is an important issue, this needs to be
studied separately and is a future work.
Fatigue crack growth rate da/dN, m/cycle
The upward arrow indicates the equivalent fracture toughness, (1 - R)Kic = 0.47 MPam0.5,
which is Ki at Kimax = Kic. Here, Kic is evaluated by a monotonic loading test on the modified
four-point bend specimen as Kic = 1.02 MPam0.5. The downward arrow indicates the
threshold, Kith, where the crack does not propagate in 106 cycles. The threshold is almost
independent of the loading frequency.
Kic
10 -1
10
-2
R = 0.54
45 5% RH
10 -3
10 -4
0.1 Hz
10 -5
1.0 Hz
10 -6
0.1
Kith
0.2
0.3
0.4
0.80.9
0.91.0
0.5 0.6 0.7
0.70.8
Stress intensity factor range Ki, MPam0.5

Fig. 9 Relationship between fatigue crack growth rate and stress intensity factor
range at frequencies of 0.1 Hz and 1.0 Hz
Figure 10 plots the Mises stress distribution along the Cu/Si interface at a = 8 mm at the
maximum load in the fatigue test of Ki = 0.22 MPam0.5 in specimen A-2. As the yield stress
of submicron-thick Cu film is about one order higher than that of bulk Cu ( 70 MPa), it
reaches about 600 MPa [25, 26]. The region where the stress exceeds 600 MPa is about 60
nm and this is much smaller than the region dominated by the singular stress field. Thus, the
crack propagates under the small scale yielding condition.
389
Von Mises stress along interface Mises, MPa
10 4
Specimen A-2
a = 8 mm
Ki = 0.22 MPam0.5
10 3
10 2
10 1
10 0
101
10 2
10 3
10 4
Distance from interface crack tip r, nm

Fig. 10 Von Mises stress distribution along the interface between Cu film and Si
substrate at a = 8 mm
Crack growth rate da/dt, m/s
The crack growth rate per time, da/dt, is plotted against the maximum stress intensity
factor, Kimax, in Fig. 11. There is no difference among the da/dt - Kimax relationships, and this
indicates that the crack growth is purely time-dependent. The effect of cyclic loading is little,
and it suggests that the environmental effect is responsible for the crack growth.
10
-2
10
-3
10
-4
10
-5
10
-6
10
-7
Kimaxc
R = 0.54
45 5% RH
0.1 Hz
1.0 Hz
0.2
Kimaxth
0.3
0.4
0.5 0.6 0.7 0.80.9

0.8 0.9 1
Maximum stress intensity factor Kimax, MPam 0.5

Fig. 11 Relationship between crack growth rate and maximum stress intensity factor
390
Conclusions
Fatigue crack growth along the interface between a submicron film and a substrate has
been rarely reported. In this study, the behavior of interface crack growth between the
submicron Cu film and the Si substrate under fatigue is experimentally investigated using a
modified four-point bend specimen with a crack, where the stress intensity can be easily
controlled. The results are summarized as follows.
(1) The interface crack grows perfectly along the interface between the Cu film and the Si
substrate under the cyclic loading.
(2) da/dN - Ki curve obtained shows the sigmoidal relationship consisting of the threes
stages: threshold, stable growth and critical growth.
(3) In the stable growth region, da/dN at the frequency of 0.1 Hz is about one order of
magnitude higher than that at the frequency of 1.0 Hz for an equal magnitude of Ki. The
threshold is almost independent of the loading frequency.
(4) The fatigue crack growth rate per time, da/dt, is correlated well with the maximum stress
intensity factor, Kimax, regardless of the loading frequency. This indicates that the crack
growth is caused by the environmental effect due to humidity in air at the crack tip.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Dauskardt, R.H., Lane, M., Ma, Q., Krishna, N., Adhesion and debonding of multi-layer thin film
structures, Engineering Fracture Mechanics 1998; 61: pp.141162.
De Boer, M.P., Kriese, M.D., Gerberich, W.W., Investigation of a new fracture mechanics specimen for
thin film adhesion measurement, Journal of Materials Research 1997; 12: pp.26732685.
Gao, L., Nakasa, K., Kato, M. and Yamamoto, Y., Evaluation of interfacial toughness of thermal barrier
coating under heat cycles, Key Engineering Material 2003; 243: pp.267-272.
Hirakata, H., Kitamura, T. and Yoshitake, Y., Evaluation of interface strength of micro-dot on substrate
by means of AFM, International Journal of Solids and Structures 2004; 41: pp. 3243-3253.
Kinbara, A., Kusano, E., Kamiya, T., Kondo, I. and Takenaka, O., Evaluation of adhesion strength of Ti
films on Si (100) by the internal stress method, Thin Solid Films 1998; 317: pp.165168.
Kitamura, T., Shibutani, T. and Ueno, T., Crack initiation at free edge of interface between thin films in
advanced LSI, Engineering Fracture Mechanics 2002; 69: pp.12891299.
Kitamura, T., Hirakata, H. and Itsuji, T., Effect of residual stress on delamination from interface edge
between nano-films, Engineering Fracture Mechanics 2003; 70: pp.20892101.
Shang, F., Kitamura, T., Hirakata, H., Kanno, I., Kotera, H. and Terada K., Experimental and theoretical
investigations of delamination at free edge of interface between piezoelectric thin films on a substrate,
International Journal of Solids and Structures 2005; 42: pp.17291741.
McNaney, J.M., Cannon, R.M. and Ritchie, R.O., Fracture and Fatigue-crack Growth along Aluminumalumina Interfaces, Acta Materialia 1996; 44: pp.4713-4728.
Mutoh, Y., Xu, J.Q., Miyashita, Y., Bernardo, G.G. and Takahashi, M., Interface Fatigue Crack
Propagation in Ceramic Thermal Barrier Coating, JSME International A 2003; 46: pp.403-408.
Hirakata, H., Kitazawa, M. and Kitamura, T., Fatigue crack growth along interface between metal and
ceramics submicron-thick films in inert environment, Acta Materialia 2006; 54: pp.89-97.
Hutchinson, J.W. and Suo, Z., Mixed mode cracking in layered materials, Advances in Applied
Mechanics 1992; 29: pp.63-191.
Suo, Z., Hutchinson, J.W., Sandwich test specimens for measuring interface crack toughness, Materials
Science and Engineering A 1998:107 pp.135-143.
Mayer, H., Papakyriacou, M., Pippan, R. and Stanzl-Tschegg, S., Influence of loading frequency on the
high cycle fatigue properties of AlZnMgCu1.5 aluminium alloy, Materials Science and Engineering A
2001; 304: pp.48-54.
Sansoz, F. and Ghonem, H., Effects of loading frequency on fatigue crack growth mechanisms in /
Ti microstructure with large colony size, Materials Science and Engineering A 2003;356: pp.81-92.
391
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
Nakai, Y. and Hiwa, C., Effects of loading frequency and environment on delamination fatigue crack
growth of CFRP, International Journal of Fatigue 2002; 24: pp.161-170.
Morrissey, J., McDowell D.L. and Nicholas, T., Frequency and stress ratio effects in high cycle fatigue
of Ti-6Al-4V, International Journal of Fatigue 1999; 21: pp.679-685.
Ruud, J.A., Josell, D., Spaepen, F., Geer, A.L., A new method for tensile testing of thin films, Journal
of Materials Research 1993; 8: pp.112-117.
Espinosa, H.D., Prorok, B.C., Peng, B., Plasticity size effects in free-standing submicron polycrystalline
FCC films subjected to pure tension, Journal of the Mechanics and Physics of Solids 2004;52: pp..667689.
Ikeda, T., Miyazaki, N. and Soda, T., Mixed mode fracture criterion of interface crack between
dissimilar materials, Engineering Fracture Mechanics 1998; 59: pp.725-735.
Chen, D.L., Weiss, B. and Stickler, R., Effect of stress ratio and loading condition on the fatigue
threshold, International Journal of Fatigue 1992; 14: pp.325-329.
Fleck, N.A., Kang, K.J. and Ashby, M.F., The cyclic properties of engineering materials, Acta
Metallurgica et Materialia 1994; 42: pp.365-381.
Magnin, T., Recent Advances for Corrosion Fatigue Mechanisms, ISIJ International 1995; 35: pp.223233.
Lane, M.W., Snodgrass, J.M., Dauskardt, R.H., Environmental effects on interface adhesion,
Microelectronic Reliability 2001; 41: pp.1615-1624.
Vinci, R.P., Zielinski, E.M. and Bravman, J.C., Thermal strain and stress in copper thin films, Thin
Solid Films 1995; 262; pp.142-153.
Balk, T.J., Dehm, G. and Arzt, E., Parallel glide: unexpected dislocation motion parallel to the substrate
in ultrathin copper films, Acta Materialia 2003; 51: pp.4471-4485.
392
Two Simple Flat Shell Elements for Elastic Analysis

of Thin Shell Structures
*
Nguyen-Ngoc Duong , Nguyen-Xuan Hung, Nguyen-Van Hieu

+
and Nguyen-Dang Hung
*
HCMC University of Architecture, Vietnam
Division of Computational Mechanics, Department of Mathematics and Informatics, University of

Natural Sciences-VNU-HCM, Vietnam
+
Abstract
This paper presents the theory and application of flat triangular and quadrilateral shell elements to
solve the problems of thin shell. These elements are built from the membrane element with drilling
degrees of freedom and the conforming plate bending element. They can simulate the geometrical
shape of the shells. They also solve two difficulties relating to coplanar elements around a common
node such as the relative compatibility of neighbouring element and the singularity of stiffness
matrix. The results of numerical investigations showed that these elements are convergent very
well.
Keywords: flat triangular shell element, flat quadrilateral shell element, thin shell, error estimation
FEM
Introduction
Structures of thin shells have been extensively applied in engineering practices. Although analytical
solutions exist for some simple shell structures, even an elastic analysis on a general shell structure
may be very complicated without aide of effectively numerical tools. One of the most popular
numerical methods is finite element method (FEM).
In FEM, shell structures are normally modelled into curved elements which give a good
approximation to the actual shell curvature. Curved elements were constructed provided membranebending coupling within the element. However, essential difficulties exist in formulating consistent
shell theories (the force-displacement relationships of curved shell elements) and an exact
representation of rigid-body motions [6]. These drawbacks can be overcame by employing an
assemblage of flat shell elements. A flat shell element can reject the element curvatures as required
in the case of a curved shell element. The formulation is normally simple and easy to code. The cost
of the stiffness formulation for a flat shell element is usual to be much less than that for a curved
shell element. Therefore, flat shell elements can be more suitable for extending to nonlinear
analysis. In flat shell elements, it is possible to obtain a good approximation of the shell geometry
by an assemblage of planar elements. The shell surface is replaced by a finite number of flat plates,
which is of triangular or quadrilateral element. Analysis of shell structures by the finite element
method was then based on the pure membrane theory with the triangular or quadrilateral constant
strain elements. A satisfactory shell element must contain both the plane stress and the plate
bending stiffness. In case the relative displacements for a planar shell element are small, the
membrane action and the bending action are uncoupled within each element. The stiffness matrix
for the shell element therefore may be obtained by superimposing two independently derived
stiffness matrices:
393
1. The stiffness matrix for a plane stress finite element.

2. The stiffness matrix for a plate bending finite element.
Details on evaluating the stiffness matrix of membrane elements with the Constant Strain Triangle
(CST) and Quadrilateral (Q4) can be found in many literatures. A procedure to constitute the flat
thin shell element is completely similar to the content of the chapter 6 in Reference [6]. Here we
focus on two bending elements that is the bending triangular element Hseigh-Clough-Tocher
(HCT) [3] and the bending quadrilateral element Linear Curvature Compatible Triangle (LCCT12) [2, 4, 10]. The HCT element was developed by T. K. Hsieh, R.W.Clough and J. Tocher for the
thin plate bending. Herein the difficulties involved in conforming displacement approaches of the
plate bending which has to satisfy the displacement and the slope compatibility conditions had
overcomed. A conforming element of the bending quadrilateral shape with macro elements triangle
was first proposed by B. Fraeijs de Veubeke [2]. A cubic deflection field is established with each
macro element delimited by the edges of quadrilateral and its diagonals. However, the cost and the
complex computation of the stiffness formulation by this way are normally increased. This paper
reconstructs two flat thin shell elements by associating membrane elements and bending elements.
The aim of the following work is to reformulate two bending elements using the cubic deflection
field in area coordinates for each sub-triangular element. This way is simple and effective.
Moreover, a perspective of the presented work is to associate with an equilibrium model [14] in
obtaining a global error based on a dual analysis [17] for the thin shell structures.
The outline of the paper is organized as follows. In the second section we present the compatible
bi-cubic Hsieh-Clough-Tocher (HCT) macro elements. Section 3 illustrates the LCCT-12 Linear
Curvature Compatible Triangle Element with 12 degrees of freedom. Several numerical
investigations are given in the section 4 to compare the accuracy and convergent rate of HCT and
LCCT-12 elements. Finally, section 5 will remark some conclusions and further works.
The conforming HCT element
The displacement function can be written in terms of the nodal degrees of freedom and shape
functions as follows:
w = Nq
(1)
where N are Hermitian shape functions and q are appropriate parameters.
To obtain a compatible triangular element with cubic expansion, the triangle is divided into three
sub-triangles (see figure 1a). The point 0 is the centroid of the original triangle, and the subtriangles are numbered corresponding to the opposite corner number. The sub-triangle number is
identified in the algebraic expressions by a superscript. Figure 1b shows the renumbering scheme
for typical sub-triangle 1.
The nodal displacement degrees of freedom to be considered in the stiffness matrix of the complete
element are shown in Figure 2.
The displacement expansion w(x, y) is now assumed over each sub-triangle
w (k ) (x, y) = (k ) (x, y)q (k ) , k = 1, 2,3
(2)
where for sub-element 1

T
q (1) = w 2 x 2 y2 w 3 x3 y3 w 0 x 0 y0 5
(3)
is the vector comprising the 10 displacement components. The set of 10 cubic Lagrangian
interpolation polynomials for a triangle are described in area coordinates = (1 , 2 , 3 ) as follows
394
w =
(4)
where
= 13 23 33 12 2
123
223
221
321
T = [ 1 2 3 4 5 6 7 8 9 10 ]
32 2 1 2 3
Figure 1: Triangular element partitioned into three sub-elements

From equation (3) the vector q of all nodal displacements of the complete element assembly is
T
q = w1 x1 x1 w 2 x 2 y2 w 3 x3 y3 4 5 6 | w 0 x 0 y0
= q TR | q TE
(5)
Figure 2: Complemented nodes for compatibility conditions

The displacements in a sub-element k may be expressed as:
q
)
) R
w (k ) () = N (k ) ()q = N (k
N (k
e
0
qE
(6)
where N (k ) () has the same form like N() , but it has additional zero components corresponding to
the nodal displacements of the complete assembly that is not associated with the sub-triangle k. The
)
)
partitions N (k
and N (k
represent the interpolation functions for the external and internal nodal
e
0
displacements, respectively. Equation (6) expresses the cubic displacement pattern in the three sub395
elements. Equation (2) does not represent an internally compatible displacement field. Introduce
points 7, 8 and 9 at the midpoints of sub-element interfaces and using condition of continuity, n w
at nodes 7, 8, 9. We establish, in matrix form, the compatibility requirements as follows:
q R + B 0q E = 0 q E = B o1Bq R = Cq R
(7)
where
[ B]3x12
(3)
B (1)
B o(1) + B (3)
7 + B7
o
(1)
(1)
(2)
(2)
= B8 + B8 , [ Bo ]3x3 = B o + B o
(3)
(3)
(2)
B (2)
B9 + B9
o + Bo
By introducing the compatibility constraint of equation (7) into equation (6)

(1)
w (1) N (1)
e + BN o C
(2) (2)
(2)
(8)
w = N e + BN o C q R
(3)
(3)
(3)
w N e + BN o C
The stiffness matrix of the triangle assembly is obtained directly by adding contributions of the
three sub-elements, as they are all expressed in terms of the same set of nodal coordinates. The
assembled triangular elements will yield optimal results for a given triangular element mesh
involving compatible cubic displacements.
We continue using the idea of static condensation to reduce the order of an algebraic system.
Specifically, we will apply this technique to the stiffness equation for the triangular in order to
eliminate the three internal degrees of freedom. If k represents the total stiffness matrix for the
triangular element, then partitioning k is
k
k = RR
k ER
k RE
k EE
(9)
and the corresponding load vector is
g T = g TR g TE
(10)
where the subscripts R and E on the block matrix partition refer to the terms of the triangles that
correspond to the remain degrees of freedom of nodes 1, 2, 3, 4, 5, 6 and the interior degrees of
freedom of nodes 0 in which degrees of freedom is eliminated, respectively. The force-displacement
relationship for element stiffness may be written in form of total energy as
k RE q R
g
k
T
T R
(11)
q TE RR
q
q
R
E

g
k ER k EE q E
E
Combining Eq.(7), (11) with the stationary of total energy to vary q R will lead to the new reduced
stiffness equation
=
1 T
q R
2
k RR q R = gR
where:
(12)
k RR = k RR + CTk ER + k REC + CTk EEC and g R = g R + CT g E
The final triangle has the twelve degrees of freedom corresponding to a deflection, two rotations at
each vertex and a slope at each mid-interface (Figure 3a). From the pragmatic point of view of data
preparation and computer analysis, the degrees of freedom of mid-side nodes as 4, 5, and 6 are
complicated for the analysis. However, if the normal slope is constrained to distribute linearly along
the side, a mid-side node may be removed. The degrees of freedom are no longer necessary and the
number of displacement components of element reducing to nine that lead to a reducing the rank of
396
the stiffness matrix. The triangle with nine degrees of freedom (Figure 3b), which is associated with
the nodes 1, 2, 3, is fully compatible with the constraint of linear distribution of slopes along
exterior edges.
Figure 3: Transformation to corner connectors
The conforming LCCT-12 element
The conforming LCCT-12 element is shown in Figure 4. It has 12 degrees of freedom with three
dofs at each node. The formulation of this element is different from other types of element in which
a sub-domain approach is used.
3 x3
w3
y3
n 7
4
w4
(2)
x 4
n 6
i
1
2
j
y4
n 8
y1
n 5
x1
y2
w2
x 2
n 6
(3)
0
(1)
(b)
w
(a)
Figure 4: The conforming quadrilateral element

The quadrilateral element is first divided into four sub-triangles by its diagonals as shown in Figure
4a. The sub-triangles are numbered in the algebraic expressions by a superscript. Figure 4b shows
the renumbering scheme for typical sub-triangle 1. In each sub-triangle, the normal displacement is
approximated by a complete cubic polynomial. The nodal displacement degrees of freedom to be
considered in the stiffness matrix of the complete element are shown in Figure 4a.
The displacement expansion w(x, y) is now assumed over each sub-triangle is similar to Eq. (2).
1
The vector q = q ( )
q ( 2)
q ( 3)
q ( 4 )
of all nodal displacements of the complete element
assembly is
397

T
q = w1 x1 y1 w 2 x2 y2 w 3 x3 y3 w 4 x4 y4 5 6 7 8 | w 0 x0 y0
= qTR | qTE
(13)
where q R , q E are the external and internal nodal displacements, respectively.

It was found more convenient to use area coordinates than to use oblique coordinates proposed by
B. Fraeijs de Veubeke. Using Eq.(4), the set of ten cubic Lagrangian interpolation polynomials for a
triangle are described in area coordinates = (1 , 2 , 3 ) .
Similar to set up the conforming HCT element, the compatibility constraints of the internal edges
restrained at the mid-nodes i,j, k have the form as follows:
q E = B 01Bq R = Cq R
(14)
where
[B]3x16
Bi(1) + Bi(2)
(3)
= B (2)
j + Bj
(3)
(4)
B k + B k
Bi0(1) + Bi0(2)
(2)
(3)
, [B0 ]3x3 = B j0 + B j0
(3)
(4)
Bk 0 + Bk 0
and
w (1)
( 2)
w
( 3) =
w
( 4)
w
N (e1)
N (e2)
N (e3)
N (e4)
N (01)C
N (02)C
qR
N (03)C
N (04)C
(15)
The following steps are similar to HCT element. Finally, we have a quadrilateral with twelve
degrees of freedom (Fig 5b) that are associated with nodal points 1, 2, 3, 4. The element is fully
compatible with the constraints of linear distribution of slopes along exterior edges.
Figure 5: Transformation to corner connectors
Numerical results
4.1 Problem 1
A barrel vault has radius of 25 ft, subtended angle of degrees, length of 50ft, and thickness of 3 in.
The structure has elastic modulus of 3 msi, Poissons ratio of 0.3 and weight of 90 lb/ft2. A loading
398
condition in z-direction is its weight. Two curved edges are assumed to be supported by rigid
diaphragms, and the other two edges are free. The figure of the structure is shown in Figure 6.
Because of symmetry, a quarter of the structure is modelled. Meshing the model with 32, 128, 200,
392, 512, 648 elements (for HCT element) (see Fig. 6a); 16, 64, 100, 196, 256, 324 elements (for
LCCT-12 element) (see Fig. 6b), we have a comparison for two kinds of element. The results are
given in the table 1.
y
B
C
C
Free
edge
x
L=
50
a)
R=2
5f
40
Diaphram
40
Diaphram
R=2
5ft
ft
Free
edge
L=
50
ft
b)
Figure 6: Illustration of barrel vault model
Table 1: The displacement (deflection) at the point C and the strain energy
Mesh
N0
4x4
8x8
10x10
14x14
16x16
18x18
Element Number
HCT
LCCT-12
32
16
128
64
200
100
392
196
512
256
648
324
Displacement (w) (ft)

HCT
LCCT-12
-0.199644 -0.279869
-0.257518 -0.292481
-0.270656 -0.295119
-0.284065 -0.297651
-0.287669 -0.298304
-0.290224 -0.298759
Strain energy
HCT
LCCT-12
9.0516e+003 1.3265e+004
1.2194e+004 1.4029e+004
1.2891e+004 1.4177e+004
1.3603e+004 1.4317e+004
1.3794e+004 1.4353e+004
1.393e+004 1.43785e+004
Vertical displacement at midsection
The convergence of deflection at the point C

0.05
HCT element
LCCT12 element
HYTCOQ element
10
HCT elements
LCCT12 elements
HYTCOQ elements
Exact solution
0
0.05
Displacement (ft)
The percentage error: (FEMexact)/exact
20
10
0.1
0.15
0.2
20
0.25
30
0.3
40
0.35
500
1000
1500
Number of DOFs
(a)
2000
2500
10
15
20
25
30
35
40
Degree
(b)
Figure 7: The convergence of the deflection at point C
399

Vertical displacement of LCCT12 element at midsection
Vertical displacement of HCT element at midsection
0.05
0.05
16 LCCT12 elements
100 LCCT12 elements
324 LCCT12 elements
Exact solution
0.05
Displacement (ft)
Displacement (ft)
0.05
0.1
0.15
0.2
0.1
0.15
0.2
0.25
0.25
0.3
0.3
0.35
32 HCT elements
200 HCT elements
648 HCT elements
Exact solution
10
15
20
25
30
35
40
0.35
10
Degree
25
1.65
x 10
FEM Extrapolation Strain Energy
Energy
1.3
1.2
1.1
1
HCT element
LCCT12 element
500
1000
1500
Number of DOFs
(c)
40
HCT energy
LCCT12 energy
ExtraHCT energy
ExtraLCCT12 energy
1.4
35
FEM and Extrapolation strain energy of conforming model
1.6
0.9
30
(b)
The convergence of energy
x 10
20
Degree
(a)
1.5
15
2000
2500
1.55
1.5
1.45
1.4
1.35
1.3
1.25
600
800
1000
1200
1400
1600
Number of dofs
1800
2000
2200
(d)
Figure 8: The convergence of displacement and strain energy

Figure 7a illustrates the convergence of the relative error of the displacement results at the point C
in the z-direction, and figure 7b shows vertical displacement at mid-section between three types of
element. The analytical solution has been calculated by Scordelis and Lo [8] (wC = -0.308 ft). The
HYTCOQ [1] gives better results than the others (with small DOFs) because DOFs of
HYTCOQ is higher than the others. However, LCCT-12 element also gives good result (see
Figure 8a). The error displacement at the point C between HYTCOO and LCCT-12 is little (1.5372%).
Figure 8c shows the convergence of the strain energy. The speed of convergence of LCCT-12
element is faster than HCT element. When mesh is refined, the energy strain of both elements is
convergent to the exact solution.
By applying the procedure of Richardsons extrapolation [15] for the displacement model, the
extrapolation of LCCT-12s energy strain is better than HCT. So, we can take the value
1.44792E+004 of LCCT-12 as a good approximation of the exact strain energy.
400
4.2 Problem 2
A hyperbolic paraboloid is clamped all along the boundary. The shell is subjected to a normal
pressure loading 0.1 N/cm2. It has length of 100 cm, height of 10 cm, and thickness of 0.8 cm. The
material has elastic modulus of 285000 N/cm2 and Poissons ratio 0.4. The figure of the structure is
shown in Figure 9. A quarter of the hyperbolic paraboloid is modeled due to symmetry.
Meshing the model with 32, 128, 512, 648 elements (for HCT element) (see Fig 9a), 16, 64, 256,
324 elements (for LCCT-12 element) (see Fig 9b); we have a comparison for two kinds of element.
The results are given in the table 2.
y
a=50cm
a=
50
cm
0c
m
a=
5
a=50cm
h=10cm
h=10cm
(a)
(b)
Figure 9: The model of hyperbolic paraboloid
Table 2: The displacement at the point A and the strain energy
0
Mesh N
4x4
8x8
16x16
18x18
Element number
HCT
LCCT-12
32
16
128
64
512
256
648
324
Displacement (w) (cm)

HCT
LCCT-12
-0.031193
-0.003995
-0.028458
-0.011431
-0.027680
-0.017051
-0.027627
-0.017765
Strain energy
HCT
LCCT-12
0.157138
0.125543
0.160499
0.141555
0.161458
0.152643
0.161553
0.154098
Figure 10a presents the convergence of the displacement at the point A. The analytical solution has
been calculated by C. Brebbia [9] (wA= -0.0245cm). The HYTCOQ [1] gives better results than
the others (with small DOFs). However, HCT element also gives better results than LCCT-12
element. A detail about the accuracy of HCT element and LCCT-12 element in the hyperbolic
paraboloid problem that using a triangular element is more suitable than a quadrilateral element was
explained in Zienkiewicz [6].
Figure 10b shows the convergence of the strain energy. The speed of convergence of HCT
element is faster than LCCT-12 element. When mesh is refined, the energy strain of both
elements is convergent to the exact solution. By applying the procedure of Richardsons
extrapolation for the displacement model, the extrapolation of the HCTs energy strain is better than
LCCT-12 element. So, we can take the value 0.162100 (HCT) as a good approximation of the
exact strain energy.
401
Figure 10a: Relative error deflection
Figure 10b: Convergence of strain energy
4.3 Problem 3
A panel of cylindrical thin shallow shell is clamped at all edges. It has radius of 762 mm, subtended
angle of 7.64 degrees, length of 76.2 mm, and thickness of 0.33 mm. The structure has elastic
modulus of 6.8948 N/m2, Poissons ratio of 0.33, and mass density 2657.3 kg/m3. Meshing the
model with 32 elements, 128 elements, 512 elements (for HCT element) (see Figure 11a); 16
elements, 64 elements, 256 elements (for LCCT-12 element) (see Figure 11b), we have a
comparison for two kinds of element. The results are given in the table 3 and 4.
y
B
z
z
D
x
a)
b)
Figure 11: The cylindrical thin shallow shell model
Table 3: The convergence of strain energy
Mesh N0
4x4
8x8
16x16
Element Number
HCT
LCCT-12
32
16
128
64
512
256
Strain energy
HCT
LCCT-12
1.017237E-09
1.044138E-09
1.033733E-09
1.040892E-09
1.038241E-09
1.039909E-09
402
Table 4: The results on the displacement at the point A

Mesh
N0
4x4
8x8
16x16
Element Number
HCT
LCCT-12
32
16
128
64
512
256
Displacement (w) (mm)

HCT
LCCT-12
Samcef solution
-3.436526E-04
-3.300356E-04
-3.327E-04
-3.315388E-04
-3.2793870E-04
-3.28E-04
-3.282081E-04
-3.273239E-04
-3.271E-04
Figure 12a: Displacement of point A
Figure 12b: Convergence of strain

energy
Figure 12a presents the convergence of the displacement at the point A. This problem is compared
with Samcef software [11]. LCCT-12 element has a very good behavior in comparison with
HCT element.
Figure 12b shows the convergence of the strain energy. The speed of convergence of LCCT-12
element is faster than HCT element. When mesh is refined, the energy strain of both elements is
convergent to the exact solution.
By applying the procedure of Richardsons extrapolation for the displacement model, the
extrapolation of LCCT-12s energy strain is better than HCT element. So, we can take the
value 1.039581E-009 of LCCT-12 as a good approximation of the exact strain energy.
Conclusions
This paper presents the finite element method for flat thin shells structure based on the coupling of
the membrane element and the bending element. Two simple flat thin shell elements are
reformulated by making use of an area coordinate. Reliability of the HCT element and the LCCT-12
element is obtained in comparing them with the analytical solutions. The numerical results show
that the calculated results can well apply into practice. Besides, the investigated results are also
compared with Samcef software [11] to certify reliable and admissible results.
This study also finds the estimated solution by using Richardsons extrapolation in order to obtain
the global error for the problems whose analytic solutions are not known. However, the procedure
of Richardsons extrapolation only finds a reference solution which depends on approximated finite
elements such as the discretized solution, the size of meshes. Therefore, in some cases, it is not sure
to be accepted an enough good solution that can replace the analytical solution in estimating the
error. A method that we want to mention in forthcoming work is a dual analysis. This method is no
need to know the exact solution, as it is always comprised between two the convergence curves.
The first curve is received from the displacement model above. The second is obtained from an
403
equilibrium model. Then the distance between the two curves is used to measure the error. Thus
Richardsons extrapolation is not necessary in the error analysis.
These proposed works will be treated in a further paper.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
Nguyen-Dang Hung, P Detroux, Ph. Falla and G. Fonder. Implementation of the duality in the finite element
analysis of shells, World congress on shell and spatial structure. Vol.2 September 1979.
B.M. Fraeijs de Veubeke, A conforming finite element for plate bending, Int.J.Solids Structures, 4, pp.95 to
108. Pergamon Press. Printed in Great Britain, 1968.
R.W.Clough and J.L. Tocher, Fininte element stiffness matrices for the analysis of plate bending, Proc. Conf.
Matrix Methods in Structural Mechanics, Wright-Patterson Air Force Base, Ohio, Otc., 1965.
Clough, R.W. and C.A., Felippa, A Refined Quadrilateral Element for the Analysis of Plate Bending, Pro. Of
the Second Conf. on Matrix Methods in Structural Mechanics, Wright Patterson Air Force Base, Ohio, Oct.
1968.
C S Krishnamoorthy, Indian Institute of Technology, Madras, Finite element analysis, Theory and
Programming ,Tata McGraw-Hill, New Delhi
O.C.Zienkiewicz, CBE, FRS, FREng and R.T. Taylor, The finite element method, Vol.2: Solid Mechanics
Butterworth-Heinemann, 2000
S. Timoshenko and S. Woinowsky-Krieger, Theory of Plates and Shells, McGraw-Hill, New York, 1959
A.C. Scordelis and K.S. Lo, Computer analysis of cylindrical shells, J.Am. Concr. Inst., 61, 539-561, (1964).
C. Brebbia, On Hyperbolic Paraboloidal Shells, PhD Thesis, Department of Civil Engineering, University of
Southampton (1967).
M. Petyt and C.C. Fleischer, University of Southampton., Vibration of curved structures using quadrilateral
finite elements, New York, Academic Press, 1972.
Samtech SA, User Manual M2-M3-M4-M10-M11-M2, Samcef User Manual, 2002.
Nguyen-Xuan Hung, Ladeveze-type compatibility error assessment for plate bending, Master thesis of EUEMMC, HoChiMinh City University of Technology, 12/2003.
Nguyen-Dang Hung, Tran-Thanh Ngoc, Analysis of cracked plates and shells using "metis" finite element
model, Finite Elements in Analysis & Design (Elsevier), Volume 40, Issue 8, pp. 855-878, May 2003.
Le-Van Canh, Nguyen-Xuan Hung, Nguyen-Dang Hung, Dual limit analysis of plate bending, Proc. of 3rd
Int. Conference on Advanced Computational Methods in Engineering, ACOMEN-2005, Ghent-Belgium.
L.F. Recharson, The approximate arithmetical solution finite differences of physical problems, Trans.
Roy.Soc.(London) A210, pages 307-357,1910.
P. N. Poulsen and L. Damkilde, A flat triangular shell element with loof nodes, Int. J. numer. methods eng.,
39, 3867-3887, 1996.
J.F. Debongnie, H.G. Zhong and P. Beckers, Dual analysis with general boundary conditions, Comput.
Methods Appl. Mech. Engrg., 122:183-192, 1995.
404
Collection papers from Prof. Nguyen-Dang Hung's former students
INDEX
A
active balancing, 218, 221-222, 225-228
arbitrary Lagrangian Eulerian formulation,
104-105, 116
B
beam-column, 294
bolted joints, 294
Bui-Cong Thanh, ii, 3, 30-31, 37, 294
Bui-Quoc Tinh, 3, 27, 34-35, 207, 217
C
Chau-Dinh Thanh, ii, vi, 3, 33, 280
cold roll forming, 104, 110, 113, 116
crack
circumferential, 119, 364-365
delamination, 43, 45-46, 49-50, 52, 166, 373-374,
376
initiation, 53, 119, 268-269, 271, 275-278, 311,
371-372, 376-378, 380-381, 391
fatigue growth, 25-26, 30, 193, 286, 290-291,
382-383, 387, 389, 391-392
plane, 289
propagation, 33, 192, 270, 277-278, 280, 287, 311,
371, 382, 385, 391
surface, 30, 184-185, 280-285, 290, 292
D
D. Crpin, 43, 53
Dang-Dinh Thi, 3, 32, 237, 253
delamination, 3, 13-14, 18, 25-27, 30-34, 43, 45-46,
49-50, 52-54, 156, 158-159, 165-167, 254, 268, 272,
275, 278, 371, 373-374, 376-377, 379-382, 391-392
Doan-Duc Bao, i, 3, 32-33
Do-Van Truong, 3, 33, 268, 279, 371, 382
Dunant C., 27
Duong-Thuy Phuong-Khanh, 1, 10
E
E. Ghorbel, 329
E. Van. Walle, 118
effective properties, 55, 58-59, 329
elastomer membrane, 255-256, 259-262, 267
element
conforming, 194, 199, 340, 394
equilibrium, 20, 27, 62-63, 71, 76, 194-195, 206,
340, 348, 352-353
flat quadrilateral shell, 393
flat triangular shell, 393, 404
hybrid, 1, 45, 49, 51, 156, 159, 253
metis, 1, 26-28, 34-35, 45, 49, 156, 163, 167,
238-239, 246-249, 253-254
energy release rate, 25, 51, 53, 156, 162-163, 165-167,
189, 273, 287, 289, 306, 371, 382
error
estimate, 27, 206, 229-230, 236
indicator, 229-233
criteria, 233
extrusion, 318, 324, 326
F
fatigue, 3, 13-14, 25-26, 30, 37, 119, 142, 145-153,
176-177, 193, 218, 280, 286-277, 290-293, 304,
307, 310-317, 339, 382-383, 386-387, 389,
391-392
fracture mechanics, i, 2-3, 12-13, 17-21, 24, 30, 34,
36-38, 43, 46, 53-55, 62, 118-120, 123, 134-135,
138-140, 156, 163, 167-168, 183-184, 193, 207, 237,
256, 268, 279-280, 293, 304, 316-317, 318, 340,
369, 371, 381-382, 391-393
fracture toughness, 118-119, 122-123, 132-133, 135,
139-140, 304, 310, 316-317, 371, 381, 386, 389
G
Gry de Saxc, i, vi, 1, 3, 10, 23-24, 28-30, 35-37, 43
53-54, 167, 182, 193
Guidoum A., 27
H
Hang-Minh Dinh, 304
high-accuracy, 237
Hiroyuki Hirakata, 268, 371, 382
Hoang Tuy, vi
Hoang-Van Long, 3, 168, 183
I
indentation, 255-256, 261-262, 267, 318, 321-323, 381
influence coefficients, 199, 218, 220, 222, 225-227
interface
edge, 268-269, 271, 273, 275, 277-278, 371-372,
374, 376, 378-381, 391
strength, 268-269, 278, 371-372, 376-377, 379-382,
391
Ionescu, 55
J
Jaime Lara, 255
J.F. Debongnie, 194, 206, 207, 217, 346, 353, 404
Jean-Philippe Ponthot, i, 104, 116-117
junctions, 43-44, 53
L
L. Papeleux, 104, 116
Le-Dinh Tuan, ii, 218, 228
Le-Hoai Long, 34, 184, 193
Le-Van Canh, 34, 340, 353, 404
lifetime prediction, 280
locking
inter, 16
phenomenon, 71
Poisson, 63
volumetric, 62-63, 67, 71, 76
loss of constraint, 118, 120, 123, 133-136, 138-139
M
M. Bringhier, 55
M. Pyrz, 43, 53
Mai Chi, ii
Marc Duflot, i, 26, 32-34, 207
405

Marc Scibetta, i, 3, 30, 118, 304, 307, 308, 311-312,
316-317
Matthieu Domaszewski, i
meshes
adaptive, 57, 203, 229, 381
distortion, 108, 237-238, 246, 248-249, 253
refinement, 46, 58-59, 120, 203, 350, 352
metal cutting, 104-105, 109, 113, 117
method, analysis
dual, 24, 26, 34, 87, 182, 194-196, 198-199,
201-206, 217, 340, 354-355, 357, 369, 394,
433-434
element free Galerkin (EFG), 27, 34-35, 63, 76, 207,
216-217
elastic-plastic, 13, 24, 32, 168-171, 178, 182, 278
equilibrium, 1, 22-23, 197, 262
extended finite element (X-FEM), 27, 55, 60
finite element (FEM), 2, 13-14, 16, 18-24, 27-28,
31, 33, 35-36, 38, 45, 54-55, 60, 62-64, 67, 71,
76-77, 123, 126, 156, 163, 184, 193-194,
205 -206, 217, 236-237, 269, 316, 352, 393,
403-404
homogenization, 56, 58, 61, 329, 334
inverse, 304
level set, 55-56, 60, 298, 316
limit, 1-2, 13-14, 16, 19, 22-23, 25, 27-29, 31,
33-34, 36-38, 88-92, 99, 101-103, 154, 168,
172, 176-177, 182, 318, 324, 327-328, 340-342,
352-356, 359, 362, 364, 366, 368-370, 404
meshfree, 207, 318-319, 310
meshless, 26, 33-34, 207, 216-217
Moving Least Square (MLS), 26, 207-208, 210
partition of unity, 76
shakedown, 2-3, 13-14, 17, 21-26, 28-32, 34, 37,
141-142, 151-152, 154, 168, 172-173, 175-179,
181-183, 353, 370
Michel Bruyneel, i, 78, 87
micromechanical modelling, 118
model
equilibrium, 34-35, 76, 194, 198, 205, 207-208, 210,
214-217, 237, 348, 351, 353, 394, 404
pure displacement, 237, 253
self-consistent, 329, 337
strength, 294
two-parameter, 280, 292
Mohammed Hjiaj, i, 88
moment-shear interaction, 294-295, 297-303
Mori-Tanaka approximation, 329, 335
N
Nguyen-Chien Thang, 32, 255
Nguyen-Dang Hung, i, iii, v-vi, 1-11, 16-19, 24, 53-54,
62, 77, 139, 154, 156, 167-168, 182-184, 193, 207,
217, 237-238, 253-254, 280, 303, 304, 318, 328,
340, 347, 353-354, 370, 393, 404
Nguyen-Hoai Son, ii, 229, 236
Nguyen-Huy Cung, 194
Nguyen-Huy Gia, 329
Nguyen-Ngoc Duong, 393
Nguyen-Thanh Son, 294
Nguyen-Tien Duong, ii, 3, 26-27, 33-35, 54, 156, 167,
254
Nguyen-Van Hieu, 3, 33, 280, 393
Nguyen-Van Yen, 294

Nguyen-Vinh Phu, 3, 27
Nguyen-Xuan Hung, 3, 27, 34, 62, 77, 194, 206, 369,
353, 393, 404
Nicolas Mos, i, vi, 3, 25, 55
non-associated plasticity, 88
non-potential operator, 88-89, 95-96, 102
numerical simulation, 59, 104-105, 109-110, 116, 194,
329, 335, 338
O
optimization
nonlinear, 354
topology, 78-83, 87
P
P. Cartraud, 55, 60
Patrick Morelle, ii, 3, 28-29, 142, 154, 354, 370
Phan-Hong Quang, ii, vi, 3, 318, 328
Pierre Duysinx, i, 87
plasticity, iii, 13, 20-24, 29, 31-32, 36, 54, 68, 76, 88,
91, 99, 103, 117, 122, 125-126, 129, 132, 138, 142,
144-146, 149, 151-153, 176, 178, 183, 274, 278-279,
318, 328, 338, 352, 370, 392
plate bending, 23, 26, 33-34, 54, 194, 199, 206, 342,
342-343, 348, 353, 393-394, 404
programming
linear, 17, 23-24, 154, 168, 171-172
protective clothing, 255-256, 267
puncture resistance, 255-256, 267
R
R. Boman, 104, 116
R. Chaouadi, 118, 140
Reinaldo Jacques Jospin, i, 3, 25, 29, 36-37, 354, 369
rubber, 255-256, 266-267
S
S. Ortola, 329, 339
semi-rigid connection, 34-35, 168, 172-173, 179-180,
182-183
stabilized conforming nodal integration, 62-63, 65-66,
76-77
Stphane Bordas, i, vi, 3, 27, 34, 62, 77, 207
strain
failure, 255, 259, 266
non-local, 62, 67
smooth, 62-63, 65, 71
stress
approach, 207, 237
intensity factor, 24-25, 29-30, 43-47, 50-54,
118-121, 140, 156, 162-167, 180, 193, 280, 289,
291, 293, 304-310, 312, 316, 369, 378, 381-382,
384-387, 389-391
singularity, 45, 156, 167, 268, 279, 371-372,
376-378, 381
structures
round bar, 118-120, 123, 139-140, 304, 311,
316-317
thin film, 268-269, 271, 273, 277-279, 371-373,
380-382, 384-385, 391-392
thin shell, 154, 393-394, 403
sphere, 141-142, 148-150, 152-153, 329
cylinder, 25, 30, 51, 71, 73-74, 141-144, 147,
149-152, 193, 249-253, 364-369
406
T
Takayuki Kitamura, 268, 371, 382
thermal loading, 30, 141-142, 146-148, 150-154
Tran-Duc Han, 3, 33, 280
t-stub effect, 294
V
variational principles, 2, 16, 23, 65, 67, 88-89, 102, 125,
171, 195
Vincent Rion, 78
Vu-Duc Khoi, ii, 3, 22, 26, 32, 34, 38, 154, 353-354,
370
Vu-Khanh Toan, 255
Yan Ai Min, ii, 3, 22, 25-26, 30-32, 34, 37-38, 139, 141,
193, 353, 370
407

BOOK 101002 Ver1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

BOOK 101002 Ver1

Uploaded by

Copyright:

Available Formats

See

Modeling in Mechanical and Civil Engineering:

Available from: H. Nguyen-Dang

International editorial advisory board

Doan-Duc Bao, Dr.

Michel Bruyneel, Dr.

Matthieu Domaszewski, Prof.

Marc Duflot, Dr.

Pierre Duysinx, Prof.

Mohammed Hjiaj, Prof.

Reinaldo Jacques Jospin, Dr.

Patrick Morelle, Dr.

Jean-Philippe Ponthot, Dr.

Marc Scibetta, Dr.

Yan Ai Min, Dr.

Vietnamese editorial advisory board

Vu-Duc Khoi, Dr.

Nguyen-Tien Duong, Dr.

Phan-Hong Quang, Ms.

Nguyen-Hoai Son, Dr.

Bui-Cong Thanh, Prof.

Chau-Dinh Thanh, Ms.

Le-Dinh Tuan, Dr.

Dedicated to Professor Nguyen-Dang Hung and his family

Nguyen-Dang Hung: the Scientist and the Man

Gry de Saxc and Duong-Thuy Phuong-Khanh

Prof. Nguyen-Dang Hungs Curriculum Vitae

International Editorial Advisory Members Papers

Gry de Saxc, D. Crpin and M. Pyrz

Image-based Micromechanics Analysis using Level Sets

Ionescu, Nicolas Mos, P. Cartraud and M. Bringhier

Smooth Strain Finite Elements: Selective Integration

Nguyen-Xuan Hung, Stphane Bordas and Nguyen-Dang Hung

Topology Optimization of Membranes Made of Orthotropic Material

Vincent Rion and Michel Bruyneel

Limit Analysis Theorems for Materials with a Non-Associated Flow Rule

Cold Roll Forming and Metal Cutting Simulation

R. Boman, L. Papeleux and Jean-Philippe Ponthot

Fracture Toughness Measurement using CircumferentiallyCracked Round Bar

Marc Scibetta, R. Chaouadi and E. Van. Walle

Analytic and Numeric Shakedown Solutions of Cylinder

EMMC & MCMC Students Papers

Nguyen-Tien Duong and Nguyen-Dang Hung

Nguyen-Dang Hung and Hoang-Van Long

Modified 8-Node Quarter-Point Element in Calculating Stress Intensity Factors

Le-Hoai Long and Nguyen-Dang Hung

Dual Analysis for Finite Element Solutions of Kirchhoff Plates

J.F. Debongnie, Nguyen-Xuan Hung and Nguyen-Huy Cung

Equilibrium Model in the Element Free Galerkin Method

Bui-Quoc Tinh, Marc Duflot, Stphane Bordas and Nguyen-Dang Hung

Active Balancing of Monorotor during Operation

A Refinement Criterion with p-Version in the Error Estimate

A Hybrid Stress Approach for High-Accuracy of Stresses

Dang-Dinh Thi and Nguyen-Dang Hung

Puncture Mechanics of Rubber Membranes

Nguyen-Chien Thang, Vu-Khanh Toan and Jaime Lara

Prediction of Delamination Strength at Interface

Do-Van Truong, Hiroyuki Hirakata, Takayuki Kitamura

Creating Macro Commands by using Bacon Commands to Simplify Meshing

Nguyen-Dang Hung, Tran-Duc Han, Chau-Dinh Thanh and Nguyen-Van Hieu

Influence of Moment-Shear Interaction to T-stub Effect in Bolted Column-Beam Joints

Nguyen-Thanh Son, Nguyen-Van Yen and Bui-Cong Thanh

Determination of the Fatigue Characteristics of Cracks by an Inverse Analysis