You are on page 1of 16

CHAPTER SIX

NONLINEAR FINITE ELEMENT SOLUTION AND


GEOMETRIC NONLINEARITY
6.1 General Introduction
In the nonlinear finite element analysis, nonlinearities occur in two different forms.
The first is material or physical nonlinearity, which result from nonlinear material
behavior. The second is geometric nonlinearity, which is a result from changes in
the geometry of the deformed body or structure.
The finite element method has been applied with great success to material
and geometric nonlinearity. The formulation of geometric nonlinearity can be
modeled with the known expression, while the material nonlinearity is still an
active area of interest in the civil engineering fields.
The earliest published paper on the material nonlinearites using the finite
element method was by Oden [22] and for the reinforced concrete using the finite
element method was by Ngo and Scordelis[21]. Soon after, geometric nonlinearty
was also included [5,18]. A great number of publications have been covering a
wide range of application to nonlinear shell problems.[1,6,13,29,33]
Iterative solution is usually used in the nonlinear finite element analysis to
guarantee the convergence towards the correct solution path. In general, time
stepping procedure method with iterative solution is used for the nonlinear
transient dynamic analysis. A reasonable number of time steps must be used and
within each time step a number of iterations are preformed to satisfy the
equilibrium equations according to a specific tolerance. A suitable selection of the
time step is required to achieve an accurate and economic solution. The selection
of the time step depends on the natural frequency of the finite element mesh for
both explicit and implicit algorithms.

In this chapter the following categories are presented:


1. geometric nonlinearity in which the total Lagrangian approach simplified
by Von Karman assumption.
2. Newmark implicit method for nonlinear transient dynamic analysis with
corrector-predictor algorithm.
3. The selection of time step which depends on the natural frequency of the
finite element mesh.
4.Graphic procedure for drawing the degenerated shell element and crack
patterns through layers.
6.2 Geometric Nonlinearity
6.2.1 General Remarks
The essential feature of geometric nonlinearity is that equilibrium equations must
be written with respect to the deformed geometry which is not known in advance.
These equilibrium equations contain stresses and strains which are continuously
changing as they depend on the initial and current configuration. Based on these
stresses and strains, two basically different formulations for geometric nonlinearity
can be employed for the description of large deformation problems.
(a) A total Lagrangian approach in which the current stress (2nd PiloaKirchhoff) and (Green-Lagrange) strain fields are referred to the original geometric
configuration and the displacement field gives the current configuration system in
relation to its initial position. The original reference frame remains stationary, and
everything is referred to it, regardless of how big the strain and rotation become.
Displacements, differentiation and integration are all carried with respect to the
original frame.
(b) Updated Lagrangian approach in which the current configuration of the
system is used to define the current (Cauchy) stress and (Almansi) strain. The

geometry of the structure is successively updated during the incremental process


and the stress and strain fields are referred to the last evaluated configuration. A
local coordinate system, called corotational system, is attached to each element.
The local system is moved with the element and therefore shares its rigid body
motion. Differentiation and integration are done with respect to local coordinates.
Both the Green-Lagrange and Almansi strains can be reduced to the
engineering strains by dropping the nonlinear terms.
Transformation from the total Lagrangian formulation to an updated
Lagrangian formulation, and vice-versa, can be performed by considering the
displacement and stretching of two natural systems of axes [2,7].
The most appropriate formulation for numerical solutions depends on the
type of analysis being considered. The use of the original configuration to refer to
the problem variables (total Lagrangain approach) is more advantageous for
degenerated shell elements. [31]
In the present study, total Lagrangian formulation with the simplified Von
Karman assumption is adopted in which large deflections and moderate rotations
are used.
6.2.2 Nonlinear Formulation for Degenerated Shell and Curved Beam
Elements
(i) Green-Lagrange Strain
The components of the Green-Lagrange strain vector expressed in terms of local
derivatives of the displacements for the degenerated shell element are[7]

x

y
xy


xz
zy

2
2
2
u
v
w




x
x
x
2
2
2
v
w
v 1 u



y 2 y

u v u u v v w w

y x x y x y x y
u w u u v v w w

z x x z x z x z
v w u u v v w w

z y y z y z y z

u 1

x 2

(6.1)

and those of the general curved beam element[31]

u 1 u 2 v 2 w 2

x
2

x
x x

(6.2)

(ii)Von Karman Assumptions


Von Karman assumption for large deflection of plates should take the following
form when applied to shells [7] and curved beams [31]:
a)

the thickness h is much smaller than the length L.

b)

the magnitude of deflection w is of the same order of the thickness h.

c)

the slope is small everywhere,

w / x 1, w / y 1.

d)

the tangential displacements, u and v , are small. Only nonlinear terms which
depend on w / x , w / y are to be retained in the strain displacement relations.

e)

All strain components are small.


Applying these assumptions, the green strain vector (equation) for shell
elements can be written as

u
u 1 w


1 w 2

x 2 x

x

2
v 2 x
v 1 w


y 1 w 2
y 2 y



u v w w u v 2 y

y x x y y x w w
u w x y
u w

0
z x
z x
v w
v w
0

z y
z y

x

y
xy

xy
zy

(6.3)

and that of the curved beam elements, equation(6.2) as;

u 1 w

x
2

x
u

u
x
v 1 w


x 2 x
u
0

x
0

0
x
0

x 0

(6.4)

The strain components can then be separated into linear 0 and nonlinear L
part
0 L

(6.5)

and

d d 0 d L

(6.6)

(iii)Strain Displacement Matrix


The strain-displacement matrix can be separated into two parts,
B B0 B L

Where
BL

(6.7)

is the linear part

B0

is the nonlinear part


For the degenerated shell elements, the nonlinear components of the strain

vector can be expressed by[33],


1 w 2
w


x
2 x 2

1 w
1 0

L 2 y w
w w 2

x y
y

0
0

0
w

y
w
x
0
0

w
x
w

(6.8)

or
L

1
AR
2

(6.9)

where
w

R x G.a
w

(6.10)

By taking the variation of equation (6.9)[33]


d L

1
1
dAR AdR AdR AGda
2
2

(6.11)

Then from the definition of strain matrix


BL AG

(6.12)

For the stiffened shell, G matrix must be transformed at some nodes to give
strains corresponding to six degrees of freedom
G T .G

(6.13)

where T is transformation matrix defined by equation(4.14)


B L A.G

(6.14)

(iv) Tangential Stiffness Matrix


Taking the variation of the equivalent nodal forces with respect to displacement
da
dP ( a )

ddV dB T dV kda

(6.15)

where
d Dd DBda

(6.16)

dB T dB LT

(6.17)

dV

(6.18)

dB T dV K da

(6.19)

P (a )

then the tangential stiffness matrix can be written as:


(6.20)

K K K

where

is defined as usual ( B T DBdV )

The geometric stiffness matrix

(6.21)

must be defined explicitly in order to

determine the tangential stiffness K. This can be done by using equations (6.12)
and (6.17)
K da dB dV G T dAT dV
T

(6.22)

where dAT can be evaluated as,

dA d

w
x

0
w
y

w

0 0 y
y
xy
w
0 0 xz


x
yz

(6.23)

x d
d
x

dAT
w

d y d

w
xy
y

w
xy
x

(6.24)

This vector can be arranged as,


x
dA
xy
T

w
d

xy x

y w
d y

(6.25) and using

the definition of G from


x xy
dAT
Gda
xy y

(6.26)

By substituting equation(6.26) in (6.22), the geometric stiffness can be


written as
K

G GdV
T

(6.27)

Following the same procedure the curved beam element, the geometric
stiffness matrix will be,
K b GbTT x Gb dV

(6.28)

where the subscript b denotes the beam elements.


6.3 Nonlinear Finite Element Solution
6.3.1 General Remark
The solutions of the equilibrium equation (4.77) governing the dynamic response
for the finite element codes are divided into two methods: direct integration
method and mode superposition. Although the two techniques may at first sight
appear to be quite different, in fact, they are closely related, and the choice for one
method or the other is determined only by their numerical effectiveness.

In direct integration, the equation (4.77) is integrated using a numerical step


by step procedure, the term direct means that prior to the numerical integration,
no transformation of the equation into a different form is carried out. In essence,
direct numerical integration is based on two ideas [2]
1.

Instead of trying to satisfy (4.77) at any time, it is aimed to satisfy (4.77) only at
discrete time intervals t apart. This means that, basically, (static) equilibrium,
which includes the effect of inertia and damping forces, is sought at discrete time
points within the interval of solution.

2.

The variation of displacements, velocities and accelerations within each time


interval is assumed. It is the form of this assumption that determines the accuracy,
stability and efficiency of each scheme.
In mode superposition a transformation is used to obtain new system
stiffness, mass, and damping matrix which have a smaller bandwidth than the
original system matrices. However, in practice, an effective transformation matrix
is established by using the displacement solution of the free vibration equilibrium
equation with damping neglected [2].
Although a form of mode superposition has been adopted in nonlinear
transient dynamic stress analysis [20], it is in general to use a time stepping
procedure. Such direct integration schemes may be broadly classified as either
explicit or implicit methods.
In the explicit integration algorithm the solution at time t t is based on the
equilibrium conditions at time t . During each time step, relatively little
computational effort is required since no formal matrix factorization is necessary.
Unfortunately, the method is conditionally stable and very small time steps are
often needed. The most popular of these methods is the central difference method
[2,23].

In implicit integration algorithm the equilibrium at time t t is imposed to


obtain the corresponding solution. In this scheme matrix factorization is required
but an unconditionally stable implicit algorithm can be selected in which the time
step length is governed by consideration of accuracy. Some of the widely used
implicit scheme are the Newmark method [19], the Houbolt method [2] and the
Wilson- method[32].
6.3.2 Newmark Method
The Newmark method, adopted in this work, is an extention of the linear
acceleration method. The dynamic equilibrium equation is linearized and written at
time

t n 1

as

Mdn 1 Cd n 1 Kd n 1 f n 1

(6.29)

and the following assumptions on the variation of displacements and velocities are
made within a typical time step:

t 2
1 2 dn 2 dn 1
d n 1 d n td n
2

d n 1 d n t 1 dn dn 1

where

dn

d n

, and

known at time t.

d
n

(6.30)

(6.31)

are the values of displacement, velocity and acceleration

Parameters

and control the stability and accuracy of the

method.
In the present work an unconditinally stable time stepping scheme is
adopted with

0.5 and 0.25

(i)Predictor-Corrector form of Newmarks Method


Hughes et al [10,11,12]have developed a predictor-corrector form of the Newmark
method which is most suitable for nonlinear transient analysis.
The Newmark formulas can be written in terms of predictor and corrector
values as
d n 1 d np1 t 2 dn 1

(6.32)

d n 1 d np1 tdn 1

(6.33)

with predictor values given as


t 2
1 2 dn
d np1 d n td n
2

(6.34)

d np1 d n t (1 ) dn

(6.35)

The terms

d n 1 and d n 1

are the corrector values while

d np1 and d np1

are the

predictor values. The corrector values for the acceleration can be obtained from
equations (6.32) and (6.33) as:

dn 1 d n 1 d np1

t
2

(6.36)

Substituting equations(6.32 and 6.36) into 4.77 an effective static problem is


formed in terms of unknown d
(6.37)

K d

where the effective stiffness matrix is

K M t 2 CT t K T

(6.38)

and the residual forces are


f n 1 Mdnp1 CT d np1 p ( d np1 )

(6.39)

When solving nonlinear problems, the linearization makes it necessary to


perform iterative correction to d to achieve equilibrium at time t t . A NewtonRaphson type scheme is used in this work [8,9,23].
(ii) Computational Algorithm for the Implicit Form of Newmark Method
The computations needed in a typical time step of the implicit form of the
Newmark method can be summarized as follows:
1.
set iteration counter i=0
2.

enter the predictor phase


d ni 1 d np1

(6.40)

d ni 1 d np1

(6.41)

dni 1 0

3.

(6.42)

solve the incremental effective static problem


K d i i

where
K M t CT t K T

and

i f n 1 Mdni 1 CT d ni 1 p d ni 1

(6.43)
(6.44)

are respectively, the effective stiffness matrix, and the residual force vector, with
K T p d

(6.45)

being the tangential stiffness matrix


4.

perform the corrector phase


d ni 11 d ni 1 d i

(6.47)

dni 11 d ni 11 d np1 t 2

(6.48)

d ni 11 d np1 tdni 11

(6.49)

5.If the convergence criterion selected is not satisfied, then set i=i+1 and go to (3)
6.set the solution at time t
d n 1 d ni 11

(6.50)

d n 1 d ni 11

(6.51)

dn 1 dni 11

(6.52)

for the use in the next time step. Also set n=n+1 and go to (1)
(6.4) Convergence Criteria
Any problem associated with iterative techniques must be concerned with the
selection of a suitable criterion for determining whether the current solutions are
sufficiently accurate to terminate the iterative process. At the end of each iteration
the solution is checked whether to continue the iteration process or to terminate it
and to consider a new load increment for static analysis or a new time step for
dynamic analysis.

In the present study the corrector-predictor Newmark iterative technique is


used for solving nonlinear dynamic problem.
The convergence criterion used for the nonlinear analysis is usually based
on: (1) displacement (2) forces
In the displacement convergence criterion the solution is considered to have
converged when the norm of the incremental displacement to the norm of the total
displacement is within a selected tolerance as.
a in
a in

where

Tn is

Tn

(6.53)

a given tolerance

The force convergence criterion is based on measuring the out of balance


forces

ri n

at the end of each iteration. These forces are checked against the current

applied forces
ri n
fin

by,

Tn

(6.54)

In the present work, the displacement criterion is adopted.


(6.5) Critical Time Step
In explicit and implicit time integration schemes the selection of an appropriate
time step is crucially important.
The time step is restricted when explicit algorithm is used to[23]:
t crit 2 max

where max is the maximum circular frequency of the finite element mesh. To
avoid the solution of the eigenvalue problem for the whole structure the proposed
method by Irons can be used[14]. Irons proved that the highest system eigenvalue
must always be less than the highest eigenvalue of the individual elements.
To avoid the exact evaluation of the highest finite element mesh frequency,
approximate expressions are usually employed. [23,28,30]

(1 )(1 2 )

t L
E (1 )

where L is the smallest length between any two nodes


In general if implicit elements are used, the numerical integration using any
of the methods[19,32] are accurate when the time step is smaller than 1/100 of the
elastic fundamental period for linear problem[2]. For nonlinear problem using
Newmark integration scheme with equilibrium iteration, a time step less than 1/100
of the elastic fundamental period can be used[3]. Bathe[3] showed that the result
obtained by using the Newmark integration scheme for time steps (1/24 and 1/126
of the fundamental period) are significantly different unless equilibrium iteration
is used. The choice of the time step for nonlinear transient analysis of reinforced
concrete ranges from 1/10 of the elastic fundamental period [27] to 1/100 [8,17].
In the present study, Newmark method with implicit algorithm is used. The
selection of the time step is based on the elastic fundamental period which is based
on the smallest natural frequency.
Subspace iteration method is used to obtain the smallest natural frequency
for the whole structure [2]. The following example is solved using subspace
iteration method and compared with other sources.
A square plate with h= 1mm, l=100 mm, E=205,000 Mpa., =7900 kg/m3
and Poissons ratio =0.3 was discretized by 16 elements and the boundaries were
assumed to be simply supported or fully clamped all round. The fundamental
frequencies obtained from the present study are compared to those obtained from
the finite strip method[15] and finite element method[16] in Table (6.3).
Table (6.3) Free vibration analysis of plates
Boundary condition
Simply supported
Fully clamped

F.S.[15]
484.28
885.01

F.E.[16]
885.48

Present study
484.6
884.459

(6.6) Mesh and Crack Drawing for the Degenerated Shell Element
Drawing is considered an important part in finite element analysis. Using the
drawing facilities makes it possible to check the input data easily and gives clear
insight for the result to be interpreted when the deflection response, cracks,
yielding points etc. are drawn.
In the present work a Visual Basic program is developed. Shape, deformed
shape and cracks for the degenerated shell element can be drawn and rotated,
Figures(6.1 and 6.2).
(i) Shape and Deformed Shape Drawing
For the drawings of the shape and deformed shape, curves are used by using spline
order. The deformed shape is drawn after adding the displacement at the top and
bottom of the shell element to the original coordinate of the elements using the
following equations:
1. for nodes with five degrees of freedom, equation(4.12) can be written:
u uk
v v hk v
2k
k
2
w wk


v1k k
k

(6.55)

2. for six degrees of freedom nodes, equations (4.12-4.14) can be written:


0
u uk
v v hk v z
3k
k
2 y
v3 k
w wk

v3zk
0
v3xk

v3yk k

v3xk k
0 k

(6.56)

The displacements on the top and bottom face are computed when

and

1 respectively.

In Figure (6.1), a curved box girder bridge is drawn using the developed
program for graphic representation. In the figure, the original shape is drawn and in
Figure (6.2), the deformed shape due to two concentrated loads is shown.

You might also like