Cours

The Finite Element Method for the Analysis of
Non-Linear and Dynamic Systems

Prof. Dr. Eleni Chatzi
Lecture 1 - 17 September, 2013
Institute of Structural Engineering
Method of Finite Elements II
Course Information
Instructor
Prof. Dr. Eleni Chatzi, email: chatzi@ibk.baug.ethz.ch
Office Hours: HIL E14.3, Wednesday 10:00-12:00 or by email
Assistant
Juan Escalln Osorio, HIL E10.2, email: ejuan@student.ethz.ch
Course Website
Lecture Notes and Homeworks will be posted at:
http://www.ibk.ethz.ch/ch/education
Suggested Reading
Nonlinear Finite Elements for Continua and Structures by T.
Belytschko, W. K. Liu, and B. Moran, John Wiley and Sons, 2000
The Finite Element Method: Linear Static and Dynamic Finite
Element Analysis by T. J. R. Hughes, Dover Publications, 2000
The Finite Element Method Vol. 2 Solid Mechanics by O.C.
Zienkiewicz and R.L. Taylor, Oxford : Butterworth Heinemann, 2000
Course Outline
Review of the Finite Element method - Introduction to
Non-Linear Analysis
Non-Linear Finite Elements in solids and Structural Mechanics
-
Overview of Solution Methods

Continuum Mechanics & Finite Deformations
Lagrangian Formulation
Structural Elements
Dynamic Finite Element Calculations

- Integration Methods
- Mode Superposition
Eigenvalue Problems
Special Topics
- Extended Finite Elements, Multigrid Methods, Meshless Methods
Grading Policy
Performance Evaluation - Homeworks (100%)

Homework
Homeworks are due in class 2-3 weeks after assignment
Computer Assignments may be done using any coding language
(MATLAB, Fortran, C, MAPLE) - example code will be
provided in MATLAB
Commercial software such as CUBUS, ABAQUS and SAP will
also be used for certain Assignments
Homework Sessions will be pre-announced and it is advised to bring
a laptop along for those sessions
Review of the Finite Element Method (FEM)

Classification of Engineering Systems
Discrete
Continuous
q|y+dy
q|x
dy
dx
q|x+dx
q|y
h1
h2
Flow
of water
Permeable Soil
Impermeable Rock
F = KX
Direct Stiffness Method
2
2x
2
2y
=0
Laplace Equation
FEM: Numerical Technique for approximating the solution of continuous

systems. We will use a displacement based formulation and a stiffness
based solution (direct stiffness method).

How is the Physical Problem formulated?
The formulation of the equations governing the response of a system under
specific loads and constraints at its boundaries is usually provided in the
form of a differential equation. The differential equation also known as the
strong form of the problem is typically extracted using the following sets
of equations:
Equilibrium Equations
aL + ax
ex. f (x) = R +
(L x)
2
Constitutive Requirements
Equations
ex. = E
Kinematics Relationships
du
ex. =
dx
Axial bar Example

q(x)=ax
x
aL
ax
f(x)
L-x

How is the Physical Problem formulated?
Differential Formulation (Strong Form) in 2 Dimensions
Quite commonly, in engineering systems, the governing equations are of a
2
second order (derivatives up to u 00 or 2 ux ) and they are formulated in terms
of variable u, i.e. displacement:
Governing Differential Equation ex: general 2nd order PDE
2
u
u u
A(x, y ) 2 ux + 2B(x, y ) xy
+ C (x, y ) 2 yu = (x, y , u, y
, y )
Problem Classification
Boundary Condition Classification
B 2 AC < 0 elliptic
B 2 AC = 0 parabolic
B 2 AC > 0 hyperbolic
Essential (Dirichlet): u(x0 , y0 ) = u0

order m 1 at most for C m1
u
Natural (Neumann): y
(x0 , y0 ) = u 0
order m to 2m 1 for C m1

Differential Formulation (Strong Form) in 2 Dimensions
The previous classification corresponds to certain characteristics for each
class of methods. More specifically,
Elliptic equations are most commonly associated with a diffusive or
dispersive process in which the state variable u is in an equilibrium
condition.
Parabolic equations most often arise in transient flow problems where
the flow is down gradient of some state variable u. Often met in the
heat flow context.
Hyperbolic equations refer to a wide range of areas including
elasticity, acoustics, atmospheric science and hydraulics.
Strong Form - 1D FEM

Reference Problem
Consider the following 1 Dimensional (1D) strong form (parabolic)
d
du
(c(x) ) + f(x) = 0
dx
dx
d
c(0) u(0) = C1
dx
u(L) = 0
Physical Problem (1D)
(Neumann BC)
(Dirichlet BC)
Diff. Equation
+ = 0
One dimensional Heat

flow
+ = 0
Axially Loaded Bar
Quantities
T=temperature
A=area
k=thermal
conductivity
Q=heat supply
u=displacement
A=area
E=Youngs
modulus
B=axial loading
Constitutive
Law
Fourier
= /
= heat flux
Hooke
= /
= stress
9
Weak Form - 1D FEM

Approximating the Strong Form
The strong form requires strong continuity on the dependent field
variables (usually displacements). Whatever functions define these
variables have to be differentiable up to the order of the PDE that
exist in the strong form of the system equations. Obtaining the
exact solution for a strong form of the system equation is a quite
difficult task for practical engineering problems.
The finite difference method can be used to solve the system
equations of the strong form and obtain an approximate solution.
However, this method usually works well for problems with simple
and regular geometry and boundary conditions.
Alternatively we can use the finite element method on a weak
form of the system. This weak form is usually obtained through
energy principles which is why it is also known as variational form.
10
Weak Form - 1D FEM
From Strong Form to Weak form

Three are the approaches commonly used to go from strong to weak
form:
Principle of Virtual Work
Principle of Minimum Potential Energy
Methods of weighted residuals (Galerkin, Collocation, Least
Squares methods, etc)
*We will mainly focus on the third approach.
11
Weak Form - 1D FEM

From Strong Form to Weak form - Approach #1
Principle of Virtual Work
For any set of compatible small virtual displacements imposed on the body
in its state of equilibrium, the total internal virtual work is equal to the
total external virtual work.
Z
Wint =
T d = Wext =
T bd +
u
ST TS d +
u
iT RC i
u
where
TS : surface traction (along boundary )
b: body force per unit area
RC : nodal loads
: virtual displacement
u
: virtual strain
: stresses
12
Weak Form - 1D FEM

Principle of Minimum Potential Energy
Applies to elastic problems where the elasticity matrix is positive definite,
hence the energy functional has a minimum (stable equilibrium).
Approach #1 applies in general.
The potential energy is defined as the strain energy U minus the work of
the external loads W
=UW
Z
1
T C d
U=
2
Z
Z
T bd +
W=
u
ST Ts dT +
u
i
T
u
i RC
(b Ts , RC as defined previously)
13
Weak Form - 1D FEM

Galerkins Method
Given an arbitrary weight function w, where
S = {u|u C 0 , u(l) = 0}, S 0 = {w |w C 0 , w (l) = 0}

C 0 is the collection of all continuous functions.
Multiplying by w and integrating over
Z
w (x)[(c(x)u 0 (x))0 + f (x)]dx = 0
[w (0)(c(0)u 0 (0) + C1 ] = 0
14
Weak Form - 1D FEM

Using the divergence theorem (integration by parts) we reduce the
order of the differential:
Z
wg dx =
[wg ]l0
gw 0 dx
The weak form is then reduced to the following problem. Also, in

what follows we assume constant properties c(x) = c = const.
Find u(x) S such that:
Z
w cu dx =
0
wfdx + w (0)C1
0
S = {u|u C 0 , u(l) = 0}
S 0 = {w |w C 0 , w (l) = 0}
15
Weak Form
Notes:
1
Natural (Neumann) boundary conditions, are imposed on the

secondary variables like forces and tractions.
u
For example, y
(x0 , y0 ) = u 0 .
Essential (Dirichlet) or geometric boundary conditions, are imposed

on the primary variables like displacements.
For example, u(x0 , y0 ) = u0 .
A solution to the strong form will also satisfy the weak form, but not
vice versa.Since the weak form uses a lower order of derivatives it can
be satisfied by a larger set of functions.
For the derivation of the weak form we can choose any weighting
function w , since it is arbitrary, so we usually choose one that satisfies
homogeneous boundary conditions wherever the actual solution
satisfies essential boundary conditions. Note that this does not hold
for natural boundary conditions!
16
FE formulation: Discretization
How to derive a solution to the weak form?
Step #1:Follow the FE approach:
Divide the body into finite elements, e, connected to each other
through nodes.
Then break the overall integral into a summation over the finite
elements:
"
#
Z xe
X Z x2e
2
w 0 cu 0 dx
wfdx w (0)C1 = 0
e
x1e
x1e
17
1D FE formulation: Galerkins Method

Step #2: Approximate the continuous displacement using a discrete
equivalent:
Galerkins method assumes that the approximate (or trial) solution, u, can
be expressed as a linear combination of the nodal point displacements ui ,
where i refers to the corresponding node number.
X
u(x) u h (x) =
Ni (x)ui = N(x)u
i
where bold notation signifies a vector and Ni (x) are the shape functions.
In fact, the shape function can be any mathematical formula that helps us
interpolate what happens at points that lie within the nodes of the mesh.
In the 1-D case that we are using as a reference, Ni (x) are defined as 1st
degree polynomials indicating a linear interpolation.
As will be shown in the application presented in the end of this lecture, for the
case of a truss element the linear polynomials also satisfy the homogeneous
equation related to the bar problem.
18

Shape function Properties:
Bounded and Continuous

One for each node
Nie (xje ) = ij , where

1 if i = j
ij =
0 if i 6= j
The shape functions can be written as piecewise functions of the x
coordinate:
This is not a convenient notation.
x x
Instead
of using the global coordinate
i1
,
xi1 x < xi
x, things become simplified when

xi xi1
xi + 1 x
Ni (x) =
using coordinate referring to the
, xi x < xi+1
x
local
system of the element (see page
i+1
i
0,
otherwise
25).
19

Step #2: Approximate w (x) using a discrete equivalent:
The weighting function, w is usually (although not necessarily) chosen to
be of the same form as u
X
w (x) w h (x) =
Ni (x)wi = N(x)w
i
i.e. for 2 nodes:

N = [N1 N2 ]
u = [u1
u2 ]T
w = [w1
w2 ]T
Alternatively we could have a Petrov-Galerkin formulation, where w (x) is

obtained through the following relationships:
w (x) =
X
he dNi
(Ni +
)wi
dx
i
= coth(
Pe e
2
) e
2
Pe
coth =
e x + e x
e x e x
20

Note: Matrix vs. Einsteins notation:
In the derivations that follow it is convenient to introduce the
equivalence between the Matrix and Einsteins notation. So far we
have approximated:
X
u(x) =
Ni (x)ui (Einstein0 s notation) = N(x)u (Matrix notation)
i
w (x) =
Ni (x)wi = N(x)w
(similarly )
As an example, if we consider an element of 3 nodes:

u(x) =
3
X
Ni (x)ui = N1 u1 + N2 u2 + N3 u3
u1
u(x) = [N1 N2 N3 ] u2 = N(x)u
u3
21

Step #3: Substituting into the weak formulation and rearranging
terms we obtain the following in matrix notation:
Z l
Z l
0
0
w cu dx
wfdx w (0)C1 = 0
0
0
l
(wT NT )0 c(Nu)0 dx
wT NT fdx wT N(0)T C1 = 0
Since w, u are vectors, each one containing a set of discrete values

corresponding at the nodes i, it follows that the above set of equations can
be rewritten in the following form, i.e. as a summation over the wi , ui
components (Einstein notation):
dN
(x)
j
dx
c
wj
dx
0
i
j

Z l X
X

f
wj Nj (x)dx
wj Nj (x)C1
=0
0

j
j
dNi (x)
ui
dx
x=0
22

This is rewritten as,
"Z
wj
0
X
i
dNi (x) dNj (x)

cui
dx
dx
#
fNj (x)dx + (Nj (x)C1 )|x=0 = 0
The above equation has to hold wj since the weighting function w (x) is
an arbitrary one. Therefore the following system of equations has to hold:
!
Z l X
dNi (x) dNj (x)
cui
fNj (x)dx + (Nj (x)C1 )|x=0 = 0 j = 1, ..., n
dx
dx
0
i
After reorganizing and moving the summation outside the integral, this
becomes:
#
"
Z l
X Z l dNi (x) dNj (x)
c
ui =
fNj (x)dx + (Nj (x)C1 )|x=0 = 0 j = 1, ..., n
dx
dx
0
0
i
23

We finally obtain the following discrete system in matrix notation:
Ku = f
where writing the intagral from 0 to l as a summation over the
subelements we obtain:
e
K = Ae K K =
x2e
NT
,x cN,x dx
x1e
f = Ae f e f e =
x2e
x1e
x2e
BT cBdx
x1e
NT fdx + NT h|x=0
where A is not a sum but an assembly (see page and, x denotes

differentiation with respect to x.
dN(x)
is known as the strain displacement
In addition, B = N,x =
dx
matrix.
24
1D FE formulation: Iso-Parametric Formulation

Iso-Parametric Mapping
This is a way to move from the use of global coordinates (i.e.in
(x, y )) into normalized coordinates (usually (, )) so that the finally
derived stiffness expressions are uniform for elements of the same
type.
Shape Functions in Natural Coordinates

x() =
Ni ()xie = N1 ()x1e + N2 ()x2e
i=1,2
1
N1 () = (1 ),
2
1
N2 () = (1 + )
2
25

Map the integrals to the natural domain element stiffness matrix.
Using the chain rule of differentiation for N((x)) we obtain:
Z 1
Z x2e
NT,x cN,x dx =
(N, ,x )T c(N, ,x )x, d
Ke =
x1e
where N, =
and x, =
d
d
1
(1
2
1
(1
2
+ )
1
2
1
2
x e x1e
h
dx
= 2
= = J (Jacobian) and h is the element length
d
2
2
,x =
d
= J 1 = 2/h
dx
From all the above,

Ke =
c
x2e x1e
1
1
1
1
Similary, we obtain the element load vector:

Z x2e
Z 1
NT ()fx, d + NT (x)h|x=0
fe =
NT fdx + NT h|x=0 =
x1e
Note: the iso-parametric mapping is only done for the integral.

26

So what is meant by assembly? (Ae )
It implies adding the components of the stiffness matrix that correspond to
the same degrees of freedom (dof).
In the case of a simple bar, it is trivial as the degrees of freedom (axial
displacement) are as many as the nodes:
2:K2
1:K1
1
Red indicates the node each

component corresponds to
3
1
1 1
Element Stiffness Matrices (2x2): 1 =

1 1
1
1
Total Stiffness Matrix (4x4): = 1 1 + 1

0
1
1
2,
0
1
1
1 1
2 =
1 1
2
3
1
2
3
27

In the case of a frame with beam elements, the stiffness matrix of the elements is
typically of 4x4 size, corresponding to 2 dofs on each end (a displacement and a
rotation):
2
u1
*The process will be shown explicitly

during the HW sessions
u2y
2
1:K1
2
2:K
u2x
Green indicates the dof each

component corresponds to
u3
u1
1
11
1
12
1
Element Stiffness Matrices (4x4): = 1
13
1
14
u2y
1
12
1
22
1
23
1
24
1
13
1
23
1
33
1
34
1
2
44
+ 22
1
Total Stiffness Matrix (2x2): = 34
0
u2x
1
14
2
u1
11
1
2
24
1
12
2
1 u2y, = 2
34
13
1 2
2
44
14
u2y
1
34
1
33
2
24
u2x
u3
2
12
2
22
2
23
2
24
2
13
2
23
2
33
2
34
0 2
2 u2y
24
2
u2x
11
3
2
14
2
24
2
34
2
44
u2x
2
u3
3
Fixed dofs are

not included in
the total
stiffness matrix
28
Axially Loaded Bar Example

A. Constant End Load
Given: Length L, Section Area A, Youngs modulus E

Find: stresses and deformations.
Assumptions:
The cross-section of the bar does not change after loading.
The material is linear elastic, isotropic, and homogeneous.
The load is centric.
End-effects are not of interest to us.
29

A. Constant End Load
Strength of Materials Approach
From the equilibrium equation, the axial force at a random point x
along the bar is:
R
A
From the constitutive equation (Hookes Law):
f(x) = R(= const) (x) =
(x)
R
=
E
AE
Hence, the deformation is obtained as:
(x) =
(x)
Rx
(x) =
x
AE
Note: The stress & strain is independent of x for this case of
loading.
(x) =
30

B. Linearly Distributed Axial + Constant End Load
From the equilibrium equation, the axial force at random point x

along the bar is:
f(x) = R +
aL + ax
a(L2 x 2 )
(L x) = R +
( depends on x)
2
2
In order to now find stresses & deformations (which depend on x)

we have to repeat the process for every point in the bar. This is
computationally inefficient.
31

From the equilibrium equation, for an infinitesimal element:
d
A = q(x)x + A( + ) A |{z}
lim
+ q(x) = 0 A
+ q(x) = 0
x
dx
x0
Also, =
du
d 2u
, = E , q(x) = ax AE 2 + ax = 0
dx
dx
Strong Form
d 2u
+ ax = 0
dx 2
u(0) = 0 essential BC

du
f(L) = R AE
dx
AE
= R natural BC
x=L
Analytical Solution
u(x) = uhom + up u(x) = C1 x + C2
ax 3
6AE
C1 , C2 are determined from the BC

32

An analytical solution cannot always be found
Approximate Solution - The Galerkin Approach (#3): Multiply by the weight function
w and integrate over the domain
L
AE
0
d 2u
wdx +
dx 2
axwdx = 0
0
Apply integration by parts

Z L
d 2u
du l
du dw
AE
wdx
=
w
AE
dx
2
dx
dx
dx dx
0
0
0

Z L
Z L
d 2u
du dw
du
du
AE 2 wdx = AE
(L)w(L) AE
(0)w(0)
AE
dx
dx
dx
dx
dx dx
0
0
AE
But from BC we have u(0) = 0, AE du

(L)w(L) = Rw(L), therefore the approximate
dx
weak form can be written as
L
AE
0
du dw
dx = Rw(L) +
dx dx
axwdx
0
33

Variational Approach (#1)
Let us signify displacement by u and a small (variation of the) displacement by u. Then
the various works on this structure are listed below:
L
Z
Wint = A
dx
0
Wext = Ru|x=L
L
qudx
Wbody =
0
In addition, = E du
dx
Then, from equilibrium: Wint = Wext + Wbody
L
E
0
du d(u)
dx =
dx dx
qudx + Ru|x=L
0
This is the same form as earlier via another path.

34

In Galerkins method we assume that the approximate solution, u can be expressed as
u(x) =
n
X
uj Nj (x)
j=1
w is chosen to be of the same form as the approximate solution (but with arbitrary
coefficients wi ),
w(x) =
n
X
wi Ni (x)
i=1
Plug u(x),w(x) into the approximate weak form:

L
AE
0
n
X
j=1
uj
Z L X
n
n
n
X
dNj (x) X dNi (x)
ax
wi Ni (x)dx
wi
dx = R
wi Ni (L) +
dx
dx
0
i=1
i=1
i=1
wi is arbitrary, so the above has to hold wi :

n Z
X
j=1

Z L
dNj (x)
dNi (x)
AE
dx uj = RNi (L) +
axNi (x)dx
dx
dx
0
i = 1...n
which is a system of n equations that can be solved for the unknown coefficients uj .
35

The matrix form of the previous system can be expressed as
Z
Kij uj = fi where Kij =
0
dNj (x)
dNi (x)
AE
dx
dx
dx
axNi (x)dx
and fi = RNi (L) +

0
Finite Element Solution - using 2 discrete elements, of length h (3 nodes)

From theiso-parametric
formulation we know the element stiffness matrix
1
1
e
AE
K = h
. Assembling the element stiffness matrices we get:
1
1
Ktot
e
K11
1
K12
=
0
Ktot
1
AE
1
=
h
0
1
K12
1
2
K22 + K11
2
K12
1
2
1
0
2
K12
2
K22
0
1
1
36
We also have that the element load vector is

Z L
axNi (x)dx
fi = RNi (L) +
0
Expressing the integral in iso-parametric coordinates Ni () we have:

d
2
= , x = N1 ()x1e + N2 ()x2e ,
dx
h
Z L
2
fi = R|i=4 +
a(N1 ()x1e + N2 ()x2e )Ni () d
h
0
37
Strong Form - 2D Linear Elasticity FEM

Governing Equations
Equilibrium Eq:
Kinematic Eq:
Constitutive Eq:
Traction B.C.:
Displacement B.C:
s + b = 0
= s u
=D
n = Ts
u = u
t
u
Hookes Law - Constitutive Equation

Plane Stress
Plane Strain
zz = xz = yz = 0, zz =
6 0
1
0
E
1
0
D=
1 2
0 0 1
2
zz = xz = yz = 0, zz 6= 0
1
E

D=
(1 )(1 + )
0
1
0
0
0
12
2
38
2D FE formulation: Discretization
Divide the body into finite elements connected to each other through
nodes
39

Shape Functions in Natural Coordinates
1
N1 (, ) = (1 )(1 ),
4
1
N3 (, ) = (1 + )(1 + ),
4
1
N2 (, ) = (1 + )(1 )
4
1
N4 (, ) = (1 )(1 + )
4
Iso-parametric Mapping
x=
y=
4
X
i=1
4
X
Ni (, )xie
Ni (, )yie
i=1
40
Bilinear Shape Functions
41
2D FE formulation: Matrices
from the Principle of Minimum Potential Energy (see slide #9)
=0Kd =f
d
where
Ke =
BT DBd,
fe =
N T Bd +
Z
eT
N T ts d
Gauss Quadrature
Z 1Z 1
I =
f (, )dd
=
1 1
Ngp
Ngp
XX
Wi Wj f (i , j )
i=1 i=1
where Wi , Wj are the weights and

(i , j ) are the integration points.
42

Lecture 10 - 4 December, 2012
Nonlinear Dynamic Analysis

Example Case
Specimen
1285
1535
Dynamic Shake Table Test of a Reinforced Concrete Column
Servo hydraulic test cylinder

Piston: + / - 125 mm
Piston force: + / - 100 kN
Damper
250
Anchorage
ETH ShakeTable
Example Case
Elevation Drawings
Example Case
Plan Drawings
Expected Behavior
RC Response in Cyclic Loading
Experimental Data
Accelerometer Measurements
Displacement Measurements
RC Response Characteristics
Strength Deterioration
Stiffness Degradation
Pinching Behavior
6. Forcedisplacement relationships observed in static cyclic tests (RFfailure of vertical reinforcement, DCdiagonal cracking,
mpression failure).
Modeling of Reinforced Concrete Behavior

Concrete Material Models
-
peak compressive stress
cu
E0
Compression
softening
strain at maximum stress
+
o
cu
tu = maximum tensile strength of concrete
Tension
+
Figure 2.5: Typical uniaxial compressive and tensile stress-strain curve for concrete (Bangash 1989)
Typical uniaxial compressive and tensile

stress-strain curve for concrete
compression, the stress-strain curve for concrete is linearly elastic up to about 30 percent of
e maximum compressive strength. Above this point, the stress increases gradually up to the
aximum compressive strength. After it reaches the maximum compressive strength cu , the
rve descends into a softening region, and eventually crushing failure occurs at an ultimate
ain cu . In tension, the stress-strain curve for concrete is approximately linearly elastic up to
e maximum tensile strength. After this point, the concrete cracks and the strength decreases
adually to zero (Bangash 1989).
2.3.1.1
FEM Input Data
Modeling of Reinforced
Concrete
Behavior
2. DISCRETE VS
SMEARED CRACK
MODELS
2.1. The discrete crack approach
Failure Criteria
forapproach
Concrete
The discrete crack
to concrete fracture is intuitively appealing: a crack is introduced as
a geometric entity. Initially, this was implemented by letting a crack grow when the nodal force
at the node ahead of the crack tip exceeded a tensile strength criterion. Then, the node is split
The determination
of the
failure
is very
important
fornode.
theWhen
proper
into two nodes and
tip of thecriteria
crack is assumed
to propagate
to the next
the
tensile strength criterion is violated at this node, it is split and the procedure is repeated, as
simulation
of
the
degrading
behavior
of
concrete
structures.
sketched in Figure 1 [1].
discrete crack approach in its original form has several disadvantages. Cracks are forced
A. DiscreteTheCracking
to propagate along element boundaries, so that a mesh bias is introduced. Automatic remeshing
The discrete
approach
to concrete
fracture
is intuitively
appealing:
allowscrack
the mesh
bias to be reduced,
if not eliminated,
and sophisticated
computer
codes with a crack
remeshing
developed by
IngraeaInitially,
and co-workers
Nevertheless,
a computational
is introduced
as awere
geometric
entity.
this [3].
was
implemented
by letting a
diculty, namely, the continuous change in topology, is inherent in the discrete crack approach
crack grow
the nodal
force
at the
node ahead
of the crack tip exceeded a
andwhen
is to a certain
extent even
aggravated
by remeshing
procedures.
The change
in topology
was to athe
largenode
extent alleviated
the advent
meshlessand
methods,
tensile strength
criterion.
Then,
is splitbyinto
two ofnodes
the tip of
such as the element-free Galerkin method [4]. Indeed, successful analyses have been carried out
the crack using
is assumed
to
propagate
to
the
next
node.
When
the
tensile
strength
these methods, but disadvantages including diculties with robust three-dimensional
largenode,
computational
demandand
compared
nite element
the
criterion isimplementations,
violated atthethis
it is split
the with
procedure
is methods,
repeated.
somewhat ad hoc manner in which the support of a node is changed in the presence of a crack [5]
Figure 1. Early discrete crack modelling.

Copyright # 2004 John Wiley & Sons, Ltd.
Int. J. Numer. Anal. Meth. Geomech. 2004; 28:583607

B. Smeared Cracking
In a smeared crack approach, the nucleation of one or more cracks in
the volume is translated into a deterioration of the current stiffness
and strength.
Generally, when the combination of stresses satisfies a specified
criterion, e.g. the major principal stress reaching the tensile strength
ft ; a crack is initiated.
This implies that at the integration point where the stress, strain and
history variables are monitored, the isotropic stress - strain relation is
replaced by an orthotropic elasticity-type relation with the n,s-axes
being axes of orthotropy; n is the direction normal to the crack and s
is the direction tangential to the crack.

surface for the concrete. Consequently, a criterion for failure of the concrete due to a
multiaxial stress state can be calculated (William and Warnke 1975).
A three-dimensional failure surface for concrete is shown in Figure 2.7. The most
significant nonzero principal stresses are in the x and y directions, represented by xp and
yp, respectively. Three failure surfaces are shown as projections on the xp-yp plane.
The mode of failure is a function of the sign of zp (principal stress in the z direction).
For example, if xp and yp are both negative (compressive) and zp is slightly positive
(tensile), cracking would be predicted in a direction perpendicular to zp. However, if zp
is zero or slightly negative, the material is assumed to crush (ANSYS 1998).
One such criterion is utilized by ANSYS accounting for both crushing & cracking.
fr
f c
fr
f c
For xp , yp 0 (compressive) and zp > 0

(tensile), cracking would be predicted in a direction
perpendicular to zp . However, if zp 0, the
material is assumed to crush.
Figure 2.7: 3-D failure surface for concrete (ANSYS 1998)
After cracking, the elastic modulus of the concrete element is set to zero in the
direction parallel to the principal tensile stress direction.
Crushing occurs when compressive stresses exceed the compressive failure
strength.
In a concrete element, cracking occurs when the principal tensile stress in any direction
lies outside the failure surface. After cracking, the elastic modulus of the concrete
element is set to zero in the direction parallel to the principal tensile stress direction.
Crushing occurs when all principal stresses are compressive and lie outside the failure
surface; subsequently, the elastic modulus is set to zero in all directions (ANSYS 1998),
and the element effectively disappears.
During this study, it was found that if the crushing capability of the concrete is turned on,
the finite element beam models fail prematurely. Crushing of the concrete started to
develop in elements located directly under the loads. Subsequently, adjacent concrete
In practice, a pure compression failure of concrete is unlikely. Therefore, crushing

13
is ignored and cracking controls failure.
iffness of the entire structure based on diaphragmatic

Steel Material Models
conventional stress-strain curves both for unconfined

parabolic stress-strain relationship with a softening
s-strain diagram with hardening is implemented. The
d in Figure 1.
Bilinear
Model
Hardening
nconfined
concrete
and b)with
for reinforcing
steel
80
fy
Ep
STRESS [ksi]
60
R=20
40
R=5
20
0.002
0.004
0.006
0.008
STRAIN [in/in]
Giuffr`e-Menegotto-Pinto Model
Figure 15. Material Parameters of Monotonic Envelope of Steel_2 Model

100
R controls the transition from elastic to

80
60
inelastic branch.
ral elements namely; beams, columns and shear walls
storey buildings. By combining such elements one
of Structural
inked together through Institute
diaphragms
at theEngineering
floor levels Method of Finite Elements II
Stress [ksi]
40
20
10

Steel Cyclic Model
Figure 2. Steel cyclic model.

11
finite element analysis textbooks for a more formal and complete introduction to basic concepts
if needed.
Modeling of Reinforced Concrete Behavior - 3D Solid

Approach
2.2 ELEMENT TYPES

Concrete Modeling using Solid Elements
2.2.1 Reinforced Concrete
A solid (3D) finite element can be used to model the concrete. For
An example,
eight-node ANSYS
solid element,
wasnode
used element
to model the
concrete.
The solid
element
has
usesSolid65,
an eight
(Solid
65) with
three
degrees
eight nodes with three degrees of freedom at each node translations in the nodal x, y, and z
of freedom
at eachis node
in the nodal
x, y,
and zorthogonal
directions.
directions.
The element
capabletranslations
of plastic deformation,
cracking
in three
The
element
is
capable
of
plastic
deformation,
cracking
in
three
directions, and crushing. The geometry and node locations for this element type orthogonal
are shown in
directions,
and crushing.
Figure
2.1.
Figure 2.1: Solid65 3-D reinforced concrete solid (ANSYS 1998)

12
Modeling of Reinforced Concrete Behavior - 3D Solid

Approach
LINK8 is a spar which may be used in a variety of engineering applications. Depending
the element
maySolid
be thought
of as a truss element, a cable element, a link el
Reinforcing Steelapplication,
Modeling
using
Elements
element, etc. The three-dimensional spar element is a uniaxial tension-compression elem
A truss element candegrees

be used
to model
steel
reinforcement.
of freedom
at eachthe
node:
translations
in the nodal x, Two
y, and znodes
directions. As in a
no bending
the element
is considered.
Plasticity,
creep, swelling, stress stiff
are required for this structure,
element.
Each of
node
has three
degrees
of freedom,
deflection capabilities are included. See Section 14.8 in the ANSYS Theory Reference fo
translations in the nodal
x, element.
y, and Az tension
directions.
about this
only compression-only element is defined as LINK10 and
Section
For example, ANSYS
uses4.10.
the LINK8, a uniaxial tension-compression
element, which is also capable of plastic deformation.
Figure 4.8-1 LINK8 3-D Spar
13
Modeling of Reinforced Concrete Behavior - Beam

Approach
Alternative View to the simulation of Degrading Hysteretic
Behavior
ents - Galerkin
The well known 1D beam element can be used as a simplified tool

reinforced column behavior in place of the
eight 3D
functions
and trial solutions
solid element
formulation.
ght functions
trial solutions
are
for theand
simulation
of the
ape functions are
This element has two degrees of freedom per node, one translational
(perpendicular to the beam axis) and one rotational.
14

Approach
The shape functions utilized from this element are the Hermite
Beam Elements - Shape Functions
Polynomials (see Lecture 6)
Hermite Polynomials
Polynomials
Hermite
15

Approach
Then, as we saw in Lecture 6, the elastic force deformation relationship, for
a prismatic beam without shearing deformations, is
Fi
12
6L
12
6L
vi
Mi EI 6L
4L2 6L 2L2
i or FE = KE v
Fj = L3 12 6L
12
6L
vj
2
2
6L 2L 6L 4L
Mj
j
Whilst , from Lecture 8, we saw that in case P-Delta effects are taken into
account, the geometric (nonlinear) stiffness is:
36
3L 36 3L
vi
Fi
2
Mi
T
3L L2
3L 4L
i
Fj = 30L 36 3L 36 3L vj or FG = KG v
3L L2 3L 4L2
j
Mj
Therefore, the total forces acting on the beam element will be:
FT = FE + FG = [KE + KG ]v = KT v
16
Moment Curvature Envelope
In order to simulate the effects of varying stiffness due to plasticity

appropriate plasticity model and a hysteretic law will be
introduced.
The hysteretic model is formulated based on an initial
moment-curvature relationship otherwise known as the backbone
skeleton curve.
Such skeleton curves must be defined for each different section type.
For instance, the bottom sections are more heavily reinforced than
the top. These curves can be either user defined or can be computed
using a fiber model.
17

Reinforced Concrete Design Calculations normally assume a simple material
model for the concrete and reinforcement so as to determine the moment
capacity of a section. The Whitney stress block for concrete along with an elasto
- plastic reinforcing steel behavior is one widely used material model.
18

However, the actual material behavior is nonlinear and can be described by
idealized stress-strain (material) models, as the ones introduced earlier.
19

Moment Curvature Analysis
is a method to accurately determine the load - deformation behavior of a
concrete section using nonlinear material stress-strain relationships.
For a given axial load there exists an extreme compression fiber strain
and a section curvature at which the nonlinear stress distribution is
in equilibrium with the applied axial load. Dividing the section into
fibers at distance z from the CG axis the strain distribution will be:
(z) = 0 + z
A unique bending moment can be calculated at this section curvature
from the stress distribution.
The extreme concrete compression strain and section curvature can be
iterated until a range of moment curvature values is obtained.
20

Software packages are available for generating Moment - Curvature
relationships by inputing the, geometry, reinforcement characteristics,
material properties and axial load for a given section.
Material properties for concrete can be obtained as a result of lab
compression tests on the utilized concrete mix.
Material properties from Steel can be directly obtained from the quality of
the reinforcing Steel
Software packages that can be used for the generation of Moment
Curvature Envelopes are:
SAP section designer (for those that have access to SAP2000)
Response 2000: http://www.ecf.utoronto.ca/ bentz/r2k.htm
MyBiaxial:
http://users.ntua.gr/vkoum/links-prog/MyBiaxial/mybiaxial.htm
21
Plasticity Model
There are different approaches for the modeling of inelastic behavior.

Concentrate Plasticity (plastic hinge approach)
Distributed Plasticity
Spread Plasticity
22
Plasticity Model
Concentrated Plasticity Model
23
Plasticity Model
Distributed & Spread Plasticity Models
24
Plasticity Model
Distributed & Spread Plasticity Models
source: Hwasung Roh, Andrei M. Reinhorn, Jong Seh Lee, Power spread plasticity model for inelastic analysis of
reinforced concrete structures, Engineering Structures, Volume 39, June 2012, Pages 148-16
25
Bouc - Wen Hysteretic Model

The smooth hysteretic model presented herein is a variation of the
model originally proposed by Bouc (1967) and modified by several
others (Wen 1976; Baber Noori 1985).
The use of such a hysteretic constitutive law is necessary for the
effective simulation of the behavior of R/C structures under cyclic
loading, since often structures that undergo inelastic deformations
and cyclic behavior weaken and lose some of their stiffness and
strength. Moreover, gaps tend to develop due to cracking causing
the material to become discontinuous.
The Bouc-Wen Hysteretic Model is capable of simulating stiffness
degradation, strength deterioration and progressive pinching effects.
(see: V. Koumousis, E. Chatzi and S. Triantafillou: Plastique A Computer Program For 3D Inelastic Analysis Of
Multi-Storey Buildings, Advances in Engineering Structures, Mechanics & Construction, Solid Mechanics and Its
Applications, 2006, Volume 140, Part 3, 367-378)
26
relation between generalized moments and curvatures is given by:
(t )
M (t ) Model
= M
+ (1 ) z (t )
Bouc - Wen Hysteretic

(1)
The
(t )
M (t ) = M y
+ (1 ) z (t )
where My is the yield moment; y is the yield curvature;
is the ratio of the post-yield to the initial
and
y
model
can be visualized as a linear
a nonlinear
element in parallel:
elastic stiffness and z(t) is the hysteretic componentdefined below.
(1)
where My is the yield moment; y is the yield curvature; is the ratio of the post-yield to the initial
elastic stiffness and z(t) is the hysteretic component defined below.
Figure 3. Bouc-Wen Hysteretic Model
The relation
betweenhysteretic
generalized
moments
is differential
given by:
The nondimensional
function z(t)
is the solutionand
of the curvatures
following non-linear

equation:
(t)
.
1
dz y
1
1 + sign
(d ) | z (t ) | + z (t )
.
(1
)z(t)
(t) = M
or alternatively
where K +
z (t ) = f ( (t), z (t )) M
=K
y = A B 2 2
nB
(2)
1 + sign(d ) | z (t ) | z (t ) Figure
1 sign(d3.
1 sign(d ) Model
| z (t ) | z (t )
| z (t ) | + z (t ) Hysteretic
E curvature;
the ratio of
where MyC is
the
yield
moment;
) Bouc-Wen
is
D
is the yield
2
2
2 y
2
2
2
the
post-yield
to
the
initial
elastic
stiffness
and
z(t)
is
the
hysteretic
The nondimensional
hysteretic
function
is the
solution
theoffollowing
differential
In the above expression
A, B, C,
D & E arez(t)
constants
which
control theofshape
the hystereticnon-linear
loop
component
defined
equation:for each
direction ofas:
loading, while the exponents n , n , n & n govern the transition from the elastic
nC
nD
z (t ) =
nE
to the plastic state. Small values of ni lead to a smooth transition, however as ni increases the transition
n
in the limit (n).
.becomes sharper
1 tending to a perfectly
dz bilinear behavior
1
1 + sign(d ) | z (t ) | + z (t ) B
or alternatively
where K z = A B
f (The
(t),program
z (t )) defaults
= Kz
are:
y
d
y
2
2
M y
1
1
, b =1 and nB = nE = n
(3)
A = 1, C = D =0 & B = n , E = n where e =
nC
nE
M y+ nD
bB
eE
1 + sign(d ) | z (t ) | z (t )
1 sign(d ) | z (t ) | + z (t )
1 sign(d ) | z (t ) | z (t )
Dgradient of the hysteretic
E occurs. Theassignment
the
unloading
C,
2The parameters
2 D control
2
2loop after
2
2

of null values for both, results to unloading stiffness equal to that of the elastic branch. Also, the
model is capable
of simulating
non symmetrical
so if the
yield moment
Institute
of Structural
Engineering yielding,
Method
ofpositive
Finite Elements
II is regarded
(2)
27
Bouc - Wen Hysteretic Model
In the above expression A, B, C, D & E are constants which control the

shape of the hysteretic loop for each direction of loading, while the
exponents nB , nC , nD & nE govern the transition from the elastic to the
plastic state. Small values of ni lead to a smooth transition, however as ni
increases the transition becomes sharper tending to a perfectly bilinear
behavior in the limit (ni ).
Plastique
Computer
forcan
Analysis
of Multi-Storey
371
Finally, A
the
flexuralProgram
stiffness
be expressed
as: Buildings
Finally, the flexural stiffness can be expressed as:
K = EI =
1
1
dM
dz
1
= M y + (1 ) = M y + (1 ) K z = EI 0 + (1 ) K z
d
d
y
y
y
(4)
6.1 Hysteretic behavior Variations

a) Stiffness Degradation
The stiffness degradation that occurs due to cyclic loading is taken into account by introducing the
parameter into the differential equation:
28
Modeling of Degradation
Stiffness Degradation
29
The parameter S depends on the damage of the section which is quantified by the Damage Index DI:
Modeling
of Degradation
S = 1 S d DI where DI =
max 1
c 1
S
dE
1 p1 diss
4 E mon
Sp2
(7)
In the above expression Sd, Sp1, Sp2 are constants controlling the amount of strength deterioration; c is
the maximum plasticity that can be reached, c = u / y ; dEdiss is the energy dissipated before
unloading occurs and finally Emon is the amount of energy absorbed during a monotonic loading until
failure as shown in Figure 4.
Figure 4. Dissipated Energy (Ediss) and Monotonic Energy (Emon).
The model can also be appropriately modified to simulate pinching.

Note: The Matlab code for the Bouc Wen Model will be provided!
30
Bouc Wen Model
Resulting Hysteretic Loops

Stiffness & Strength Degradation
Pinching
31
Bouc Wen Model
Dynamic Equation of Motion

g ). The general
This is a dynamic problem (input: base excitation U
equation of motion is therefore written as:
g (t)
MX(t)
+ CX(t)
+ KX(t) = MU
The Newmark Constant acceleration method outlined in Lecture 8 can
be used for the direct integration of the above equation.
(You can neglect the effect of damping for this project)
In order to achieve equilibrium within each time step, it is necessary to
implement a Newton - Raphson iterative scheme as outlined in Lecture 3.
32

Eigenvalue Problems
Where are these mostly encountered?

Vibration problems
Mode Shapes and Natural Frequencies
Buckling problems
Buckling Shapes and Critical Loads
Modal Analysis problems
For noise and durability issues
Eigenvalue Problems
Generalized Eigenvalue Problem
Ki = i Mi
where
K
Stiffness matrix
Mass matrix (or equivalent)
Eigenvector(mode shape)
Eigenvalue(natural frequencies or critical loads)
*This section is based on Prof. H. Waismans notes of the FEM II course - CEEM Department, Columbia University
Vibration Problems
Consider the following 1D physical problems, governed by the
Wave Equation
c2
2y
2y
= 2
2
x
t
Vibration Problems
General Wave Equation (Hyperbolic Equation)
c2
2y
2y
=
2x
2t
where c is the wave speed
An analytical solution to this problem can be found using separation

of variables:
y(x, t) = F (x)G(t)
substituting in the original expression we obtain:
c2
F 00
G00
=
= = constant
F
G
where > 0 to exclude exponentially growing solutions, hence:
G00 + G = 0 =2
y(x, t) = (Asint + Bcost)(Csin x + Dcos x)
c F 00 + F = 0
c
c
2
Longitudinal vibration of rods

Equilibrium Equation using = E
u
(constitutive relation)
x
u ( x, t )

u
2u
EA
= A 2
x
x
t
The above can be written in the form of the wave equation

s
2
2u
E
u
2
= 2
c
c=
x2
t
u(l, t)
=0
x
Boundary Conditions u(0, t) = 0
*In general we will also have initial conditions
Looking at the spatial part c2 F 00 + F = 0 u(x) = Asin x + Bcos x

c
c
u(0) = 0 B = 0
u(l)
= 0 cos l = 0
x
c
cos l = 0
c

3 5
(1)
l = , , , ..., n
c
2 2 2
2
We obtain the following Eq:
n = 1, 2, ...
The natural frequencies become

c 3c 5c
(1) c
n =
,
,
, ..., n
2l 2l 2l
2
l
n = 1, 2, ...
*Note: Natural frequencies are Directly dependent upon the B.C.

The mode shapes (eigenfunctions) that correspond to n are:
#
"
s
1 E
x
n (x) = Cn sin n
2 l

Modes Shapes are orthogonal with respect to a scalar product
Z l
k
j
< k (x), j (x) >=
Cj Ck sin xsin x dx = ... = 0 for k 6= j
c
c
0
Longitudinal vibration of rods-3

Note:
are
orthogonal
with
to a scalar product
Themodes
modesshapes
shapes
are
normalized
byrespect
requiring
1 =< k (x), k (x) >=
Ck2 sin2
weshapes
obtainare
thenormalized
following by
orthonormal
TheThus,
modes
requiring
r
n (x) =
k
x dx
c
set
h
r
xi
sin (2n 1)
Al
2 l
We obtain the following orthonormal set
The general solution is given by

u(x, t) =
(An sinn t + Bn cosn t)sin
The generaln=1
solution is given by
n
x
c

Finite Element Approach
Finite Element approach
The weak formulation of this problem is
Longitudinal
vibrationvibration
of rods
Longitudinal
of rods-5
The corresponding Galerkin formulation (with standard shape
Galerkin formulation
(with standard shape functions):
functions)
is
10
Longitudinal vibration of rods-6
The stiffness matrix
The Consistent Mass Matrix is defined by
Leads to the matrix eigenvalue problem
orthogonality condition
11
Free
vibrations
of thin beams
Free
vibrations
of thin beams-1
Consider the Euler-Bernoulli beam model
Clamped-Free
Assume a solution by separation of variables
The general solution is given by
12
Free vibrations of thin beams

The
Characteristic
Equation in of
this case
becomes
Free
vibrations
thin
beams-2
cosl coshl = 1
The characteristic equation
Finite Element approach (Hermite polynomials)
13
FreeFree
vibrations
of thin beams of
vibrations
thin beams-2
Stiffness matrix:
Consistent Mass Matrix:
14
Buckling Analaysis
Linear Buckling Analysis:
Linear buckling analysis can estimate the maximum load that can be
supported prior to structural instability or collapse. This method
corresponds to the textbook approach to elastic buckling analysis.
Nonlinear Buckling Analysis:
Imperfections and nonlinearities prevent most real-world structures
from achieving their theoretical elastic buckling strength. Nonlinear
buckling analysis provides a detailed buckling assessment and can
include geometric, material and boundary effects. This usually
involves a static structural analysis with large deflection effects
turned on. (large displacement approach Lectures 4-6)
15
Buckling Analaysis
Nonlinear Buckling Analysis:
Reminder, from Lecture 3, Standard Newton-Raphson methods perform
poorly for the case of buckling problems.
The Arc Length method can overcome these issues.
16
buckling strength. Thus, linear buckling analysis often yields quick but non-conservative
results.
Buckling
Analaysis
Governing Equation for Elastic Buckling
Here
we will deal with Elastic (linear) Buckling - also known as
Eigenvalue buckling
Consider a
a buckled
simply-supported
column of
length Lof
under
an external
axial compression
Consider
buckled
simply-supported
column
length
L under
an external
force F, as shown in the left schematic below. The transverse displacement of the buckled
axial
compression
column
is representedforce
by w. F , as shown in the left schematic below.
The right schematic shows the forces and moments acting on a cross-section in the buckled
The
rightMoment
schematic
shows
thelower
forces
acting
oninternal
a bending
column.
equilibrium
on the
free and
body moments
yields a solution
for the
moment M (u has
in place
of what isMoment
noted by w equilibrium
in the figure) on the lower free
cross-section
in been
the used
buckled
column.
body yields a solution for the internal
moment M
Fu Mbending
=0
17
Buckling Analysis
Equilibrium Equation
Fu M = 0
Recall the relationship between the moment M and the transverse
displacement w for an Euler-Bernoulli beam
M = EI
d2 u
dx2
Eliminating M from the above two equations results in the

governing equation for the buckled column
F u + EI
d2 u
=0
dx2
18
Buckling Analaysis
Ultimately the buckling equation can be written as (Pcr being the critical value for load F):
19
Buckling Analaysis
In this consideration,
an important
parameter
many
modes
Institute
of Structural
Engineeringis to decide
Methodhow
of Finite
Elements
II you would like to include 20

Review
Previous Lectures Overview
1
Part I - Nonlinear Analysis

Material Nonlinearity
Nonlinearities due to Large Displacements/Strains
Special Types of Problems (Contact, P- Effects)
Part II - Dynamic Analysis

Solution Methods
- Direct Integration
- Modal Superposition
Eigenvalue Problems
- Typical Formulations
- Solution for Eigenproblems (current lecture)
Solution Techniques for Eigenvalue Problems

Generalized Eigenvalue Problem
Ki = i Mi
where
K
M
i
i
Stiffness matrix (n n)
Mass matrix (or equivalent) (n n)
Eigenvector - mode shape (n 1)
Eigenvalue - natural frequencies (i2 ) or critical loads
The eigenvectors satisfy:

Ti Mj = ij
Ti Kj = i ij
M-orthonormality
K-orthogonality
If additionally K is positive definite (as is the case for structural problems)

0 1 2 ... n
Solution Techniques for Eigenvalue Problems

In practice, we are generally interested in the lowest modes.
Thus, we wish to extract:
modes
{k k }nk=1
where
nmodes << n
Solution Methods
1
Characteristic Polynomial Technique
Vector Iteration Methods
Transformation Methods
Characteristic Polynomial Technique

As already presented in Lecture 9 (Modal Superposition Methods),
solve for:
(K i M)i = 0 = det(K i M) = 0
We can use polynomial iteration techniques (even Horners scheme
for low order polynomials) to find the roots of the above equation.
The solution will yield:
n eigenvalues i = i2 ,
i = 1, ..., n
with 0 1 ...n (the eigenfrequencies)

and the solution of
2 M = K (eigenproblem)
will yield:
n eigenvectors (or modal vectors) i

are based on the notion of Rayleigh Quotient defined as
R() =
T K
T M
scalar quantity
Consider the Standard Eigenvalue Problem

Ai = i i
This is equivalent to our generalized problem with A = M1 K
is the best fit or least square eigenvalue:

is orthogonal to the component (A y)
y = being the projection of A on
T (A y) = T (A ) = 0
T K
scalar quantity
(T K = M) = R() = T
M

include the following alternative schemes
Inverse Iteration
Power Iteration
Shifting in Vector Iteration
Rayleigh Quotient Iteration
Matrix Deflation and Gram-Schmidt Orthogonalization

Differences amongst Schemes
Inverse Iteration
In general we can calculate all Eigenvalues, but with very small
convergence. Only the first Eigenvalue has a good convergence
Power Iteration
Gives the last Eigenvalue
Able to yield Eigenvalues in between the highest and lowest one
using the inverse iteration
Gram-Schmidt Orthogonalization
Gives the starting vector to calculate all Eigenvalues using the
inverse iteration

Inverse Iteration
The inverse iteration algorithm assumes a starting iteration vector x1
and an associated vector y1 = Mx1 .
The method is described by the following steps:
For k = 1, 2, ... Solve
K
xk+1 = yk
y
k+1 = M
xk+1
x
T yk
(
xk+1 ) = T k+1
x
k+1 y
k+1
yk+1
yk+1 = T
(
xk+1 y
k+1 )1/2
where provided that y1T 1 6= 0,
yk+1 M1
and (
xk+1 ) 1
as k
10

Forward Iteration
This converges to the largest eigenvalue by solving M = 1 K
The algorithm assumes a starting iteration vector x1 and an
associated vector y1 = Kx1 . The method is described by the
following steps:
For k = 1, 2, ... Solve
M
xk+1 = yk
y
k+1 = K
xk+1
T
x
y
k+1
(
xk+1 ) = k+1
T
x
k+1 yk
yk+1
yk+1 = T
(
xk+1 yk )1/2
where provided that Tn y1 6= 0,
yk+1 Kn
and (
xk+1 ) n
as k
11

Assuming that a shift is applied then we consider the
Eigenproblem:
(K M) = M
where the eigenvalues of the original problem K = M and of
the shifted problem are related as:
i = i
i = 1, 2, ..., n
12

Inverse iteration is again applied assuming a starting iteration vector
x1 and an associated vector y1 = Mx1 .
The method is described by the following steps:
For k = 1, 2, ... Solve
(K M)
xk+1 = yk
y
k+1 = M
xk+1
x
T yk
+
(
xk+1 ) = T k+1
x
k+1 y
k+1
yk+1
yk+1 = T
(
xk+1 y
k+1 )1/2
13

Remarks
The convergence rate of inverse iteration can be actually improved
by shifting since we can in this way effectively moving towards areas
closer to an eigenvalue. In practice, the difficulty lies in choosing an
appropriate shift. One possibility is to use the Rayleigh quotient,
which is an approximation to the eigenvalue sought:
(
xk+1 ) =
x
Tk+1 yk
x
Tk+1 y
k+1
If a new shift is evaluated in each iteration we obtain the Rayleigh

quotient iteration method Assuming that a shift is applied then we
consider the eigenproblem:
14

Rayleigh Quotient Method
In this procedure we again assume a starting iteration vector x1 and
an associated vector y1 = Mx1 , but also a starting shift (
x1 ) which
is usually zero and then evaluate
For k = 1, 2, ...
(K (
x1 )M)
xk+1 = yk
y
k+1 = M
xk+1
x
T yk
(
xk+1 ) = T k+1
+ (
x1 )
x
k+1 y
k+1
yk+1
yk+1 = T
(
xk+1 y
k+1 )1/2
where now
yk+1 Mi
and (
xk+1 ) i
as k
i , i depend on the initial choices of x1 and (

x1 ) .
15

Matrix Deflation
This method is employed when we have already calculated an
Eigenpair, say (k , k ), using another method and we require the
solution of another Eigenpair.
The method is used in order to deflate the original Eigenpair and
then input it in the inverse iteration process (or forward iteration) so
that we converge to a separate pair. The basis of vector deflation is
that in order for an iteration vector to converge in forward or inverse
iteration to a required eigenvector, the iteration vector must be
orthogonal to it. Hence, we eliminate the possibility that the
iteration converges to any of the already specified solutions.
A particular vector orthogonalization procedure is the
Gram-Schmidt method.
16

Gram-Schmidt Orthogonalization
In order to consider a general case, assume that we have calculated in
inverse iteration the eigenvectors 1 , 2 ,...,m and that we want to
M-orthogonalize x1 to these eigenvectors. This is calculated using:
1 = x1
x
m
X
i i
(1)
i=1
where the coefficients i are obtained using

Ti M
x1 = 0, i = 1, ..., m
Ti Mj = ij
Premultiplying both sides of Eqn (1) by Ti M we obtain:

x1
i = Ti M
i = 1, ..., m
1 as the starting iteration vector

In the inverse iteration we now use x
instead of x1 .
17
The Transformation Methods comprise a group of eigensystem
solution procedures that employ the basic properties of the
eigenvectors for matrix ,
T K = 2
T M = I
Since , of order n n, diagonalizes K, M we can try to construct

it by iteration.
The Jacobi Method (1845) enforces such an approach. The
method is developed for the standard Eigenproblem (equivalent to
our generalized problem with A = M1 K):
Ai = i i
for which the Eigenvectors have the property:
T A = 2
18
One of the oldest techniques introduced by Jacobi in 1845.
Jacobi Method
The idea is to apply a series of matrix multiplications
The starting transformation matrix is
and then
19
Solution Techniques for eigenproblem

Jacobi Method
This matrix is known as

Givens rotation matrix
The idea is to solve the 2x2 matrix (select ) such that
20
Solution Techniques for eigenproblems
Jacobi
Jacobi Method
Algorithm
Basic step
1. Choose an index pair i,k
2. Select (c,s) for the rotation such that the off diag. terms are
zeroed. Note that
3. Overwrite A with
note that the only changes are in rows i,k
4. Choice of (c,s):
21
Solution Techniques for eigenproblems

JacobiAlgorithm
Method
Jacobi
Example: We wish
to find the eigenvalues
of the matrix using
Jacobi method
22
Solution
Jacobi
Method
Techniques for eigenpr
The first rotation matrix will zero
term:(1,3)
23
Jacobi Method
After six more iterations
The eigenvectors are the
24

Final Project Analysis Guide

Dynamic Shake Table Test
Specimen
1285
1535
Use an FE approach in order to simulate the experimentally observer

response of the column specimen in cyclic loading. The experiment
simulates the response of an R/C bridge pier.
Damper
Servo hydraulic test cylinder

Piston: + / - 125 mm
Piston force: + / - 100 kN
250
Anchorage
ETH ShakeTable
Note: If preferable, you may use commercial FE software for the

simulation, instead of the suggested MATLAB scheme.

Elevation Drawings
=276Kg

Plan Drawings
AVAILABLE DATA
Accelerometer Measurements
Displacement Measurements
OBJECTIVE
The aim is to simulate the Top
Displacement - Base Shear Force
Hysteretic Loop
Initialization
Define
Geometry
Number of Finite Elements
Connectivity

Step 1 - Choice of Concrete Material Model
-
cu
peak compressive stress

E0
Compression
softening
strain at maximum stress
+
o
cu
tu = maximum tensile strength of concrete
Tension
+
Figure 2.5: Typical uniaxial compressive and tensile stress-strain curve for concrete (Bangash 1989)
The properties will be defined

from the curve provided from the
results of the compression test
performed for two cylindrical
samples of the material.
Self Compacting Concrete was
utilized for the tests.
For preliminary calculations you
may assume a cylinder strength of
60 MPa.
Typical uniaxial compressive and tensile

stress-strain
curveis linearly
for concrete
mpression, the stress-strain
curve for concrete
elastic up to about 30 percent of
aximum compressive strength. Above this point, the stress increases gradually up to the
mum compressive strength. After it reaches the maximum compressive strength cu , the
descends into a softening region, and eventually crushing failure occurs at an ultimate
Institute
of Structural
Engineering
Method
cu . In tension, the stress-strain curve
for concrete
is approximately
linearly elastic
up to of Finite Elements II
iffness of the entire structure based on diaphragmatic
conventional stress-strain curves both for unconfined

parabolic stress-strain relationship with a softening
s-strain diagram with
- hardening is implemented. The
d in Figure 1.
Step 2 Choice of Steel Material Model
Bilinear
Model
Hardening
nconfined
concrete
and b)with
for reinforcing
steel
You may use a simplified bilinear

material model with an
assumption for hardening.
The data for reinforcing steel can
be obtained from the excel file
entitled:
Steel data.xls which will be
uploaded on the website
ral elements namely; beams, columns and shear walls

storey buildings. By combining such elements one
inked together through Institute
diaphragms
at theEngineering
floor levels Method of Finite Elements II
of Structural
ents - Galerkin
Using the
beam
element
ght functions
and trial
solutions
are
model the
concrete
eight to
functions
andreinforced
trial solutions
column behavior
ape functions are
This element has two degrees of freedom per node, one translational
(perpendicular to the beam axis) and one rotational.
You may ignore the effects of Geometric Nonlinearity - P- Effect during
the analysis.
As discussed in Lecture 10, in order to simulate the effects of varying

stiffness due to plasticity appropriate plasticity model and a hysteretic
law will be introduced.
For these we needMethod
an initial
Moment-Curvature relationship.
of Finite Elements II
25

Step 3 - Deriving the Moment Curvature Envelope (M )
These envelopes must be defined for each different section type. For
instance, the bottom sections have more shear reinforcement and a larger
axial load than the top ones. These curves can be either user defined or
can be computed using a fiber model approach implemented by any of the
following software:
SAP section designer (for those that have access to SAP2000)
Response 2000: http://www.ecf.utoronto.ca/ bentz/r2k.htm
MyBiaxial:
http://users.ntua.gr/vkoum/links-prog/MyBiaxial/mybiaxial.htm
After obtaining the M envelopes for the different sections, approximate
them with a bilinear model. In this way a yield point (y , My ) will be
specified for each section. Next, discretize the specimen into finite beam
elements with a corresponding (M ) diagram for each end section.
10

Step 4 - Initiate the Newmark Method loop
g . (This is
This is a dynamic problem with input: base excitation U
measured using an accelerometer at the base of the table and will be
provided as a text file)
Dynamic Equation of Motion
g (t)
MX(t)
+ KX(t) = MU
The Newmark Constant acceleration method ( = 0.5, = 0.25)
outlined in Lecture 8 can be used for the direct integration of the above
equation.
You can neglect the effect of damping for this project (C = 0)
Step 5 - Initiate the Newton Raphson loop

Due to yielding Nonlinear phenomena occur leading to progressive change
of stiffness, therefore we need a nonlinear solution approach for the solution
of the balance equations at each time step
11

Step 6
Formulate the Element Stiffness Matrix using a Spread Plasticity Model
A stiff.file will be uploaded on the website with:

Input the previous yield penetration lengths , for the element, the
elastic shear and breding stiffness for the element, the length of the
element and the yield moments as well as the current stiffness and moment
at each end section of the element
Output the previous yield penetration lengths , for the element and the
4 4 stiffness matrix for a beam element.
12

Step 7 - Total Matrices
Assemble the total Stiffness Matrix using the elemental matrices and
taking into account the connectivity of the elements. You already have the
code for this from previous HWs. This time though notice that there is two
degrees of freedom per node.
Assemble the total Mass Matrix using the elemental matrices.
Translations dofs: Make the assumption of lumped masses, meaning that
for each element you will use a diagonal mass matrix with a diagonal terms
equal to half of the mass of the finite element. The only exception is the
top node (final horizontal dof) where the concentrated mass is assumed to
be fixed.
Rotational dofs: You need only consider the moment of inertia of the
concentrated mass at the top. The rest rotational diagonal terms can be
assumed to be zero.
Damping will be neglected.
13

Step 8 - Total Matrices
Formulate the Effective Stiffness and load matrices for the Newmark
Method
g Pint in order to formulate the external force vector.
Use t R = MU
Proceed to formulating the effective load vector t R

Solve for the current iteration incremental displacement U and derive the
and velocity U .
current total displacement U,acceleration U
Also, calculate the additional internal forces F=K*U for each element
F=[V M] - It obviously involves both a shear and a moment
component, corresponding to the translational and rotational dofs.
14
relation between generalized moments and curvatures is given by:
Final Project AnalysisM (Guide

(t )
t) = M
+ 1 z (t )
y
(1)
Step where
9 Bouc
- Wen Hysteretic Model
M is the yield moment; is the yield curvature; is the ratio of the post-yield to the initial
y
elastic stiffness and z(t) is the hysteretic component defined below.
Figure 3. Bouc-Wen Hysteretic Model

The nondimensional hysteretic function z(t) is the solution of the following non-linear differential
equation:
A bouc.m file will be uploaded on the website with:

Input previous
curvature ( ), additional curvature ( = Mi /EIi ),
. (t ) = f (.(t), z (t )) 1 or alternativelyi dz = K 1 where K = A B 1 + sign(d ) | z (t ) | i+ z (t )
2
2 1,2.
i =
previous zinternal
moment
(Mdi ), for
each end
section
(2)
1 + sign(d curvature,
) | z (t ) | z (t )
1 sign(d ) |internal
z (t ) | + z (t ) element
1 sign(d ) force
| z (t ) | z (t(moment,
)
Output current
current
shear),
C

2
2
2
2
2
2

current bending stiffness
(EIi )
nB
nC
nD
nE
In the above expression A, B, C, D & E are constants which control the shape of the hysteretic loop
for each direction of loading, while the exponents nB, nC, nD & nE govern the transition from the elastic
The file towill

alsostate.
define
the offollowing
parameters:
transition, however as n increases the transition
the plastic
Small values
n lead to a smooth
becomes sharper tending to a perfectly bilinear behavior in the limit (n).
Ni (controls
the
smoothness
of
the
hysteretic
loop), Sk (controls the
The program defaults are:
(controls
M
1 Sp1
1 [0
stiffness degradation
effect),
1]
the strength
, E=
where e =
, b =1 and n = n = n
(3)
A = 1, C = D =0 & B =
M
b
e
deterioration effect), A, zm , Zs (control the pinching effect)
i
nB
nE
y
+
y
The parameters C, D control the gradient of the hysteretic loop after unloading occurs. The assignment
of null values
for both,
results to unloading
stiffness equal
to that
of theElements
elastic branch.
Also, the
Institute
of Structural
Engineering
Method
of Finite
II
15

Tuning of the aforementioned parameters will produce hysteretic loops of
different shapes enabling you to closer approximate the experimentally
observed response.
Example of resulting Hysteretic Loops
Stiffness & Strength Degradation
Pinching
16

Step 10
Check Convergence
Check if the error in the residual force vector is less than a specified
tolerance:
g Pint MU
tol
Perr = MU
If not continue with the Newtorn-Raphson iterations

If yes go to the next time step t + T
Plot the computed top Relative Displacement versus the
experimentally measured one and plot the Moment Curvature
relationship for the bottom element (Hysteretic Loop).
17

Special Topics
The standard finite element method has been used with great success
in many fields with both academic and industrial applications. It is
however not without limitations. Due to mesh-based interpolation,
distorted or low quality meshes lead to higher errors, necessitate
remeshing, a time and human labor consuming task, which is not
guaranteed to be feasible in finite time for complex three-dimensional
geometries. Additionally, they are not well suited to treat problems
with discontinuities that do not align with element edges.
Alternative/Extended Approaches
Meshless Methods
Multigrid Methods
Extended Finite Element methods
Special Topics
Meshless Methods
Traditional simulation algorithms relied on a grid or a mesh, Meshless
methods in contrast use the geometry of the simulated object directly for
calculations. Meshless (or Meshfree) methods (MMs) were born with the
objective of eliminating part of the difficulties associated with reliance on a
mesh to construct the approximation.
Special Topics
Meshless Methods
Meshless approximations for a scalar function u in terms of the
material (Lagrangian) coordinates can be written as
u(x, t) =
Ni (x)ui (t)
iS
where Ni : R are the shape functions and the ui are the nodal
values at particle i located at position xi and S is the set of nodes i
for which Ni (x) 6= 0. In contrast to FEM, the shape functions are
only approximants and not interpolants, since ui 6= u(xi ).
Therefore, special techniques are needed to treat displacement
boundary conditions.
Special Topics
Meshless Methods
Disadvantages
Advantages
problems with moving
discontinuities such as crack
propagation and phase
transformation can be treated
with ease
The MMs shape functions

are rational functions which
requires high-order
integration scheme to be
correctly computed
large deformation can be

handled more robustly
The treatment of essential

boundary conditions is not as
straightforward
higher-order continuous
shape functions
non-local interpolation
character
no mesh alignment sensitivity.
They do not satisfy the

Kronecker delta property
In general, the computational
cost of MMs is higher than
one of FEM.
Special Topics
Multigrid Methods
Multigrid (MG) methods in numerical analysis are a group of
algorithms for solving differential equations using a hierarchy of
discretizations. They are a class of techniques very useful in
problems exhibiting multiple scales of behavior. They are fast linear
iterative solvers based on the multilevel or multi-scale paradigm.
The typical application for

multigrid is in the
numerical solution of
elliptic partial differential
equations in two or more
dimensions.
Special Topics
Multigrid Methods
Special Topics
Multigrid Methods
There are many variations of multigrid algorithms, but the common
features are that a hierarchy of discretizations (grids) is considered.
The important steps are:
Smoothing Perform some steps of a basic method in order to
smooth out the error.
Restriction Restrict the current state of the problem to a
subset of the grid points, the so-called coarse grid, and solve
the resulting projected problem.
Interpolation or Prolongation Interpolate the coarse grid
solution back to the original grid, and perform a number of
steps of the basic method again.
Special Topics
Multigrid Methods - Adaptive Mesh
Adaptive multigrid exhibits adaptive mesh refinement, that is, it

adjusts the grid as the computation proceeds, in a manner dependent
upon the computation itself. The idea is to increase resolution of the
grid only in regions of the solution where it is needed.
Special Topics
Multigrid Methods
Advantages
MG can be applied in combination with any of the common
discretization techniques.
Multigrid is among the fastest solution techniques known today.
In contrast to other methods, multigrid is general in that it can
treat arbitrary regions and boundary conditions.
Multigrid does not depend on the separability of the equations
or other special properties of the equation.
MG is also directly applicable to more complicated,
non-symmetric and nonlinear systems of equations.
Can be very easily parallelized.
10
Special Topics
The Extended Finite Element Method (XFEM)
The Extended Finite Element Method is a numerical method that
enables a local enrichment of approximation spaces. The enrichment
is realized through the partition of unity concept.
The method is useful for the approximation of solutions with
pronounced non-smooth characteristics in small parts of the
computational domain, for example near discontinuities and
singularities. In these cases, standard numerical methods such as the
FEM often exhibit poor accuracy.
The XFEM offers significant advantages by enabling optimal
convergence rates for these applications.
11
Special Topics
XFEM - Motivation
Modeling of discontinuities, ex. structural flaws (cracks, holes)
12
Special Topics
XFEM - Motivation
Modeling of complicated crack geometries (ex. non planar cracks)
A T-Joint with Crack on a

Given Curved Plane
13
Special Topics
XFEM vs FEM
- XFEM locally enrich the standard FE approximation with local
partitions of unity enrichment functions
- This allows for the separation of geometry from the mesh.
Standard FEM
X-FEM
14
Special Topics
XFEM Scheme
The conventional shape function space is enriched with a set of
enrichment functions H(x) and Fj (x)
nJ
nT
4
n
X
X
X
X
NI (x)
NI (x)uI +
NI (x)H(x)aI +
Fj (x)bjI
uh (x) =
I=1
I=1
I=1
j=1
Tip Enrichment
Fj (r(x), (x)) = { r sin , r cos , r sin sin , r cos sin }j=1,..,4

2
2
2
2
Jump Enrichment
(
1
H(X) =
0
above +
c
bellow
c
15
Special Topics
XFEM Scheme - Tip and Jump Enrichments
16
Special Topics
XFEM Applications - Composite Materials
G XFEM = 4.762
G ABAQUS = 4.703
Gexp eriment = 4.784
[lb in/in^2]
Hart D., NSWCCD [2006]

5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
coordinate along crack
17
Special Topics
XFEM Applications - Crack Growth
Load P
6.25
Unit: mm
Simplification
35.0
15.63
BC 1
Fix U,V
7.813
BC 2
Fix U
6.25
37.5
100
18
Special Topics
Crack Initiation Site:

Experimental Observation
19
Special Topics
20
Introduction to FEM Software

*The section on Introduction to FEM Software is based on

Prof. M. Fabers notes of the FEM II course - Fall 2009
21

Overview
FEM development
Introduction to the use of Finite Elements
Modeling the physical problem
Finite elements as a tool for computer aided design and
assessment
An overview of FEM software (open source and commercial)
Example: ABAQUS
22

FE development
FEM development
FE is the confluence of three ingredients: matrix structural
analysis, variational approach and computer
1950s, M.J. Turner at Boeing (aerospace industry in general):
Direct Stiffness Method
Academia: J.H. Argyris, R.W. Clough (name finite element),
H.C. Martin and O.C. Zienkiewicz popularization
1960s, Melosh and De Veubeke: Variational Approach
Commercial finite element computer codes
23

FE Basics
Within the framework of continuum mechanics dependencies
between geometrical and physical quantities are formulated on a
differentially small element and then extended to the whole
continuum
As a result we obtain differential, partial differential or integral
equations for which, generally, an analytical solution is not
available they have to be solved using some numerical
procedure
MFE is based on the physical discretization of the observed
domain, thus reducing the number of the degrees of freedom;
moreover the governing equations are in general algebraic
24

FE Steps
Continuum is discretized in a mesh of finite elements
Elements are connected at nodes located on element boundaries
State of deformation, stresses, etc. in each element is described
by interpolation (shape) functions and corresponding values in
the nodes; these node values are basic unknowns of the method
of finite elements
The way in which these three steps are approached has a great
influence on the results of the calculations
25

1.3
Modeling the Physical Problem

Ideal
Mathematical
model
THE FEM ANALYSIS PRO
generally
irrelevant
CONTINUIFICATION
SOLUTION
Physical
system
FEM
Discrete
model
IDEALIZATION &
DISCRETIZATION
Discrete
solution
VERIFICATION
solution error
simulation error= modeling + solution error
VALIDATION
26

Modeling the Physical Problem
MFE is only a way of solving the mathematical model
The solution of the physical problem depends on the quality of
the mathematical model the choice of the mathematical model
is crucial
Thus, mathematical model must be reliable and effective
The chosen mathematical model is reliable if the required
response can be predicted within a given level of accuracy
measured on the response of a very comprehensive
mathematical model
The most effective mathematical model for the analysis is the
one that gives the required response with sufficient accuracy
and at least costs
27

Example
Complex physical problem (steel bracket) modeled by a simple
mathematical model
Complexphysicalproblem(steelbracket)modelled
p
p y
p
(
)
byasimplemathematicalmodel
28

Example
Detailed reference model 2D plane stress model for FE analysis
29

Example
Choice of mathematical model must correspond to desired
response
The most effective mathematical model delivers reliable answers
with the least amount of effort
Any solution (including MFE) of a mathematical model is
limited to information contained in or fed into the model: bad
input bad output (garbage in garbage out)
Assessment of accuracy is based on comparisons with the
results from very comprehensive models but in practice it has
to be based on experience (experiments)
30
FE as a Tool for CAD

Practical application requires that solutions obtained by MFE
are reliable and efficient
However, it is also necessary that the use of FE is robust this
implies that minor changes in any input to an FE analysis
should not change the response quantity significantly
The engineer (user) should be able to judge the quality of the
obtained results (i.e. for plausibility)
31
Implementation of FE
Calculation of system matrices K, M, C and R whichever
applicable (nodal point and element information are read;
element stiffness matrices, mass and damping matrices and
equivalent loads are calculated; structure matrices are
assembled)
Solution of equilibrium equations
Evaluation of element stresses
32

Implementation of FE - Flowchart
33

Open Source Packages
CalculiX is an Open Source FEA project. The solver uses a
partially compatible ABAQUS file format.
OpenSees, the Open System for Earthquake Engineering
Simulation, is an object-oriented, open source software
framework created at the NSF-sponsored Pacific Earthquake
Engineering (PEER) Center.
FEAP is a general purpose finite element analysis program
which is designed for research and educational use.
http://www.ce.berkeley.edu/projects/feap/
ForcePAD: educational software forcepad.sourceforge.net
Sundance: a software package developed at Sandia National
Laboratories
34

Commercial Packages
ABAQUS: Franco - American software from SIMULIA
ADINA R&D, Inc. See http://www.adina.com/
ANSYS: American software
COMSOL Multiphysics
COSMOSWorks: A SolidWorks module
GTSTRUDL30
LS-DYNA, LSTC - Livermore Software Technology Corporation
Nastran: American software
CUBUS Swiss Software Structurally Oriented
SAP: Structurally Oriented
35
ABAQUS Features
Wide material modeling capability
Ability to be customized
A good collection of multiphysics capabilities, such as coupled
acoustic - structural piezoelectric etc.
Attractive for simulations where multiple fields need to be
coupled
36

ABAQUS Multiphysics Example
Example: Shock Response and Acoustic Radiation Analysis for marine
structures
Most naval organizations require some form of shock survival assessment to
be performed on marine structures before they are commissioned
Besides shock survivability, another important consideration in submarine
design is stealth. Since their inception, submarines have been deployed in
strategic situations where susceptibility to detection is equated with
reduced effectiveness of the vessel. Various acoustic treatments are used to
reduce the target strength and acoustic signature of a submarine.
ABAQUS provides fully coupled structural acoustic capabilities in the
frequency domain that predict near and far-field sound pressure levels due
to machinery-induced vibrations.
37

The structure was meshed with quadrilateral shells and the fluid with
acoustic tetrahedra.
38
14 and Figure 15 show the Mises stress

Introduction to FEM Software Figure
distribution on the external surface of the submarine
and in the internal compartments, respectively.

Stresses are notably lower in the areas where the
hulls overlap, as is expected due to the enhanced
thickness there.
Figure 14. Mises stress distribution on the exterior

surface of the submarine after 10 ms.
Figure 12. Acoustic pressure distribution in the

outer water after 10 ms.
presence of the terminating fluid boundary. This
indicates that the fluid zone is sufficiently large, so
there are no significant spurious reflections from the
Institute
of Structural
Engineering
boundary. Since the effect
of hydrostatic
pressure
is
Figure 15. Mises stress distribution in the internal

structures of the submarine after 10 ms.
The results of the acoustic radiation analysis are
Method
of Finite
II Figure 17, and Figure 18
examined
next. Elements
Figure 16,
39

Lecture 2 - 2 October, 2012
Non Linear FE - Background and Motivation
Linear Analysis Assumptions

Infinitesimally small displacements
Linearly elastic material
No gaps or overlaps occurring during deformations - The
displacement field is smooth
Nature of Boundary Conditions remains unchanged
Steady State Assumption
No dependence on time
What kind of problems are not steady state and linear?

Material behaves Nonlinearly
Geometric Nonlinearity (ex. p- effects, follower force)
Contact Problems (Hertzian stress)
Loads vary fast compared to the eigenfrequencies of the
structure
Varying Boundary conditions (ex. freezing of water, welding)
General feature:
Response becomes load path dependent
What is the added value of being able to assess the Nonlinear

non-steady state response of structures?
Assessing the:
Structural response of structures to extreme events (rock-fall,
earthquake, hurricanes)
Performance (failures and deformations) of soils
Verifying simple models
Collapse Analysis of the World Trade Center

Ultimate collapse capacity of jacket structure

Analysis of soil performance

Analysis of bridge response

Steady state problems (Linear/Nonlinear):
The response of the system does not change over time
KU = R
Propagation problems (Linear/Nonlinear):
The response of the system changes over time
MU(t)
+ CU(t)
+ KU(t) = R(t)
Eigenvalue problems:
No unique solution to the response of the system
Introduction
to Nonlinear
analysis analysis
Introduction
to non-linear
Classification
of of
Nonlinear
Classification
non-linear analyses
analyses
Type of analysis
Description
Materially-nonlinear Infinitesimal
only
displacements and
strains; stress train
relation is nonlinear
Large
Displacements and
displacements, large rotations of fibers
rotations but small
are large; but fiber
strains
extensions and
angle changes
between fibers are
small; stress strain
relationship may be
linear or non-linear
Large
Displacements and
displacements, large rotations of fibers
rotations and large
are large; fiber
strains
extensions and
angle changes
between fibers may
also be large; stress
strain relationship
may be linear or
non-linear
Typical
formulation used
Materiallynonlinear-only
(MNO)
Stress and strain

measures used
Engineering strain
and stress
Total Lagrange (TL) Second PiolaKirchoff stress,

Green-Lagrange
strain
Updated Lagrange
(UL)
Cauchy stress,
Almansi strain
Total Lagrange (TL) Second PiolaKirchoff stress,

Green-Lagrange
strain
Updated Lagrange
Cauchy stress,
(UL)
Logarithmic strain
10
Introduction to Nonlinear analysis

Page 21
Swiss Federal Institute of Technology
Linear
Elastic to non-linear analysis
Introduction
Classification of non-linear analyses
P
2
P
2
= P/ A
= / E
=L
E
1
< 0.04
Linear elastic (infinitesimal displacements)

Infinitesimal Displacements
11

Page 22
Material
Nonlinearity
only
Introduction
to non-linear
analysis
P
2
L
P/ A
Y
= P/ A
Y Y
=
+
E
P
2
ET
ET
E
1
< 0.04
Materially nonlinear only (infinitesimal
displacements, but
relation)
Infinitesimal Displacements,
butnonlinear
Nonlinearstress-strain
Stress Strain
relation
12

Large displacements, small strains
< 0.04
= L
Linear or Nonlinear material behavior

13
Introduction to Nonlinear
analysisanalysis
Introduction
to non-linear
Large displacements, large strains
Linear or Nonlinear material behavior
Large
displacements,
large rotations and
14
Introduction
Nonlinear
analysis
Swissto
Federal
Institute of Technology
Introduction to non-linear analysis
Change in BC for displacement

P
2
P
2
Chang in boundary conditions

15
Example: Simple Bar Structure

Material Nonlinearity
Swiss Federal Institute ofonly
Technology
Assumptions: Small displacements, strains, load is applied slowly.
Area = 1cm
u
t
Section a
Section b
ET
Lb = 5cm
La = 10cm
Y = 0.002
4
3
E = 107 N / cm 2
ET = 105 N / cm 2
Y : yield stress
Y : yield strain
Calculate the displacement at the point of load application.

16

Area = 1cm 2
E = 107 N / cm 2
Section a
ET = 105 N / cm 2
Section b
ET
Y : yield stress
Y : yield strain
Lb = 5cm
La = 10cm
= 0.002
t
t
u t
u
, b =
La
Lb
a =
R + b A = a A
4
3
2
(elastic region)
= Y +
ET
(plastic region)
(unloading)
E
17

In the beginning both Sections are elastic
18

Section A is elastic while Section B is plastic
Since the stress on Section B is higher, it will yield first at time t :
Unloading occurs before Section A yields.

19
Introduction to Nonlinear Analysis

Conclusion from the previous example:
The basic problem in general Nonlinear analysis is to find a state of
equilibrium between externally applied loads and element nodal forces
t
R t F = 0
R =t RB +t RS +t RC
XZ
t (m)T t (m) t
t
B
dV(m)
F=
m
tV
where RB : body forces, RS : surface forces, RC : nodal forces

We must achieve equilibrium for all time steps when
incrementing the loading
Very general approach
Includes implicitly also dynamic analysis!
20
Types of Response Diagrams

Basic Types
F
21

Complex Types
R
U
22
Solution Algorithms for NL equations

Root finding for single variable NL problems f (x) = 0
Fixed Point Iteration
Bisection Method
Assumption: f [a, b] < and

continuous
If f (a) > 0, f (b) < 0 a x
b f (
x) = 0
Write f (x) = 0 in the form f (x) = x q(x),

the solution x satisfies x = q(
x)
Recurrence relation: xk+1 = g (xk )
Convergence: If g 0 (x) is defined over [a, b] and a
positive constant K exists with |g (x)| K ,
x [a, b] then g (x) has a unique fixed point
x [a, b].
23

Newton (Raphson) Method
Secant Method
Defined by the recurrence relation

xk+1 = xk
f(xk )
f 0 (xk )
terminate when |xk+1 xk | , <<

Convergence: quadratic
|
x xk+1 | C |
x xk+1 |2

x x
k1
xk+1 = xk f(xk ) f(xkk)f(x
k1 )
Convergence:
superlinear order
= 1+2 5 (golden ratio)
24
Incremental Analysis
The basic approach in incremental analysis is:
Find a state of equilibrium between externally applied loads and
element nodal forces
t+t
Assuming that
t+t R
R t+t F = 0
is independent of the deformations we have

t+t
R =t F + F
We know the solution t F at time t and F is the increment in the

nodal point forces corresponding to an increment in the
displacements and stresses from time t to time t + t. This we can
approximate by
F =t KU
25
Newton-Raphson Method
Assume the tangent stiffness matrix:
t F
t U
We may now substitute the tangent stiffness matrix into the
equilibrium relation
t
K=
KU =t+t R t F
which gives us a scheme for the calculation of the displacements:

t+t
U =t U + U
The exact displacements at time t + t correspond to the applied

loads at t + t, however we only determined these approximately as
we used a tangent stiffness matrix thus we may have to iterate to
find the solution.
26
We may use the Newton-Raphson iteration scheme to find the
equilibrium within each load increment
t+t
K(i1) U(i) =t+t R tt F(i1)
(out of balance load vector)

t+t
U(i) =t+t U(i1) + U(i)
with Initial Conditions

t+t
U(0) =t U;
t+t
K(0) =t K;
t+t (0)
=t F
27
The first two analysis types, although significantly simplified, can lead to valuable conclusions
concerning the behavior of the structure and the possible collapse mechanism. The applied procedure
Modified
(Raphson)Method
can be described Newton
in brief as follows.
In the case of 2-D analysis the structure is assumed to consist of a
finite number of nodes interconnected by a finite number of elements. The types of elements have
been
described
in section 3.to
In calculate
the case of the
3-D tangent
analysis the
structure matrix.
is assumed
consist of the
It may
be expensive
stiffness
Intothe
aforementioned
2-D frames, assuming a rigid
diaphragm
assemblage
theircalculated
horizontal dofs
per floor
Modified Newton-Raphson
iteration
scheme
it is of
only
in the
slab. Loads may be applied at the nodes or along the elements. In both cases though, they are
beginning
of each
transformed
to nodal
loads.new load step
Figure 5. Modified Newton Raphson Method
In the
iteration
schemes
the secant
stiffness
matrix
After
the quasi-Newton
formation of the stiffness
matrix
the equilibrium
equations
are solved
by is
anused
efficient
algorithm
based
on
the
Gaussian
elimination
method.
The
structure
stiffness
is
stored
in
a
banded
form
instead of the tangent matrix
28
Simple Bar Example - Revisited
( t K a + t K b ) u ( i ) =
t +t
R ( t +t Fa ( i 1) t +t Fb ( i 1) )
t +t
u ( i ) = t +t u ( i 1) + u ( i )
with initial conditions
t +t
t +t
u (0) = t u;
Ka =
CA
;
La
Fa (0) = t Fa
Kb =
t +t
Fb (0) = t Fb
CA
Lb
= E if section is elastic
C
= ET if section is plastic
29

Load step 1: t = 1:
( 0 K a + 0 K b )u (1) = 1R 1Fa(0) 1Fb(0)
2 104
= 6.6667 103
1 1
107 ( + )
10 5
Iteration 1: (i = 1)
u (1) =
= 1u (0) + u (1) = 6.6667 103
1 (1)
1 (1)
a
1 (1)
b
1
1 (1)
u
= 6.6667 104 < Y (elastic section!)
La
1 (1)
u
Lb
Fa(1) = 6.6667 103 ;
( K a + K b ) u
0
(2)
Fb(1) = 1.3333 10 4
= R Fa(1) 1Fb(1) = 0
1
Convergence in one iteration!

u = 6.6667 `103
30
Load step 2: t = 2 :
( 1K a + 1K b ) u (1) = 2 R 2 Fa(0) 2 Fb(0)
(4 104 ) (6.6667 103 ) (1.333 104 )

= 6.6667 10 3
1
7 1
10 ( + )
10 5
u (1) =
u (1) = 2u (0) + u (1) = 1.3333 10 2
a(1) = 1.3333 10 3 < Y (elastic section!)
b(1) = 2.6667 103 > Y (plastic section!)
2
1
Fa(1) = 1.3333 10 4 ;
Fb(1) = ( E T ( 2 b(1) Y ) + Y ) A = 2.0067 10 4
( 1K a + 1K b ) u (2) = 2 R 2 Fa(1) 2 Fb(1) u (2) = 2.2 103
31

The procedure is repeated and the results of successive iterations are
tabulated in the accompanying table.
32

Introduction to Nonlinear Analysis

Conclusion from the previous example:
The basic problem in general Nonlinear analysis is to find a state of
equilibrium between externally applied loads and element nodal forces
t
R t F = 0
R =t RB +t RS +t RC
XZ
t (m)T t (m) t
t
B
dV(m)
F=
m
tV
where RB : body forces, RS : surface forces, RC : nodal forces

We must achieve equilibrium for all time steps when
incrementing the loading
Very general approach
Includes implicitly also dynamic analysis!

Basic Types
F

Complex Types
R
U

Bisection Method
Fixed Point Iteration
Write f (x) = 0 in the form f (x) = x q(x),

the solution x satisfies x = q(
x)
Recurrence relation: xk+1 = g (xk )
Assumption: f [a, b] < and continuous
Convergence: If g 0 (x) is defined over [a, b]

If f (a) > 0, f (b) < 0 a x
and a positive constant K exists with
b such that f (
x) = 0
|g 0 (x)| K , x [a, b] then g (x) has a
c=(a+b)/2, check sign of f(c) and so on unique fixed point x [a, b].

Newton (Raphson) Method

xk+1 = xk
f (xk )
f 0 (xk )
terminate when |xk+1 xk | , <<

Convergence: quadratic
|
x xk+1 | C |
x xk+1 |2
Secant Method

x x
xk+1 = xk f (xk ) f (xkk)f k1

(xk1 )
Convergence:
superlinear order
= 1+2 5 (golden ratio)

Newtons method for FE

Assume the general case where K is a nonlinear function of the
displacement U:
K(U)U = F
In order to write it in the form f (x) = 0, we define the residual r(U):
r(U) = K(U)U F r(U) = 0
Then, the Newton iteration formula becomes:
Uk+1 = Uk T1 (Uk )r(Uk )
The slope of the tangent (the tangent stiffness) is
dr(Uk )
dK(U)U F
=
|U=Uk
dU
dU
dK(Uk )
T(Uk ) = K(Uk ) +
Uk
dU
T(Uk ) =
The basic approach in incremental analysis is:
Find a state of equilibrium between externally applied loads and
element nodal forces between successive time steps t
t+t
Assuming that
t+t R
R t+t F = 0
is independent of the deformations we have

t+t
R = tF + F
We know the solution t F at time t and F is the increment in the

nodal point forces corresponding to an increment in the
displacements and stresses from time t to time t + t. This we can
approximate by
F = t KU
Newton-Raphson Method
Assume the tangent stiffness matrix:
t F
t U
We may now substitute the tangent stiffness matrix into the
equilibrium relation
t
K=
KU = t+t R t F
which gives us a scheme for the calculation of the displacements:

t+t
U = tU + U
The exact displacements at time t + t correspond to the applied

loads at t + t, however we only determined these approximately as
we used a tangent stiffness matrix thus we may have to iterate to
find the solution.
We may use the Newton-Raphson iteration scheme to find the
equilibrium within each load increment
t+t
K(i1) U(i) = t+t R t+t F
(i1)
(out of balance load vector)
t+t
(i1)
U(i) = t+t U
+ U(i)
with Initial Conditions

t+t
U(0) = t U;
t+t
K(0) = t K;
t+t (0)
= tF
10
Modified Newton (Raphson)Method

It may be expensive to calculate the tangent stiffness matrix. In the
Modified Newton-Raphson iteration scheme it is only calculated in the
beginning of each new load step
t+Dt
R
Standard NR
Modified NR
t+Dt
t+Dt
R
DU
t
DU
DU
t+Dt
t+Dt
t+Dt
t+Dt
In the quasi-Newton iteration schemes the secant stiffness matrix is used

instead of the tangent matrix
11
Special Considerations
Standard Newton-Raphon methods perform poorly for bucking

problems, where the slope at limit points is exactly equal to 0
12
Special Considerations
The Arc-Length Method for Nonlinear Post-Buckling
Also called Modified Riks Method.
Control the size of the load step using a
parameter .
Solve for both and U in each Newton
iteration.
Assume F = independent of geometry. Then,
can be thought of as a normalized load
parameter and the residual is given by
r(U, ) = K(U)U F
The load increment is computed using
p
= s 2 Un2
where the reference arc length is
s02 =
F
nloadstep
13
14

Load step 1: t = 1:
( 0 K a + 0 K b )u (1) = 1R 1Fa(0) 1Fb(0)
2 104
= 6.6667 103
1 1
107 ( + )
10 5
u (1) =
= 1u (0) + u (1) = 6.6667 103
1 (1)
1 (1)
a
1 (1)
b
1
1 (1)
u
La
1 (1)
u
Lb
Fa(1) = 6.6667 103 ;
( K a + K b ) u
0
(2)
Fb(1) = 1.3333 10 4
= R Fa(1) 1Fb(1) = 0
1
Convergence in one iteration!

u = 6.6667 `103
15
Load step 2: t = 2 :
( 1K a + 1K b ) u (1) = 2 R 2 Fa(0) 2 Fb(0)
(4 104 ) (6.6667 103 ) (1.333 104 )

= 6.6667 10 3
1
7 1
10 ( + )
10 5
u (1) =
u (1) = 2u (0) + u (1) = 1.3333 10 2
a(1) = 1.3333 10 3 < Y (elastic section!)
b(1) = 2.6667 103 > Y (plastic section!)
2
1
Fa(1) = 1.3333 10 4 ;
Fb(1) = ( E T ( 2 b(1) Y ) + Y ) A = 2.0067 10 4
( 1K a + 1K b ) u (2) = 2 R 2 Fa(1) 2 Fb(1) u (2) = 2.2 103
16

The procedure is repeated and the results of successive iterations are
tabulated in the accompanying table.
17
The Continuum Mechanics Incremental Equations

The basic Problem
Establish the solution using an incremental formulation. Two main
approaches exist for establishing equilibrium
Lagrangian Formulation:
Track the movement of all particles of the body (located in a
Cartesian coordinate system), in their motion from the original to the
final configuration (pathline)
Eulerian Formulation:
The motion of the material through a stationary control volume is
considered (streamlines). Mainly used in fluid mechanics.
18
Lagrangian vs. Eulerian Formulation - 1D Example

Spatial or Eulerian coordinates (x): These coordinates are used to locate a
point in space with respect to a fixed basis.
Material or Lagrangian coordinates (X): These coordinates are used to label
material points. If we sit on a material point, the label does not change with time.
Example: Assume that the motion is
x = (X, t) = X(1 + 2t + t 2 )
The inverse of the map gives us X in terms of x, i.e.,
X = 1 (x, t) =
x
(1 + 2t + t 2 )
Then, the displacement of the material point X is

u(X, t) = (X, t) (X, 0) = X(2t + t 2 )
The velocity of the material point is (Langrangian Description)
v(X, t) =
u
= 2X(1 + t)
t
Alternatively we can express the velocity in terms of x (Eulerian Description)
v(X, t) = v(1 (x, t), t) =

1x(1 + t)
(1 + 2t + t 2 )
19
In solids we use the Lagrangian approach as the solution process moves from time
t to t + t iteratively following elements of the body in their motion.
20

The Continuum Mechanics Incremental Equations

The basic Problem
Establish the solution using an incremental formulation. Two main
approaches exist for establishing equilibrium
Lagrangian Formulation: Track the movement of all particles of the

body, in their motion from an initial to a final configuration (pathline)
Eulerian Formulation: The motion of the material through a
stationary control volume is considered (streamlines). Mainly used in
fluid mechanics (but also in large strain plasticity theories - e.g.
generalized plasticity).
Lagrangian vs. Eulerian Formulation - 1D Example

Spatial or Eulerian coordinates (x): These coordinates are used to locate a
point in space with respect to a fixed basis.
Material or Lagrangian coordinates (X): These coordinates are used to label
material points. If we sit on a material point, the label does not change with time.
Example: Assume that the motion is
x = (X, t) = X(1 + 2t + t2 )
The inverse of the map gives us X in terms of x, i.e.,
X = 1 (x, t) =
x
(1 + 2t + t2 )
Then, the displacement of the material point X is

u(X, t) = (X, t) (X, 0) = X(2t + t2 )
The velocity of the material point is (Langrangian Description)
v(X, t) =
u
= 2X(1 + t)
t
Alternatively we can express the velocity in terms of x (Eulerian Description)

v
(X, t) = v(1 (x, t), t) =
2x(1 + t)
(1 + 2t + t2 )
FE Nonlinear Analysis in Solids and Structural Mechanics

How can we evaluate stresses and forces at time t since both the surface and the
volume of the body are unknown ? (In the linear case stiffness and equilibrium
were evaluated based on the initial configuration)
We need to properly map both current strains and stresses to corresponding

measures evaluated at previous configurations
In the further we introduce an appropriate notation:
t
xi = 0 xi + t ui , i =1, 2, 3
t+t
xi = 0 xi + t+t ui
Increments in displacements from time t to t + t are related as:

ui = t+t ui t ui , i =1, 2, 3
Reference configurations are indexed as e.g.
s
0 fi
t+t
where the left subscript indicates the reference configuration and the left
superscript indicates at which configuration the quantity occurs.
Note if those quantities are the same the left subscript maybe omitted e.g:
t+t
t+t
ij
Differentiation is indexed as:

t+t
0 ui,j =
t+t
0 ui
0
xj
The deformation gradient, strain and stress tensors

As mentioned we must try to establish a description of the volume we
consider such that we can express the internal virtual work in terms of
an integral over a volume we know!
Further we would like to be able to decompose the stresses and strains
in an efficient manner keeping track of how the volume stretches and
rotates (rigidly).
We consider a body under deformation at times 0 and t:
The deformation gradient

We now consider the change of an infinitesimal gradient vector
Definition
The deformation gradient maps d0 x onto dt x through the following relation
dt x =
t
0
0 Xd x

We can write the deformation gradient as the Jacobian of the
current configuration at time t, with respect to the initial
configuration at time 0.
t
X
=
t x1
0 x1
t x2
0 x1
t x3
0 x1
t x1
0 x2
t x2
0 x2
t x3
0 x2
t x1
0 x3
t x2
0 x3
t x3
0 x3
The deformation gradient describes

the stretches and rotations that the
material fibers have undergone from
time zero to time t
t
0X
= (0 t xT )T ,
where
It can be shown that
t
0X
0 =
0 x1
0 x2
0 x3
and
t T
x =
x1
x2
x3
= ( 0t X)1

Then we introduce the Cauchy-Green deformation tensor
The deformation gradient is also used to measure the stretch of a material
fiber and the change in angle between fibers due to the deformation. For
this we use the
t
0C
6=
t
0
t T t
0 X 0 X:
B=
Right Cauchy-Green deformation tensor
t T
t
0X 0X :
Left Cauchy-Green deformation tensor
The deformation gradient can be decomposed into a unique product of two

matrices
t
0X
t
0 U:
Symmetric stretch matrix
t
0 R:
Orthogonal rotation matrix
t
t
0R 0U
This is referred to as a Polar Decomposition


Decomposition of the deformation gradient
We continue by rewriting the deformation gradient
X = RU = RURT R = VR
where U: the right stretch matrix and V: the left stretch matrix
Further it can be shown that:
U = RL RT
L
where : the principal stretches and RL : the Direction of principal
stretches
V = RE RT
E
where RE : the Base vectors of principal stretches in the stationary
coordinate system
10
Page 22
Th deformation
The
d f
ti
gradient,
di t strain
t i and
d stress
t
tensors
t
Consider a bar under stretch and rotation

We consider a bar under stretch and rotation
t
X = RU
x1
WeDecomposition
continue by rewriting
(Ex 6.8) the
deformation
gradient
It is instructive to consider the
deformation in two steps
2L
Stretching
2 0
Method
of Finite Elements II
U=
0 1 0
0 0 1
X = RU Decomposition
2 0 0
Stretching U = 0 1 0
consider the
It is simpler
to
0 0 1in
two steps
deformation
0
x1
0 1 0
Rotation
R = 1 0 0
Rotation
0 0 1
cos
R=
sin
sin
cos
0
0
= 1
1
0
1
0
0
11

General Case
Assuming both rotation U and stretch R: X = RU
iss Federal Institute of Technology

0
0
cos sin 0
U = 0 h 0 R = sin cos 0
deformation
d f
ti gradient,
di
t strain
t i and
t 0 1tensors
t
H
0d stress
0 0 1
which yields
ample beam element
x2 , t x2
h
l
h
L cos H sin 0
l
h
RU X =
sin
cos
0
L
H
0
0
1
Page 2
H
0
x1 , t x1
L
12

Using the decomposition of the deformation gradient we may rewrite
the right and left Cauchy-Green deformation tensors:
The right Cauchy-Green deformation tensor:
C = XT X = (RU)T RU = UT RT RU = U2
since R is orthogonal, hence RRT = RT R = I

The left Cauchy-Green deformation tensor:
B = XXT = VRT RV = V2
13
The strain tensor

From deformations to strains:
l
The strain may be understood in terms of the stretch = .
L
In a previous lecture we saw that the one dimensional equivalents where:
Using time 0 as reference

1 l2
1 2
Green - Lagrange strain: E =
1
1 =
2 L2
2
1
Tensor Equivalent: t+t
0 = (C I)
2
Using time t as reference

1
1
L2
Almansi strain: A =
1 2 =
1 2
2
l
2
1
t+t
1
Tensor Equivalent:
)
t = (I B
2
14

In terms of tensor components,
Green - Lagrange strains:
1
ij =
2
X uk uk
ui
uj
+ 0 +
0
xj
xi
0 xi 0 xj
k=1
Almansi strains:
1
ij =
2
X uk uk
ui
uj
+ t +
t
xj
xi
t xi t xj
k=1
15
The stress tensors

Finally we need to establish the stresses
We start by introducing the Cauchy stresses:
The Cauchy stress tensor relates forces at the current configuration

to areas at the current configuration
16
The stress tensors

Since the Cauchy tensor is not known a-priori, forces and areas are
mapped through the deformation gradient to the reference
configuration.
Definition
The Second Piola-Kirchoff stress tensor
0
t
0S
0 t 0 T
X tX
t t
0
where t is the mass density of the body at time t and
= det(t0 X)
These are so-called work conjugate to the Green - Lagrange strains

The mapping retains the symmetry of the Cauchy tensor
Rigid body motions (translations/rotations) do not induce
strains/stresses
The components of the Piola-Kirchoff stress tensor have little physical
meaning and in practice, Cauchy stresses must be calculated
17
Total and Updated Lagrangian Formulation

Remember, a configuration C is a snapshot of the set of motions of
all particles
Definition-Initial Configuration
The configuration defined as the origin of Undeformed displacements.
Strain free but not necessarily stress free
Definition-Reference Configuration
Configuration to which stepping computations in an incremental solution
process are referred
Definition-Target Configuration
Equilibrium configuration accepted after completing the an increment step
18
19
Total Lagrangian (TL)

The initial configuration is
the reference configuration.
Stress & Strain measures at
the target configuration
t + t are computed with
respect to the initial
configuration at time 0.
Derivatives and Integrals are
taken with respect to 0 V
(initial conf.)
Updated Lagrangian (UL)

The previous configuration is
the reference configuration.
Stress & Strain measures at
the target configuration
t + t are evaluated with
respect to the configuration
at time t.
Derivatives and Integrals are
taken with respect to t V
20

We originally set out to solve the following equation:
t+t
t+t eij dt+t V = t+t R
t+t V
Two schemes have been formulated for this namely:

The Total Lagrangian (TL) formulation
Z
t+t
t+t
0
t+t
R
0 S 0 ij d V =
0V
The Updated Lagrangian (UL) formulation

Z
t+t
t+t
t
t+t
R
t S t ij d V =
tV
21
Setting up the governing equations

Therefore for the Total Lagrangian (TL) Formulation we obtain :
Z
t+t
t+t
0
t+t
R
(1)
0 Sij 0 ij d V =
0V
where the Green-Lagrange strain component has been defined as:

!
3
X
1 t+t
t+t
t+t
t+t
t+t
0 ui,j + 0 uj,i +
0 uk,i
0 uk,j
0 ij =
2
k=1
If we denote 0 ij , 0 ui as increments in the strains and displacements

respectively we can write:
t+t0 ij = 0t ij +0 ij
(2)
22

The incremental strain in eq (2) can be further decomposed to
0
ij =0 eij +0 ij
(3)
where the linear incremental strains are:

1
0 eij =
2
ui,j +0 uj,i +
3
X
!
t
0
uk,i 0 uk,j + 0 uk,i t0 uk,j
k=1
where the nonlinear incremental strains are:

3
1X
0 ij =
0 uk,i 0 uk,j
2
k=1
Furthermore, stresses can be written incrementally as:

t+t0 Sij = 0t Sij +0 Sij
(4)
23
Noting that the variation 0t+t ij at configuration time t+t is only
dependent on the variation of the increment, hence
t+t0 ij = 0 ij
and plugging eqns (3),(4) into eqn(2) we obtain the Equation of
Motion with Incremental Decompositions
Z
Z
Z
t
0
t
0
t+t
0
R
0 Sij 0 ij d V +
0 Sij 0 eij d V
0 Sij 0 ij d V =
0V
0V
0V
24
Remarks
R
Term 0 V t0 Sij 0 eij d0 V is known, so it can be moved to the
right hand side of the equation of motion.
R
Term 0 V t0 Sij 0 ij d0 V is linear with respect to the
incremental displacements ui .
R
Term 0 V 0 Sij 0 ij d0 V is highly nonlinear but we can use
Taylor expansion to approximate it
25

Since 0 ij is of higher order we can approximate:
ij = 0 eij + 0 ij = 0 eij
Using the above and the Taylor approximation of the highly

nonlinear term ultimately yields
Z
0
0V
where 0 Cijrs
Sij 0 ij d V =
0
0V
Cijrs ers 0 eij d0 V

t0 Sij
= t

0 rs
26
Finally, the Linearized Equation of Motion for the (TL) formulation is:
Z
0
0 Cijrs ers 0 eij d V +
0V
t
0
0V
Sij 0 ij d0 V = t+t R
t
0
0V
Sij 0 eij d0 V
Similarly, for the (UL) formulation we get:

Z
tV
t
t Cijrs ers t eij d V +
ij t ij dt V = t+t R
tV
ij t eij dt V
tV
where t0 Sij , t ij are the Piola-Kirchhoff and Cauchy stresses at time t
27
Total Lagrangian Formulation
28
Updated Lagrangian Formulation
In practice, it is often sufficient to account for only material non-linearity. In this case
the TL and the UL formulations become identical.
29
Element Matrices
The general matrix equations vector depend on the assumed type of
analysis and the approach
A) Material nonlinearity
Static Analysis
t
KU = t+t R t F
Dynamic Analysis, Implicit Integration scheme

+ t KU = t+t R t F
Mt+t U
Dynamic Analysis, Explicit Integration scheme

= tR tF
Mt U
30
Element Matrices
B) TL Formulation
Static Analysis
(t0 KL +t0 KN L )U = t+t R t0 F

+ (t KL +t KN L )U = t+t R t F
Mt+t U
0
0
0

= tR t F
Mt U
0
31
Element Matrices
C) UL Formulation
Static Analysis
(tt KL +tt KN L )U = t+t R tt F

Mt+t U
t
t
t

= tR tF
Mt U
t
32
Element Matrices
where the following notation has been used

M
t
=
t
0
t
t
K, KL , KL =
t
0
t
t
KN L , KN L =
t+t
t
0
time independent Mass matrix

linear strain incremental Stiffness matrices
nonlinear strain incremental Stiffness matrices
vector of externally applied nodal point loads at time t + t
F, F, F
vector of nodal point forces equivalent to the element stresses
vector of increments in nodal point displacements
t
t
33
Element Matrices
34
Element Matrices
HS , H
the shape function matrices (usually denoted by N)
0 S
vector of surface and body forces at time 0
f , 0f B
BL , t0 BL , tt BL = linear strain displacement transformation matrices

t
0
BN L , tt BN L
nonlinear strain displacement transformation matrices
C, t C
incremental stress strain material property matrices
matrix and vector of Cauchy stresses
matrix and vector of 2nd Piola-Kirchhoff stresses
S, t S
t
0
These matrices depend on the type of Finite Element considered
35
Structural Elements
Truss and Cable Elements
A Truss Element is a structural element capable of transmitting
stresses only in the direction normal to the cross-sectional area
Consider a truss element

that has an arbitrary
orientation. It is usually
described by two to four
nodes and is subjected to
large displacements and
large strains.
36
Truss and Cable (Bar) Elements
The nodal point coordinates determine the spatial configuration of

the bar at time 0 and t using:
0
x1 (r) =
n
X
Nk xk1
0
x2 (r) =
and
x1 (r) =
Nk xk2
0
x3 (r) =
Nk t xk1
x2 (r) =
k=1
n
X
Nk t xk2
n
X
Nk 0 xk3
k=1
k=1
k=1
n
X
n
X
x3 (r) =
k=1
n
X
Nk t xk3
k=1
37
Shape Functions
n = 2 nodes at r1 = 1, r2 = 1
1
1
(1 r), N2a = (1 + r)
2
2
n = 3 nodes at r1 = 1, r2 = 1, r3 = 0
1
1
N1b = N1a (1 r2 ), N2b = N2a (1 r2 ), N3b = (1 r2 )
2
2
1
1
n = 4 nodes at r1 = 1, r2 = 1, r3 = , r4 =
3
3
1
1
(9r3 + r2 + 9r 1), N2c = N2b +
(9r3 + r2 9r 1),
N1c = N1b +
16
16
1
1
N3c = N3b +
(27r3 + 7r2 27r 7), N4c =
(27r3 9r2 + 27r + 9)
16
16
N1a =
38

Isoparametric Elements t ui (r) =
Pn
t k
k=1 Nk ui
i=1,2,3
Element Matrices
Since the only stress is the normal stress we consider only the
corresponding longitudinal strain along s: t0 11 .
From the TL Formulation we obtain:
t
0
11 =
3
X
d0 xi dt ui
i=1
d0 s d0 s
1 dt ui dt ui
2 d0 s d0 s
Incremental Decomposition
11 =
e
3
X
d0 xi dui
d0 s d0 s
i=1
dt ui dui
d0 s d0 s
11 =
0
1 X dui dui
2
d0 s d0 s
i=1
39

Matrix Equivalent Formulation
Strain Displacement Matrices
t
0
BL = (0 J 1 )2 (0 xT NT,r N,r + t uT NT,r N,r )
t
0
BN L = (0 J 1 )N,r :independent of orientation
dr
and 0 s(r) is the arc length (coordinate) at point
d0 s
(x1 (r), x2 (r), x3 (r)) given by
where 0 J 1 =
s(r) =
n
X
Nk 0 sk
k=1
and the tangent Stress

The only non zero stress component is t0 S
11
Strain relationship therefore is:
C1111 =
t0 S
11
t0 11
40

Example
Develop the tangent stiffness matrix and force vector at time t.
Consider large displacement and large strain conditions.
41
Truss and Cable Elements - Example

Since the element is aligned with the 0 x1 axis at time 0 we need not introduce
the arc length s in our calculations. From the relevant TL formulation table we
have that the linear and nonlinear components of the Green- Lagrange strain
increments will be given as:
0 eij
0 ij
1
=
2
1
=
2
0 ui,j
+ 0 uj,i +
3
X
!
(t0 uk,i 0 uk,j
t
0 uk,i 0 uk,j )
k=1
3
X
!
0 uk,i 0 uk,j
k=1
Since we are interested in 0 e11 , 0 11 , and since the displacements are restricted in
the 0 x1 , 0 x2 plane the above expressions become
0 e11
u1
t u1 u1
t u2 u2
= 0 + 0
+ 0
0
x1
x1 x1
x1 0 x1
0 11
1
=
2
"
u1
0 x1
2

+
u2
0 x1
2 #
(5)
42

From the Geometry the nodal coordinates at time t are:
t 1
u1
t 1
u2
= 0,
= 0,
t 2
u1
= (0 L + L)cos 0 L
t 2
u2
= (0 L + L)sin
The displacement at a point within the element (at a distance from the center)
is given by
ui =
2
X
Nk t uki
with N1 =
k=1
Also, 0 J =
1
1
(1 ), N2 = (1 + )
2
2
0
0 x2
L
0 x1
=
=
. Then we obtain
N1 () t 1 N2 () t 2
t u1
=
u1 +
u1
0 x1
0 x1
0 x1
h
i 1
(0 L + L)cos
t u1
1
= 0 + (0 L + L)cos 0 L 0 J =
0L
0
x1
0
t
( L + L)sin
u2
Similarly,
=
0L
0 x1
(6)
(7)
(8)
43

In addition, the incremental displacements are also written as:
ui =
2
X
Nk uki
with N1 =
k=1
which yields:
1
1
(1 ), N2 = (1 + )
2
2
u1
u2
1
1
= 0 (u21 u11 ), 0
= 0 (u22 u12 )
L
L
0 x1
x1
Substituting expressions (6) - (9) into equation (5) we obtain the

representation:
"
0

1
( L + L)cos
1 0 1 0 +
1
0 e11 = 0
0L
L
1
u1
#
0

u12
( L + L)sin
0
1
0
1
+
2
0L
u1
u22
1
u1
u12
(0 L + L)
cos sin cos sin
0 e11 =
u2
(0 L)2
1
u22
(9)
matrix equivalent
44

Hence,
t
0 BL
(0 L + L)
cos
(0 L)2
sin
cos
sin
The same could be obtained by directly applying the formula:

t
0 BL
= (0 J
1 2 0 T
t T T
) ( x NT
, N, + u N, N, )
Next, the linear part of the stiffness matrix is then obtained as:
t
0 KL
Z
=
0V
T
t
t
0
0 BL 0 C0 BL d V
t0 S11
. If we use that the original ratio is equal to the
t0 11
elasticity modulus, we have 0 C = E. Also, 0 V = 0 A0 L. Then,
where for the truss element 0 C =
t
0 KL
= 0A0 C
(0 L
L)2
+
(0 L)3
cos2
cossin
sin2
cos2
sincos
cos2
Symm
cossin
2
sin
sincos
2
sin
45

In order to derive the nonlinear part of the stiffness matrix we first need to evaluate the
tP
Piola-Kirchhoff stress. We know that the Cauchy stress at time t is equal to t = t ,
A
directed along the axis of the element at time t. Using the rotational matrix we can
rotate the stress tensor from the element axis system to the original reference system
(t x1 , t x2 ). Denoting the rotated tensor as t we have:
t

= R
0
0
RT ,

R=
sin
cos
cos
sin
Then, from Lecture 4 we know that the Piola-Kirchhoff stress is given as:
t
0S
0
t
0
t 0 T
tX
tX

or
0
0
RT =
t
T
t
X t0 S t0 X
0 0
(10)
where the deformation gradient, t0 X, can be obtained as the product of a rotational and
a stretch component as t0 X = RU. For this example the stretch matrix is obviously
(elongation along x):
U=
0L
+ L
0L
0
0L
0 U1 =
1
0L
+ L
0
0
1
46

Then equation (10) can be rewritten as:

t

t
0
RT = 0 RU t0 S RUT R
R
0
0
t

t

t
S 0
0
t
=
= 0 U 0 S UT 0
0
0
0
0
0
0
U1
0
0
t
RT =
t
t
0
0
0
RU t0 S UT RT
(U1 )T
Ultimately, carrying out the calculations yields:

t
0 S11
Since mass=const
0
t
(0
0L
tP
L + L
)2 t
0 0
L0 A = t 0 L + Lt A t0 S11 =
0L
0L
tP
+ L
0A
Note how the components of t0 S do not depend on rotation, hence the tensor retains only
the S11 one component. Then, the nonlinear part of the stiffness matrix is derived as:
Z
T
t
t
t
0
K
=
N
L
0
0 BN L 0 S0 BN L d V
0V
where t0 BN L = (0 J 1 ) N, =
1
0
0
1
1
0
0
1
47
Finally, t0 KN L
1
0
= 0
L + L 1
0
tP
0
1
0
1
1
0
1
0
0
1
0
1
and, t0 KN = t0 KL + t0 KN L
and the force vector is obtained as (see relevant table for (TL) formulation):
t
0F
Z
=
0V
Tt
t
0
0 BL 0 S11 d V
(0 L + L)
(0 L)2
cos
0L
tP
sin
0
cos 0 L + L 0 A V
sin
cos
sin
t
t
0F = P
cos
sin
48


From the last Lecture we obtained the Total Lagrangian (TL)
Formulation of the principle of Virtual Displacements:
Z
t+t
t+t
0
t+t
R
0 Sij 0 ij d V =
(1)
0V
where the Green-Lagrange strain component has been defined as:

!
3
X
1 t+t
t+t
t+t
t+t
t+t
0 ij =
0 ui,j + 0 uj,i +
0 uk,i
0 uk,j
2
k=1
If we denote 0 ij , 0 ui as increments in the strains and displacements

respectively we can write:
t+t0 ij = 0t ij +0 ij
(2)

The incremental strain in eq (2) can be further decomposed to
0
ij =0 eij +0 ij
(3)
where the linear incremental strains are:

1
0 eij =
2
ui,j +0 uj,i +
3
X
!
t
0
uk,i 0 uk,j + 0 uk,i t0 uk,j
k=1
where the nonlinear incremental strains are:

3
1X
0 ij =
0 uk,i 0 uk,j
2
k=1
Furthermore, stresses can be written incrementally as:

t+t0 Sij = 0t Sij +0 Sij
(4)
3
Noting that the variation 0t+t ij at configuration time t+t is only
dependent on the variation of the increment, hence
t+t0 ij = 0 ij
and plugging eqns (3),(4) into eqn(2) we obtain the Equation of
Motion with Incremental Decompositions
Z
Z
Z
t
0
t
0
t+t
0
R
0 Sij 0 ij d V +
0 Sij 0 eij d V
0 Sij 0 ij d V =
0V
0V
0V
Remarks
R
Term 0 V t0 Sij 0 eij d0 V is known, so it can be moved to the
right hand side of the equation of motion.
R
Term 0 V t0 Sij 0 ij d0 V is linear with respect to the
incremental displacements ui .
R
Term 0 V 0 Sij 0 ij d0 V is highly nonlinear but we can use
Taylor expansion to approximate it

Since 0 ij is of higher order we can approximate:
ij = 0 eij + 0 ij = 0 eij
Using the above and the Taylor approximation of the highly

nonlinear term ultimately yields
Z
0
0V
where 0 Cijrs
Sij 0 ij d V =
0
0V
Cijrs ers 0 eij d0 V

t0 Sij
= t

0 rs
Finally, the Linearized Equation of Motion for the (TL) formulation is:
Z
0
0 Cijrs ers 0 eij d V +
0V
t
0
0V
Sij 0 ij d0 V = t+t R
t
0
0V
Sij 0 eij d0 V
Similarly, for the (UL) formulation we get:

Z
tV
t
t Cijrs ers t eij d V +
ij t ij dt V = t+t R
tV
ij t eij dt V
tV
where t0 Sij , t ij are the Piola-Kirchhoff and Cauchy stresses at time t
Element Matrices
The general matrix equations vector depend on the assumed type of
analysis and the approach
A) Material nonlinearity
Static Analysis
t
KU = t+t R t F

+ t KU = t+t R t F
Mt+t U

= tR tF
Mt U
Element Matrices
B) TL Formulation
Static Analysis
(t0 KL +t0 KN L )U = t+t R t0 F

Mt+t U
0
0
0

= tR t F
Mt U
0
Element Matrices
C) UL Formulation
Static Analysis
(tt KL +tt KN L )U = t+t R tt F

Mt+t U
t
t
t

= tR tF
Mt U
t
10
Element Matrices

M
t
=
t
0
t
t
K, KL , KL =
t
0
t
t
KN L , KN L =
t+t
t
0
time independent Mass matrix

linear strain incremental Stiffness matrices
nonlinear strain incremental Stiffness matrices
vector of externally applied nodal point loads at time t + t
F, F, F
vector of nodal point forces equivalent to the element stresses
vector of increments in nodal point displacements
t
t
11
Element Matrices
12
Element Matrices
HS , H
the shape function matrices (usually denoted by N)
0 S
vector of surface and body forces at time 0
f , 0f B
BL , t0 BL , tt BL = linear strain displacement transformation matrices

t
0
BN L , tt BN L
nonlinear strain displacement transformation matrices
C, t C
incremental stress strain material property matrices
matrix and vector of Cauchy stresses
matrix and vector of 2nd Piola-Kirchhoff stresses
S, t S
t
0
These matrices depend on the type of Finite Element considered
13
Structural Elements
A Truss Element is a structural element capable of transmitting
stresses only in the direction normal to the cross-sectional area
Consider a truss element

that has an arbitrary
orientation. It is usually
described by two to four
nodes and is subjected to
large displacements and
large strains.
14

The nodal point coordinates determine the spatial configuration of the bar
at time 0 and t using:
0
x1 (r) =
n
X
Nk 0 xk1
x2 (r) =
k=1
and
x1 (r) =
n
X
n
X
Nk 0 xk2
x3 (r) =
k=1
Nk t xk1
x2 (r) =
k=1
n
X
Nk t xk2
n
X
Nk 0 xk3
k=1
t
x3 (r) =
k=1
n
X
Nk t xk3
k=1
Shape Functions
n = 2 nodes at r1 = 1, r2 = 1
1
1
(1 r), N2a = (1 + r)
2
2
n = 3 nodes at r1 = 1, r2 = 1, r3 = 0
1
1
N1b = N1a (1 r2 ), N2b = N2a (1 r2 ), N3b = (1 r2 )
2
2
1
1
n = 4 nodes at r1 = 1, r2 = 1, r3 = , r4 =
3
3
1
1
N1c = N1b +
(9r3 + r2 + 9r 1), N2c = N2b +
(9r3 + r2 9r 1),
16
16
1
1
N3c = N3b +
(27r3 + 7r2 27r 7), N4c =
(27r3 9r2 + 27r + 9)
16
16
N1a =
15

Isoparametric Elements t ui (r) =
Pn
t k
k=1 Nk ui
i=1,2,3
Element Matrices
Since the only stress is the normal stress we consider only the
corresponding longitudinal strain along s: t0 11 .
From the TL Formulation we obtain:
t
0
11 =
3
X
d0 xi dt ui
i=1
d0 s d0 s
1 dt ui dt ui
2 d0 s d0 s
Incremental Decomposition
11 =
e
3
X
d0 xi dui
d0 s d0 s
i=1
dt ui dui
d0 s d0 s
11 =
0
1 X dui dui
2
d0 s d0 s
i=1
16

Matrix Equivalent Formulation
Strain Displacement Matrices
t
0
BL = (0 J 1 )2 (0 xT NT,r N,r + t uT NT,r N,r )
t
0
BN L = (0 J 1 )N,r :independent of orientation
dr
and 0 s(r) is the arc length (coordinate) at point
d0 s
(x1 (r), x2 (r), x3 (r)) given by
where 0 J 1 =
s(r) =
n
X
Nk 0 s k
k=1
and the tangent Stress

The only non zero stress component is t0 S
11
Strain relationship therefore is:
C1111 =
t0 S
11
t0 11
17

Example
Develop the tangent stiffness matrix and force vector at time t.
Consider large displacement and large strain conditions.
18

Since the element is aligned with the 0 x1 axis at time 0 we need not introduce
the arc length s in our calculations. From the relevant TL formulation table we
have that the linear and nonlinear components of the Green- Lagrange strain
increments will be given as:
0 eij
0 ij
1
=
2
1
=
2
0 ui,j
+ 0 uj,i +
3
X
!
(t0 uk,i 0 uk,j
t
0 uk,i 0 uk,j )
k=1
3
X
!
0 uk,i 0 uk,j
k=1
Since we are interested in 0 e11 , 0 11 , and since the displacements are restricted in
the 0 x1 , 0 x2 plane the above expressions become
0 e11
u1
t u1 u1
t u2 u2
= 0 + 0
+ 0
0
x1
x1 x1
x1 0 x1
0 11
1
=
2
"
u1
0 x1
2

+
u2
0 x1
2 #
(5)
19

From the Geometry the nodal coordinates at time t are:
t 1
u1
t 1
u2
= 0,
= 0,
t 2
u1
= (0 L + L)cos 0 L
t 2
u2
= (0 L + L)sin
The displacement at a point within the element (at a distance from the center)
is given by
ui =
2
X
Nk t uki
with N1 =
k=1
Also, 0 J =
1
1
(1 ), N2 = (1 + )
2
2
0
0 x2
L
0 x1
=
=
. Then we obtain
N1 () t 1 N2 () t 2
t u1
=
u1 +
u1
0 x1
0 x1
0 x1
h
i 1
(0 L + L)cos
t u1
1
= 0 + (0 L + L)cos 0 L 0 J =
0L
0
x1
0
t
( L + L)sin
u2
Similarly,
=
0L
0 x1
(6)
(7)
(8)
20

In addition, the incremental displacements are also written as:
ui =
2
X
Nk uki
with N1 =
k=1
which yields:
1
1
(1 ), N2 = (1 + )
2
2
u1
u2
1
1
= 0 (u21 u11 ), 0
= 0 (u22 u12 )
L
L
0 x1
x1
Substituting expressions (6) - (9) into equation (5) we obtain the

representation:
"
0

1
( L + L)cos
1 0 1 0 +
1
0 e11 = 0
0L
L
1
u1
#
0

u12
( L + L)sin
0
1
0
1
+
2
0L
u1
u22
1
u1
u12
(0 L + L)
cos sin cos sin
0 e11 =
u2
(0 L)2
1
u22
(9)
matrix equivalent
21

Hence,
t
0 BL
(0 L + L)
cos
(0 L)2
sin
cos
sin
The same could be obtained by directly applying the formula:

t
0 BL
= (0 J
1 2 0 T
t T T
) ( x NT
, N, + u N, N, )
Next, the linear part of the stiffness matrix is then obtained as:
t
0 KL
Z
=
0V
T
t
t
0
0 BL 0 C0 BL d V
t0 S11
. If we use that the original ratio is equal to the
t0 11
elasticity modulus, we have 0 C = E. Also, 0 V = 0 A0 L. Then,
where for the truss element 0 C =
t
0 KL
= 0A0 C
(0 L
L)2
+
(0 L)3
cos2
cossin
sin2
cos2
sincos
cos2
Symm
cossin
2
sin
sincos
2
sin
22

In order to derive the nonlinear part of the stiffness matrix we first need to evaluate the
tP
Piola-Kirchhoff stress. We know that the Cauchy stress at time t is equal to t = t ,
A
directed along the axis of the element at time t. Using the rotational matrix we can
rotate the stress tensor from the element axis system to the original reference system
(t x1 , t x2 ). Denoting the rotated tensor as t we have:
t

= R
0
0
RT ,

R=
sin
cos
cos
sin
Then, from Lecture 4 we know that the Piola-Kirchhoff stress is given as:
t
0S
0
t
0
t 0 T
tX
tX

or
0
0
RT =
t
T
t
X t0 S t0 X
0 0
(10)
where the deformation gradient, t0 X, can be obtained as the product of a rotational and
a stretch component as t0 X = RU. For this example the stretch matrix is obviously
(elongation along x):
U=
0L
+ L
0L
0
0L
0 U1 =
1
0L
+ L
0
0
1
23

Then equation (10) can be rewritten as:

t

t
0
RT = 0 RU t0 S RUT R
R
0
0
t

t

t
S 0
0
t
=
= 0 U 0 S UT 0
0
0
0
0
0
0
U1
0
0
t
RT =
t
t
0
0
0
RU t0 S UT RT
(U1 )T
Ultimately, carrying out the calculations yields:

t
0 S11
Since mass=const
0
t
(0
0L
tP
L + L
)2 t
0 0
L0 A = t 0 L + Lt A t0 S11 =
0L
0L
tP
+ L
0A
Note how the components of t0 S do not depend on rotation, hence the tensor retains only
the S11 one component. Then, the nonlinear part of the stiffness matrix is derived as:
Z
T
t
t
t
0
K
=
N
L
0
0 BN L 0 S0 BN L d V
0V
where t0 BN L = (0 J 1 ) N, =
1
0
0
1
1
0
0
1
24
Finally, t0 KN L
1
0
= 0
L + L 1
0
tP
0
1
0
1
1
0
1
0
0
1
0
1
and, t0 KN = t0 KL + t0 KN L
and the force vector is obtained as (see relevant table for (TL) formulation):
t
0F
Z
=
0V
Tt
t
0
0 BL 0 S11 d V
(0 L + L)
(0 L)2
cos
0L
tP
sin
0
cos 0 L + L 0 A V
sin
cos
sin
t
t
0F = P
cos
sin
25

2D Axisymmetric, Plane Strain and Plane Stress Elements
The nodal point coordinates determine the spatial configuration of the

element at time 0 and t using:
0
x1 (r) =
n
X
Nk 0 xk1
x2 (r) =
k=1
and
x1 (r) =
n
X
Nk 0 xk2
k=1
Nk t xk1
x2 (r) =
k=1
t
n
X
Isoparametric Elements ui (r) =
n
X
Nk t xk2
k=1
Pn
k
k=1 Nk ui ,
t
i=1,2

Shape Functions
n = 4 nodes
1
1
1
N1a = (1 + r)(1 + s), N2a = (1 r)(1 + s), N3a = (1 r)(1 s),
4
4
4
1
N4a = (1 + r)(1 s)
4
n = 8 nodes
1
1
1
N5b = (1 r2 )(1 + s), N6b = (1 s2 )(1 r), N7b = (1 r2 )(1 s),
2
2
2
1
N8b = (1 s2 )(1 + r)
2
1
1
1
1
1
1
N1b = N1a N5b N8b , N2b = N2a N6b N7b , N3b = N3a N6b N7b ,
2
2
2
2
2
2
1
1
N4b = N4a N7b N8b
2
2
n = 9 nodes
N9c = (1 r2 )(1 s2 ),
1
Nic = Nib N9c , i=1,2,3,4
4
1
Njc = Njb N9c , i=5,6,7,8
2
Example
Establish the Total Lagrangian Strain - Displacement matrices for
the following element (assuming large displacements/large strains):
Example
tu1
The nodal coordinates at time

t are:
u11
t 2
u1
t 3
u1
t 4
u1
=1
=0
=0
=1
u12
t 2
u2
t 3
u2
t 4
u2
2=0.5
tu1
1=1
= 0.5
= 0.5
=0
=0
The Jacobian for a 2D system is defined as:

Example6.18
t

x1 t x2
3
r
0
r
2
J = tx
t x2 =
1
0
s
s
0
3
2
Example - Linear Strain Displacement Matrix component t0 BL0

Using the 4 node 2D element shape functions and the relevant
matrix we obtain:
Example - Nonlinear Strain Displacement Matrix component t0 BL1

We first need to calculate the following terms:
10

Example - Linear Strain Displacement Matrix component t0 BL1
11

Example - Nonlinear Strain Displacement Matrix component t0 BN L
Using the 4 node 2D element shape functions and the relevant matrix we
obtain:
12
Beams/Plates/Shell elements .vs. Solids elements
*The section on Beams/Plates/Shell elements is taken from Prof.

H. Waismans notes of the FEM II course - CEEM Department,
Columbia University
F-16 Aeroelastic Structural Model
Exterior
model
95% are
shell
elements
FEM model:
150000 Nodes
Internal structure
zoom. Some Brick
and tetrahedral
elements
http://www.colorado.edu/engineering/CAS/Felippa.d/FelippaHome.d/Home.html
13
Beams/Plates/Shell elements .vs. Solids elements

FE Model of a car
Car body
(shells and
beams)
http://www.itcsoftware.com/general_finite-eg-automotive.htm
14
Beam Elements
Two beam theories:
Euler-Bernoulli theory (Engineering beam theory) -slender
beams
Timoshenko theory thick beams
Euler - Bernoulli Beam
15
Beam Elements
Euler Bernoulli Beam Assumptions - Kirchhoff Assumptions
Normals remain straight (they do not bend)
Normals remain unstretched (they keep the same length)
Normals remain normal (they always make a right angle to the
neutral plane)
16
Beam Elements - Strong Form

The Euler-Bernoulli Beam theory (small deformations)
- normal stress
M - bending moment
- normal strain
v - displacement of the centroid
EI - bending stiffness
17

Equilibrium
distributed load per unit length
shear force
Combining the equations
18

(1)
Free end with applied load
(2)
(S)
(3)
Simple support
(4)
Clamped support
(5)
19
Beam Elements - Strong Form to Weak Form

Multiply
(4)w(5)
byintegrate
w and integrate
the domain
Multiply Eqns.
(1), Eq.
(4) (1),
(5) by
and
over theover
domain
First integration by parts

by parts
First integration
First integration
First integration
by parts by parts
Second integration
by by
parts
gives
Second integration
parts gives
Second integration by parts gives
Second integration by parts gives
20
Beam Elements - Strong Form to Weak Form

Arrive at
Arrive
at the
the weak
weak form
form
(W)
Note:
Note:1. The spaces are C1 continuous, i.e. the derivative must also be
continuous
1. The
spaces are C 1 continuous, i.e. the derivative must also be
continuous
2. The left side is symmetric in w and v (bi-linear form: a(v,w)=a(w,v)
2. The
left
is to
symmetric
instiffness
w and vmatrix
(bi-linear form:
this
willside
lead
symmetric
a(v,w)=a(w,v)) this will lead to symmetric Stiffness Matrix
21
Beam Elements - FE Formulation
Physical domain
Natural domain
Element
displacement
vector
Element
force
vector
22

Hermite Polynomials
Polynomials
Hermite
23

The displacement is approximated by
However note :
From coordinate transformation (mapping)
24
Beam Elements - Galerkin

Finally, the weight functions and trial solutions are
Finally, the weight functions and trial solutions
Where the shape functions are
25
Beam Elements - FE Matrices
From the weak form, we had

From the weak form, we had
The
second
derivative
The
second
derivative
becomesbecomes
and
26
Beam Elements - FE Matrices

Stiffness matrix
Force vector
Assuming constant distributed force
27
Beam Elements - Example

Consider a clamped-free
Consider a clamped-free beam

with EI = 104 N m2
s = 20N
m = 20N m
Pre-processing
28
For element (1)
[1]
For element (2)
[2]
[3]
[4]
[3]
[4]
[5]
[6]
Assembly into a Global Stiffness Matrix
29

Boundary force matrix
Element (1) has no boundary on S or M

[3]
For element (2) we have
[4]
[5]
[6]
Assembly to global boundary force vector
30

Body force vector
(distributed
loads)
(Point loads)
For element (1) Given:
For element (2) Given:
31

The global force vector
9
15.
3
4
15.
3
20
20
=KNOWN
Post-processing
32
Timoshenko Beam
In a Timoshenko Beam, a plane normal to
the beam axis before deformation doesnt
remain normal after deformation (short and
thick beams, sandwich composite beams).
dv
:
i.e. 6=
dx
dv
= +
Transverse Shear strain
dx
d
Normal strain
= y
dx
The strain energy of an element may be written as
G - shear modulus
b - widthh
- correction factor
h - height
33
Timoshenko Beam
98
Chapter 9: THE TL TIMOSHENKO PLANE BEAM ELEMENT
2-node C 1 (cubic) element

for Euler-Bernoulli beam model:
plane sections remain plane and
normal to deformed longitudinal axis
2-node C 0 linear-displacement-and-rotations
element for Timoshenko beam model:
plane sections remain plane but not
normal to deformed longitudinal axis
(b)
(a)
C element
with same DOFs
Figure 9.6. Sketch of the kinematics of two-node beam finite element models based on
(a) Euler-Bernoulli beam theory, and (b) Timoshenko beam theory. These
models are called C 1 and C 0 beams, respectively, in the FEM literature.
34
Timoshenko Beam
Substituting
the the
relations
intointo
U U
Substituting
relations
Substituting the relations into U
- moment of inertia
II -moment
of inertia
- moment
inertia
A
- of
cross-sectional
area
A-cross-sectional
A
- cross-sectional area
area
Galerkins
Approximation
(Linear
shape
functions)
Galerkins
Approximation
(Linear
shape
functions)
Galerkins Approximation (Linear shape functions)
Element
stiffness
Element
stiffness
Element stiffness
35
Timoshenko Beam
Exact integration
Shear Locking
Example: Cantilever Beam,

20 elements, different
span L to depth h
Reduced 1-point quadrature integration
Ratio of tip displacement for

Timoshenko beam theory and
Euler-Bernoulli theory
36
Plates vs. Shell Structures

A plate is a three dimensional body with
(i) one of the plate dimensions much smaller than the other two
(ii) the curvature of the plate mid-surface in the reference
configuration is zero
A shell is a three dimensional body with

(i) one of the plate dimensions much smaller than the other two
(ii) the curvature of the plate mid-surface in the current configuration
is non-zero
37
3D Solids vs. (Plates & Shells)
Plates and Shells are a special case of 3D solids (which are much
easier to handle).
When t is very small compared to other dimensions, full 3D numerical

treatment is costly and leads to serious ill-conditioning problems
When employing plates or shell models it is important to consider the

slenderness of the loading
However, if we are interested in global behavior, local details may be

neglected
38
Plate Theories
Several different plate-theories exist
Several different plate-theories
Kirchhoff Bernoulli - Euler Equivalent

The particles located normal to the neutral surface of an undeformed plate, after
deformation, remain on a straight line but this straight line is not necessarily
normal to the deformed neutral surface
39
V@
sides
andand
thethenodal
, andnormal
normalrotations
rotations8,,8,,and
andBy,By,atateach
eachnode
nodei iasasdegrees-ofdegrees-ofsides
nodaldeflection
deflectionw ,wand
freedom
(Figures
2a2aand
3a)3a)belong
freedom
(Figures
and
belongtotothis
thiscategory.
category.Typical
Typicalexamples
examplesofofsuch
suchelements
elementsare
arethe
the
of
Reference
2,
the
element
of
Reference
3
and
the
discrete
Kirchhoff
elements
elements
elements of Reference 2, the element of Reference 3 and the discrete Kirchhoff elementsofof
4 and
5. 5.
References
4 and
References
Plate Theories
Triangular Kirchhoff & Mindlin Elements

(a)
(a)
Wl
Wl
(b)
(b)
.-.-
w, w,
6,l
6,l
2 2
6x1
6x1
22
Figure
2 . 2Triangular
Kirchhoff
Figure
. Triangular(a)(a)
Kirchhoffand
and(b)(b)Mindlin
Mindlinplate
plateelements
elements
Quadrilateral Triangular Kirchhoff & Mindlin Elements
44
2 2
2 2
Figure
3. 3.
Quadrilateral
Kirchhoff
Mindlin
Figure
Quadrilateral(a)(a)
Kirchhoffand
and(b)(b)
Mindlinplate
plateelements
elements
40
Kirchhoff Plate
41
Kirchhoff Plate
Define the curvature of the plate
In-plane strain-displacement equations become
Assuming plane-stress conditions
1
0
E
1
D=
0
1 2
1
0 0
(Plane-stress)
42
Kirchhoff Plate
Moments are defined as
Equilibrium Conditions
Q - Shear forces
p distributed pressure
Elimination of shear forces
biharmonic equation
bending rigidity
43
Classic Plate Bending Element

Tocher, J.L. Analysis of plate bending using triangular elements, PhD thesis, UC Berkeley 1962
Consider 3-node triangular element (3-dofs per node)
stiffness matrix
from: J.L Tocher, Analysis of plate bending using triangular elements
44
The Beam Element in Large Displacements

Assumptions:
Plane sections initially normal to the neutral axis remain plane
The longitudinal ( ) and the two shear stresses( , ) are the only non
zero ones.
A general 3D beam element would then be:
45

Along the lines of the truss element formulation we need to express the
coordinates of a random point within the beam element. Using (r, s, t) as
the Cartesian coordinates at a point within an element with N nodal
points, this is written:
xi =
N
X
k=1
Nk t xki +
k=1
k=1
sX
tX
ak Nk t Vtik +
bk Nk t Vsik ,
2
2
i = 1, 2, 3
Vectors Vs and Vt define the orientation of the cross-section for the beam:
They are normal to the axis of the beam and to each other. The values a
and b define the size of the cross section of the beam.
The relative displacement components would be:
ui = t xi 0 xi
ui = t + t xi t xi
are the incremental components
46

This leads to the following formulations
ui =
N
X
Nk t uki +
k=1
and ui =
N
X
k=1
k=1
k=1
tX
sX
k
k
ak Nk (t Vtik 0 Vti
)+
bk Nk (t Vsi
0 Vsik )
2
2
Nk uki
k=1
k=1
sX
tX
ak Nk Vtik +
bk Nk Vsik ,
+
2
2
i = 1, 2, 3
where
k
Vti
= t + t Vtik t Vtik
k
Vsi
= t + t Vsik t Vsik
47

We need to express the components Vtik , ZVsik of the vectors Vtk , Vsk in
terms of the nodal rotational degrees of freedom per node k:
kT =
xk
yk
zk
This in linear analysis would be written as:

k
Vtk = k t Vt
Vsk = k t Vs
but in the case of large displacements a second order Taylor approximation

needs to be used:
1
k
k
Vtk = k t Vt + k (k t Vt )
2
1
k
k
Vsk = k t Vs + k (k t Vs )
2
48

The finite element equations in this case are:
..
.
uk
k
..
.
t + t
=
R tF
(1)
Having solved Eqn (1) for bf uk , k , we obtain the approximations for the
nodal point displacement and direction vectors:
t + t
uk = t uk + uk
Z
k
t + t
Vt = t Vtk +
dk Vtk
Vk
Z
k
t + t
Vs = t Vsk +
dk Vsk
Vk
The above formulation corresponds to an iteration of the Newton-Raphson

method
49

Example:
Assume the following 2-node beam element. Evaluate the coordinate and
displacement interpolations and derivatives required for the setup of the
strain displacement matrices in the TL and UL formulation.
50

Example:
First we need to calculate the nodal coordinates at times t and 0 (t x1 , t x2 ),
(0 x1 , 0 x2 ). Using the corresponding geometries and the following formula:
t
xi =
N
X
k=1
Nk t xki +
k=1
k=1
tX
sX
ak Nk t Vtik +
bk Nk t Vsik ,
2
2
i = 1, 2, 3
where Nk are linear shape functions, we obtain:
51

Example:
Hence, the displacement components are at time t given by the following
formula:
t
ui =
N
X
k=1
Nk t uki +
N
N
tX
sX
k
k
ak Nk (t Vtik 0 Vti
)+
bk Nk (t Vsi
0 Vsik ),
2
2
k=1
i = 1, 2
k=1
hence:
52

Example:
For the calculation of the incremental displacements we will need to
calculate the rotational components Vts (since Vtk = 0). Using the
following formula:
1
k
k
Vsk = k t Vs + k (k t Vs )
2
we get:

e1
e2
e3
k cos(t k )

k
0
0
= k sin(t k )
k t Vs =
t
t
sin( k ) cos( k ) 0
0
and k (k
k
t
Vs )

e1
e2
e3
k sin(t k )

0
0
= k2 cos(t k )
=
k cos(t k ) k sin(t k ) 0
0
53

Example:
Then, the incremental displacements are given by
and ui =
N
X
k=1
Nk uki +
k=1
k=1
tX
sX
ak Nk Vtik +
bk Nk Vsik ,
2
2
i = 1, 2
hence:
54

Example:
Finally, the required derivatives for the TL and UL formulation will be:
where we assumed t L = 0 L = L.
55

Example:
Also, the Jacobian will be:
And the associated displacement derivatives are:
These will be used to derive the strain-displacement matrices B and finally the
stiffness matrix K
56

Lecture 7 - 5 November, 2010
Constitutive Relations
Previously we examined the kinematic equations formulation
(displacement, strain displacement relations)
The next step is to determine appropriate constitutive relationships
of the form:
= f ()
ex. linear analysis = E
When dealing with incremental analysis this is written in tensor
form for time t:
t
= t Cijrs t rs
It is necessary that kinematic and constitutive relations are

appropriate (ex. The Second Piola-Kirchhoff stress tensor is to
be used with the Green Lagrange strain tensor).
Ultimately, the position of the observer frame should not affect
the constitutive relations of a material (material frame
indifference or objectivity of the material response).
Overview of Material Descriptions
We can discriminate amongst the following major classes of

material behavior
Elastic, linear or nonlinear
Hyperelastic
Hypoelastic
Elastoplastic
Creep
Viscoplastic
Solution Flowchart
General Solution process in incremental nonlinear FE
Known Solution at t:
Stresses
, strains ,
Internal material parameters
Known Quantities at iterations i- 1 :

Nodal Displacements at first Iteration:
and hence
Element strains
Repat till Convergence
Calculate at t+ t:
Stresses +
Tangent stress strain matrix

Elastic Analysis: directly obtain

+
, from
Inelastic Analysis: Integrate to get
Calculate:
Incremental Displacement Vector i :
+
i =
Then,
+ i
Notation
Main Stress - Strain pairs:

Material Nonlinearity (small deformations)
Engineering Stress
Engineering Strain
TL formulation (large deformations)
2nd Piola-Kirchhoff Stress S
Green-Lagrange Strain
UL formulation (large deformations)
Cauchy Stress
Almansi Strain A
Elastic Material
For an elastic material the stress is a function of strain only The
stress path is the same both in loading and unloading
Linear Elastic
The elasticity (constitutive)
tensor components, Cijrs are
constant
Nonlinear Elastic
a function of strain
Example: Almost all materials under small stress
Elastic Material
For the case of an elastic material we already saw that the TL
Formulation (used for large deformation analysis) yields:
t
0
Sij = t0 Cijrs t0 rs
The elasticity tensor for 3D stress conditions is defined as:

t
Cijrs = ij rs + (ir js + is jr)
where and are the Lame constants and ij is the Kronecker delta,
E
,
()(1 2)

0 i 6= j
ij =
1 i=j
E
2(1 + )
Elastic Material
Important Note
The 2nd Piola-Kirchhoff (PK2) stress and Green-Lagrange strain
tensor components are invariant to rigid body motions.
For problems with small strains we can take advantage of this
observation and use any constitutive relationship that has been
developed for engineering stress and strain measures by just
substituting with the PK2 stress and Green-Lagrange strain
This observation can be extended to all problems with large
deformations but small strain conditions such as the elastic or
elastoplastic buckling problem and the collapse analysis of slender
structures.
Elastic Material
UL Formulation
We now write
t
ij = tt C ijrs tt Ars
where the elasticity tensor components of the UL constitutive matrix

C , are related to the one of the TL formulation C, through the
following relationship:
t
t
C ijrs =
t
0
t
0
xi,m
t
0
xj,n
t
0
Cmnpq
t
0
xr,p
t
0
xs,q
Also the Almansi strain tensor is related to the Green-Lagrange one

through
t A
t
= 0t XT 0t 0t X
Therefore the two formulations are equivalent and interchangeable.

10
Example
Consider the four node element shown below. Examine the effect of using two
different stress measures and the same constitutive matrix on the Cauchy stresses.
A. TL formulation Using the PK2 stress measure and the Green-Lagrange strain
tensor we have:
t
t
t
(1)
0 Sij = 0 Cijrs 0 rs
11
Example
In order to define the required stresses and strains we need to determine the deformation
gradient t0 S, which in 2D is written as:
t
x1
t x1
t
0 x2
0 x1
(2)
0X = tx
t x2
2
0 x1
0 x2
The coordinates of a random point within the element are given as the weighted sum of
the nodal coordinates, where the weights are the shape functions. However, the shape
functions are written with respect to the r, s system (isoparametric representation) as:
1
1
(1 + r)(1 + s) N2 = (1 r)(1 + s)
4
4
1
1
N3 = (1 r)(1 s) N4 = (1 + r)(1 s)
4
4
N1 =
(3)
From the given figure however we see that the 0 x1 , 0 x2 system is related to the r, s
system through:
0
x1 = r + 1
x2 = s + 1
1
N 1 = 0 x1 0 x2
4
1
N2 = (2 0 x1 )(0 x2 )
4
1
1
0
0
N3 = (2 x1 )(2 x2 ) N4 = 0 x1 (2 0 x2 )
4
4
(4)
12
Example
Therefore, we ultimately have:
4
X
t xi
=
0
xj
k=1
Nk
0 xj
t k
xi
(5)
The nodal coordinates at time t are:

1
(t x1 , t x12 ) = (2 + 4t, 2)
(t x31 , t x32 ) = (0, 0)
(t x21 , t x22 ) = (4t, 2)

(t x41 , t x42 ) = (2, 0)
(6)
By substituting (4), (5) and (6) into Eqn (2) we obtain:

"
#
1 2t
t
0X =
0 1
From Lecture 4 we know that the Green-Langrange strain tensor will then be:
t
0
1 t Tt
(0 X 0 X I)
2
"
#
0
t
t

=
0
t 2t2
13
Example
Also, for plane strain strain conditions the constitutive tensor is:
1
0
6731 2885
E 1
0
E=5000,=0.3
2885 6731
C=
1 2
1
0
0
0 0
2
1923
Now, using the assumption of small strain we can use the above constitutive
tensor for the relationship between the PK2 stress and the Green-Lagrange strain.
Hence from Eqn (1):

S11
5770t2

2
S22 = 13462t
S12
3846t
14
Example
Then, the Cauchy stress at time t can be obtained from the PK2
stress as (Lecture 4):
t
= 0 t0 X t0 S t0 XT
21000t2 + 54000t4
11

13000t2
22 =
12
3800t + 27000t3
15
Example
B. Jaumann stress rate formulation
This formulation uses the following constitutive relationship:
t
ij = t Cijrs
Drs
(7)
where the velocity strain tensor t D is computed using the velocity gradient L (see
Lecture 4) :
"
#
0 2
1 =
L = XX
0 0
L can be decomposed to s symmetric part D = DT (the velocity strain tensor) and a
skew symmetric part W = WT (the spin tensor):
L = D + W which in this case yields
"
#
"
#
0 1
0
1
D=
, W=
1 0
1 0
Now we use the same constitutive matrix C and Eqn (7) to obtain the Jaumann stress at
time t:
11
0
22 = 0
12
3846
16
Example
The Jaumann stress is connected to the Cauchy stress through:
ij = ij + ip Wpj + jp Wpi
which from the above yields the follwoing formula for the Cauchy components ij
11

22 =
12
212
212
3846 + 22 11
The above system of ordinary differential equations can be solved to finally get:

11
1900(1 cos2t)
22 = 1900(1 cos2t)
12
1900sin2t
The results from methods A and B are rather close for small values of the deformation
measure t but grow quite different as t gets larger than 0.1, indicating that the same C
can no longer be used.
17
Hyperelastic Material
Hyperelastic (rubberlike) materials

exhibit an incompressible response,
path independence and no energy
dissipation.
The stress is now calculated through
the strain energy functional W
t
0
Sij =
W
t0 ij
Figure: Stress-strain curves for various
hyperelastic material models.
18
Hyperelastic Material Models
Saint Venant-Kirchhoff model
W () = [tr()]2 + tr(2 )

2
and the second Piola-Kirchhoff stress can be derived as
S = [tr()]I + 2
, are the Lame constants
Mooney-Rivlin model
W () = C1 (I1 3) + C2 (I2 3)
where C1 and C2 are empirically determined material constants and
I1 = tr(C) = C11 + C22 + C33
where C is the Cauchy-Green deformation tensor (see Lecture 4) and
I2 =
1
[(I1 )2 tr(C)2 ]
2
19
Hypoelastic Material
In this case, stress increments are calculated from strain increments
dij = Cijrs drs

The material moduli Cijrs are defined as functions of
stress
strain
fracture criteria
loading and unloading parameters
maximum strains reached and so on
Example Concrete models
(*Hypoelasticity is not an elastic type of behavior in the sense that it
does not exhibit path independence)
20
Inelasticity
Elastoplasticity, Creep and Viscoplasticity are types of Inelastic
behavior
Elastic behavior stresses can be directly calculated from the strain
Inelastic behavior the stress at time t depends on the stress strain
history
In the incremental analysis of inelastic response we had three main scenarios
Small displacements-rotations / small strains use linear elastic
solution, engineering stress and strain measures
Large displacements-rotations / small strains use TL formulation
by substituting the appropriate stress - strain measures (PK2,
Green-Lagrange) in the place of the engineering stress and strain
measures
Large displacements-rotations / large strains use either TL or UL
formulation, more complex constitutive laws
21
Elastoplasticity
In this formulation we encounter a linearly elastic behavior until yield
and usually a gardening post yield behavior
Examples Metals, soild and Rocks when subjected to high stresses

22
Elastoplasticity
The strain and stress increments are given by:
E
dij = Cijrs (drs drs )
drs = drs + drs

E
where Cijrs are the components of the elastic constitutive tensor and drs , drs ,
P
drs are the components of the total strain increment.

To calculate the plastic strains we use the following three properties:
P
Yield Function fy (, )
fy < 0 Elastic behavior
fy = 0 Plastic behavior
fy > 0 Inadmissible
Flow rule
The yield function is used in the flow rule in order to obtain the plastic
strain increments
P
dij =
fy
ij
is a scalar to be determined
Hardening rule
This specifies how the yield function is modified during plastic flow
23
Elastoplasticity
Example: Von Mises yield criterion (in 3D):
2
2
2
fy = 0 (11 22 )2 + (22 33 )2 + (11 33 )2 + 6(12
+ 23
+ 31
) 2y2 = 0
24
Elastoplasticity
Isotropic & Kinematic hardening Rules
In the case of isotropic hardening, the yield surface expands
uniformly.
In the case of kinematic hardening, the size of the yield surface
remains unchanged and the center location of the yield surface
is shifted.
25
Elastoplasticity
Response for cyclic loading
Isotropic hardening: the yield stress is higher as the cyclic loading progresses
Kinematic hardening: the difference between unloading stress and new yield
stress in the opposite direction of loading is constant and equal to 2y .
Figure: isotropic hardening in tension

Figure: kinematic hardening

26
Thermoelastoplasticity and Creep

This behavior exhibits the time effect of increasing strains under constant
loads or decreasing stress under constant deformations (relaxation)
Typical examples of such behavior are metals at high temperatures
The thermal strain( = T ) and the creep strain now enter the
formulation of the stress strain relationships.
Creep
Creep is the tendency of a solid material to slowly move or deform
permanently under constant stresses. Creep tests measure the strain
response due to a constant stress. The classical creep curve represents the
evolution of strain as a function of time in a material subjected to uniaxial
stress at a constant temperature. The creep test, for instance, is performed
by applying a constant force/stress and analyzing the strain response of the
system. In general, this curve usually shows three phases or periods of
behavior.
27
Creep
1. A primary creep stage, also known as transient creep, is the starting stage
during which hardening of the material leads to a decrease in the rate of flow
which is initially very high. (0 1 ).
2. The secondary creep stage, also known as the steady state, is where the strain
rate is constant. (1 2 ).
3. A tertiary creep phase in which there is an increase in the strain rate up to the
fracture strain. (2 R ).
28
Relaxation
A relaxation test is defined as the stress response due to a constant strain for a
period of time. In viscoplastic materials, relaxation tests demonstrate the stress
relaxation in uniaxial loading at a constant strain. The decompositon of strain
d
de dvp
rate is
=
+
dt
dt
dt
d
de
The elastic part of the strain rate is given by
= E1
dt
dt
For the flat region of the strain-time curve, the total strain rate is zero.
dvp
d
= E1
Hence we have,
dt
dt
Therefore the relaxation curve can be used to determine rate of viscoplastic strain
and hence the viscosity of the dashpot in a 1D viscoplastic material model.
29
Viscoplasticity
Viscoplasticity describes the rate-dependent inelastic behavior of
solids. Rate-dependence in this context means that the deformation
of the material depends on the rate at which loads are applied[1].
The inelastic behavior that is the subject of viscoplasticity is plastic
deformation which means that the material undergoes unrecoverable
deformations when a load level is reached. Rate-dependent plasticity
is important for transient plasticity calculations.
The main difference between rate-independent plastic and
viscoplastic material models is that the latter exhibit not only
permanent deformations after the application of loads but continue
to undergo a creep flow as a function of time under the influence of
the applied load.
Typical examples of such behavior are Polymers and Metals
30
Viscoplasticity
The elastic response of viscoplastic materials can be represented in one-dimension
by Hookean spring elements. Rate-dependence can be represented by nonlinear
dashpot elements.
Plasticity can be accounted for by
adding sliding frictional elements. In the
figure E is the modulus of elasticity, is
the viscosity parameter and N is a
power-law type parameter that
represents non-linear dashpot
d 1/N
=
. The sliding element can
dt
have a yield stress (y) that is strain
rate dependent, or even constant, as
shown in Figure (c).
31
Viscoplasticity
Stress-strain response of a
viscoplastic material at different
strain rates. The dotted lines
show the response if the
strain-rate is held constant. The
blue line shows the response when
the strain rate is changed
suddenly.
32
NL FE Special Considerations - The Contact Problem
Difficult non linear behavior = contact between two or more

bodies
Contacts = From frictionless in small displacement to friction in
general large strain conditions
Nonlinearity of the analysis is not only geometric and material
but also contact conditions
33
Contact Conditions
Usual term
Contribution of contact
forces
Consider N bodies that are in contact at
time t:
t
Sc is the complete area of contact

t c
fi : Components of the contact
tractions t fiS : components of the known
externally applied traction.
Then the virtual work for the N bodies at time t is give by:
34
Contact Conditions
Usual term
Cont
force
We denote the two bodies as I and J. Each body is supported such that
without contact no rigid motion is possible.
35
Definitions used in Contact Analysis

Let t f IJ : vector of contact surface traction on body I due to contact with
body J then t f IJ = t f JI . Hence, the virtual work due to the contact
traction can be written:
Z
Z
Z
uIi fiIJ dS IJ +
uJi fiJI dS JI =
uIJi fiIJ dS IJ
S IJ
S IJ
S IJ
where
where
uIJi = uIi uJi
We call the pair of surfaces S IJ , S IJ a contact surface pair.

S IJ is named the contactor surface and
S JI is named the target surface
Hence, the right hand side of the previous eqn represents the virtual work
of the contact tractions over the relative virtual displacements uIi , uJi
36
37
Contact Analysis-Interface Conditions

Normal Conditions
38
39

Tangential Conditions
Coulombs Law
of friction
Definitions
Coulombs
lawStates:
of friction states :
40
Contact Analysis
Complexities
In the previous we consider pseudo-static contact conditions
In dynamic analysis:
Body forces include inertial force effects and the kinematic interface
conditions must be satisfied at all instances of time, requiring
displacement, velocity and acceleration compatibility between the
contacting bodies.
The time integrations schemes (ex. trapezoidal rule) do not
automatically satisfy compatibility which therefore has to be imposed
separately on the step by step solution
Various algorithms have been proposed to solve contact problems in
Finite Element analysis.
41
Contact Analysis-Solution Approach

The Constraint Function Method
Let w be a function of and g such that the solutions of w(g, ) = 0
satisfy the Normal conditions
Let v be a function of and u such that the solutions v(u,
) = 0 satisfy
the Tangential Conditions. Then, the contact conditions are given by:
w(g, ) = 0
v(u,
) = 0
These can now be imposed on the principle of virtual work using either a
penalty approach or a Lagrange Multiplier method. Variables and can
be considered Lagrange multipliers, and so we consider their variations ,
. By multiplying by the variations and integrating in the domain we
obtain the constraint equation:
Z
[ w(g, ) + u(u,
)]dS IJ = 0
S IJ
The governing equations of motion in this case are now both the principle
of virtual work and the constraint equation
42

Previously we examined the kinematic equations formulation
(displacement, strain displacement relations)
The next step is to determine appropriate constitutive relationships
of the form:
= f ()
ex. linear analysis = E
When dealing with incremental analysis this is written in tensor
form for time t:
t
= t Cijrs t rs
It is necessary that kinematic and constitutive relations are

appropriate (ex. The Second Piola-Kirchhoff stress tensor is to
be used with the Green Lagrange strain tensor).
Ultimately, the position of the observer frame should not affect
the constitutive relations of a material (material frame
indifference or objectivity of the material response).
Overview of Material Descriptions
We can discriminate amongst the following major classes of

material behavior
Elastic, linear or nonlinear
Hyperelastic
Hypoelastic
Elastoplastic
Creep
Viscoplastic
Solution Flowchart
General Solution process in incremental nonlinear FE
Known Solution at t:
Stresses
, strains ,
Known Quantities at iterations i- 1 :

Nodal Displacements at first Iteration:
and hence
Element strains
Repat till Convergence
Calculate at t+ t:
Stresses +
Tangent stress strain matrix

Elastic Analysis: directly obtain

+
, from
Inelastic Analysis: Integrate to get
Calculate:
Incremental Displacement Vector i :
+
i =
Then,
+ i
Notation
Main Stress - Strain pairs:

Material Nonlinearity (small deformations)
Engineering Stress
Engineering Strain
TL formulation (large deformations)
2nd Piola-Kirchhoff Stress S
Green-Lagrange Strain
UL formulation (large deformations)
Cauchy Stress
Almansi Strain A
Elastic Material
For an elastic material the stress is a function of strain only The
stress path is the same both in loading and unloading
Linear Elastic
constant
Nonlinear Elastic
a function of strain
Example: Almost all materials under small stress
Elastic Material
For the case of an elastic material we already saw that the TL
Formulation (used for large deformation analysis) yields:
t
0
Sij = t0 Cijrs t0 rs
The elasticity tensor for 3D stress conditions is defined as:

t
Cijrs = ij rs + (ir js + is jr)
where and are the Lame constants and ij is the Kronecker delta,
E
,
()(1 2)

0 i 6= j
ij =
1 i=j
E
2(1 + )
Elastic Material
Important Note
The 2nd Piola-Kirchhoff (PK2) stress and Green-Lagrange strain
tensor components are invariant to rigid body motions.
For problems with small strains we can take advantage of this
observation and use any constitutive relationship that has been
developed for engineering stress and strain measures by just
substituting with the PK2 stress and Green-Lagrange strain
This observation can be extended to all problems with large
deformations but small strain conditions such as the elastic or
elastoplastic buckling problem and the collapse analysis of slender
structures.
Elastic Material
UL Formulation
We now write
t
ij = tt C ijrs tt Ars
where the elasticity tensor components of the UL constitutive matrix

C , are related to the one of the TL formulation C, through the
following relationship:
t
t
C ijrs =
t
0
t
0
xi,m
t
0
xj,n
t
0
Cmnpq
t
0
xr,p
t
0
xs,q
Also the Almansi strain tensor is related to the Green-Lagrange one

through
t A
t
= 0t XT 0t 0t X
Therefore the two formulations are equivalent and interchangeable.

10
Example
Consider the four node element shown below. Examine the effect of using two
different stress measures and the same constitutive matrix on the Cauchy stresses.
A. TL formulation Using the PK2 stress measure and the Green-Lagrange strain
tensor we have:
t
t
t
(1)
0 Sij = 0 Cijrs 0 rs
11
Example
In order to define the required stresses and strains we need to determine the deformation
gradient t0 S, which in 2D is written as:
t
x1
t x1
t
0 x2
0 x1
(2)
0X = tx
t x2
2
0 x1
0 x2
The coordinates of a random point within the element are given as the weighted sum of
the nodal coordinates, where the weights are the shape functions. However, the shape
functions are written with respect to the r, s system (isoparametric representation) as:
1
1
(1 + r)(1 + s) N2 = (1 r)(1 + s)
4
4
1
1
N3 = (1 r)(1 s) N4 = (1 + r)(1 s)
4
4
N1 =
(3)
From the given figure however we see that the 0 x1 , 0 x2 system is related to the r, s
system through:
0
x1 = r + 1
x2 = s + 1
1
N 1 = 0 x1 0 x2
4
1
N2 = (2 0 x1 )(0 x2 )
4
1
1
0
0
N3 = (2 x1 )(2 x2 ) N4 = 0 x1 (2 0 x2 )
4
4
(4)
12
Example
Therefore, we ultimately have:
4
X
t xi
=
0
xj
k=1
Nk
0 xj
t k
xi
(5)
The nodal coordinates at time t are:

1
(t x1 , t x12 ) = (2 + 4t, 2)
(t x31 , t x32 ) = (0, 0)
(t x21 , t x22 ) = (4t, 2)

(t x41 , t x42 ) = (2, 0)
(6)
By substituting (4), (5) and (6) into Eqn (2) we obtain:

"
#
1 2t
t
0X =
0 1
From Lecture 4 we know that the Green-Langrange strain tensor will then be:
t
0
1 t Tt
(0 X 0 X I)
2
"
#
0
t
t

=
0
t 2t2
13
Example
Also, for plane strain strain conditions the constitutive tensor is:
1
0
6731 2885
E 1
0
E=5000,=0.3
2885 6731
C=
1 2
1
0
0
0 0
2
1923
Now, using the assumption of small strain we can use the above constitutive
tensor for the relationship between the PK2 stress and the Green-Lagrange strain.
Hence from Eqn (1):

S11
5770t2

2
S22 = 13462t
S12
3846t
14
Example
Then, the Cauchy stress at time t can be obtained from the PK2
stress as (Lecture 4):
t
= 0 t0 X t0 S t0 XT
21000t2 + 54000t4
11

13000t2
22 =
12
3800t + 27000t3
15
Example
B. Jaumann stress rate formulation
This formulation uses the following constitutive relationship:
t
ij = t Cijrs
Drs
(7)
where the velocity strain tensor t D is computed using the velocity gradient L (see
Lecture 4) :
"
#
0 2
1 =
L = XX
0 0
L can be decomposed to s symmetric part D = DT (the velocity strain tensor) and a
skew symmetric part W = WT (the spin tensor):
L = D + W which in this case yields
"
#
"
#
0 1
0
1
D=
, W=
1 0
1 0
Now we use the same constitutive matrix C and Eqn (7) to obtain the Jaumann stress at
time t:
11
0
22 = 0
12
3846
16
Example
The Jaumann stress is connected to the Cauchy stress through:
ij = ij + ip Wpj + jp Wpi
which from the above yields the follwoing formula for the Cauchy components ij
11

22 =
12
212
212
3846 + 22 11
The above system of ordinary differential equations can be solved to finally get:

11
1900(1 cos2t)
22 = 1900(1 cos2t)
12
1900sin2t
The results from methods A and B are rather close for small values of the deformation
measure t but grow quite different as t gets larger than 0.1, indicating that the same C
can no longer be used.
17
Hyperelastic (rubberlike) materials

exhibit an incompressible response,
path independence and no energy
dissipation.
The stress is now calculated through
the strain energy functional W
t
0
Sij =
W
t0 ij
Figure: Stress-strain curves for various
hyperelastic material models.
18
Hyperelastic Material Models
Saint Venant-Kirchhoff model
W () = [tr()]2 + tr(2 )

2
and the second Piola-Kirchhoff stress can be derived as
S = [tr()]I + 2
, are the Lame constants
Mooney-Rivlin model
W () = C1 (I1 3) + C2 (I2 3)
where C1 and C2 are empirically determined material constants and
I1 = tr(C) = C11 + C22 + C33
where C is the Cauchy-Green deformation tensor (see Lecture 4) and
I2 =
1
[(I1 )2 tr(C)2 ]
2
19
Hypoelastic Material
In this case, stress increments are calculated from strain increments
dij = Cijrs drs

The material moduli Cijrs are defined as functions of
stress
strain
fracture criteria
loading and unloading parameters
maximum strains reached and so on
Example Concrete models
(*Hypoelasticity is not an elastic type of behavior in the sense that it
does not exhibit path independence)
20
Inelasticity
Elastoplasticity, Creep and Viscoplasticity are types of Inelastic
behavior
Elastic behavior stresses can be directly calculated from the strain
Inelastic behavior the stress at time t depends on the stress strain
history
In the incremental analysis of inelastic response we had three main scenarios
Small displacements-rotations / small strains use linear elastic
solution, engineering stress and strain measures
Large displacements-rotations / small strains use TL formulation
by substituting the appropriate stress - strain measures (PK2,
Green-Lagrange) in the place of the engineering stress and strain
measures
Large displacements-rotations / large strains use either TL or UL
formulation, more complex constitutive laws
21
Elastoplasticity
In this formulation we encounter a linearly elastic behavior until yield
and usually a gardening post yield behavior
Examples Metals, soild and Rocks when subjected to high stresses

22
Elastoplasticity
The strain and stress increments are given by:
E
dij = Cijrs (drs drs )
drs = drs + drs

E
where Cijrs are the components of the elastic constitutive tensor and drs , drs ,
P
drs are the components of the total strain increment.

To calculate the plastic strains we use the following three properties:
P
Yield Function fy (, )
fy < 0 Elastic behavior
fy = 0 Plastic behavior
fy > 0 Inadmissible
Flow rule
The yield function is used in the flow rule in order to obtain the plastic
strain increments
P
dij =
fy
ij
is a scalar to be determined
Hardening rule
This specifies how the yield function is modified during plastic flow
23
Elastoplasticity
Example: Von Mises yield criterion (in 3D):
2
2
2
fy = 0 (11 22 )2 + (22 33 )2 + (11 33 )2 + 6(12
+ 23
+ 31
) 2y2 = 0
24
Elastoplasticity
Isotropic & Kinematic hardening Rules
In the case of isotropic hardening, the yield surface expands
uniformly.
In the case of kinematic hardening, the size of the yield surface
remains unchanged and the center location of the yield surface
is shifted.
25
Elastoplasticity
Response for cyclic loading
Isotropic hardening: the yield stress is higher as the cyclic loading progresses
Kinematic hardening: the difference between unloading stress and new yield
stress in the opposite direction of loading is constant and equal to 2y .
Figure: isotropic hardening in tension

Figure: kinematic hardening

26
Thermoelastoplasticity and Creep

This behavior exhibits the time effect of increasing strains under constant
loads or decreasing stress under constant deformations (relaxation)
Typical examples of such behavior are metals at high temperatures
The thermal strain( = T ) and the creep strain now enter the
formulation of the stress strain relationships.
Creep
Creep is the tendency of a solid material to slowly move or deform
permanently under constant stresses. Creep tests measure the strain
response due to a constant stress. The classical creep curve represents the
evolution of strain as a function of time in a material subjected to uniaxial
stress at a constant temperature. The creep test, for instance, is performed
by applying a constant force/stress and analyzing the strain response of the
system. In general, this curve usually shows three phases or periods of
behavior.
27
Creep
1. A primary creep stage, also known as transient creep, is the starting stage
during which hardening of the material leads to a decrease in the rate of flow
which is initially very high. (0 1 ).
2. The secondary creep stage, also known as the steady state, is where the strain
rate is constant. (1 2 ).
3. A tertiary creep phase in which there is an increase in the strain rate up to the
fracture strain. (2 R ).
28
Relaxation
A relaxation test is defined as the stress response due to a constant strain for a
period of time. In viscoplastic materials, relaxation tests demonstrate the stress
relaxation in uniaxial loading at a constant strain. The decompositon of strain
d
de dvp
rate is
=
+
dt
dt
dt
d
de
The elastic part of the strain rate is given by
= E1
dt
dt
For the flat region of the strain-time curve, the total strain rate is zero.
dvp
d
= E1
Hence we have,
dt
dt
Therefore the relaxation curve can be used to determine rate of viscoplastic strain
and hence the viscosity of the dashpot in a 1D viscoplastic material model.
29
Viscoplasticity
Viscoplasticity describes the rate-dependent inelastic behavior of
solids. Rate-dependence in this context means that the deformation
of the material depends on the rate at which loads are applied[1].
The inelastic behavior that is the subject of viscoplasticity is plastic
deformation which means that the material undergoes unrecoverable
deformations when a load level is reached. Rate-dependent plasticity
is important for transient plasticity calculations.
The main difference between rate-independent plastic and
viscoplastic material models is that the latter exhibit not only
permanent deformations after the application of loads but continue
to undergo a creep flow as a function of time under the influence of
the applied load.
Typical examples of such behavior are Polymers and Metals
30
Viscoplasticity
The elastic response of viscoplastic materials can be represented in one-dimension
by Hookean spring elements. Rate-dependence can be represented by nonlinear
dashpot elements.
Plasticity can be accounted for by
adding sliding frictional elements. In the
figure E is the modulus of elasticity, is
the viscosity parameter and N is a
power-law type parameter that
represents non-linear dashpot
d 1/N
=
. The sliding element can
dt
have a yield stress (y) that is strain
rate dependent, or even constant, as
shown in Figure (c).
31
Viscoplasticity
Stress-strain response of a
viscoplastic material at different
strain rates. The dotted lines
show the response if the
strain-rate is held constant. The
blue line shows the response when
the strain rate is changed
suddenly.
32
Difficult non linear behavior = contact between two or more

bodies
33
Contact Conditions
Usual term
Contribution of contact
forces
time t:
t
Sc is the complete area of contact

t c
fi : Components of the contact
tractions t fiS : components of the known
34
Contact Conditions
Usual term
Cont
force
35

Z
Z
Z
uIi fiIJ dS IJ +
uJi fiJI dS JI =
uIJi fiIJ dS IJ
S IJ
S IJ
S IJ
where
where
uIJi = uIi uJi

36
37

Normal Conditions
38
39

Coulombs Law
of friction
Definitions
Coulombs
lawStates:
40
Contact Analysis
Complexities
contacting bodies.
41
Contact Analysis-Solution Approach

) = 0 satisfy
w(g, ) = 0
v(u,
) = 0
Z
[ w(g, ) + u(u,
)]dS IJ = 0
S IJ
42


The Contact Problem is a special consideration of Nonlinear Analysis
from the point of view that it enforces two extra constraints that
need to be satisfied at the contact area.
Difficulties
Complex non linear behavior = contact between two or more
bodies
Contact Conditions

time t:
t
Sc is the complete area

t c
fi : Components of the
of contact
contact
tractions
t S
fi : components of the known
denote 2 body I and J

ch body is supported such that without contact no rigid motion is possible
vector of contact surface traction on body I due to contact with body J then tfIJ = - tfJI
e virtual work due to the contact traction can be written :
Contact Conditions
Usual term
Contribu
forces
Definitions used in Contact Problem

Z
Z
Z
uIi fiIJ dS IJ +
uJi fiJI dS JI =
uIJi fiIJ dS IJ
S IJ
S IJ
S IJ
where
where
uIJi = uIi uJi

Definitions used in Contact Problem
Contact Problem-Interface Conditions

Normal Conditions

Coulombs Law
of friction
Definitions
Coulombs
lawStates:
Contact Problem
Complexities
contacting bodies.
10
Contact Problem-Solution Approach

) = 0 satisfy
w(g, ) = 0
v(u,
) = 0
Z
[ w(g, ) + u(u,
)]dS IJ = 0
S IJ
11

The discretization of the principle of virtual work and the constraint
equation at time t + t yields:
t+t
F(t+t U) = t+t R t+t Rc (t+t U, t+t )
t+t
Fc (t+t U, t+t ) = 0
where the parameter vector

written as:
t+t
t + t
for m contactor nodes will be
= [1 , 1 , ..., k , k , ..., m , m ]
12

The nodal force vector t+t Rc can be derived from the visualization of the two
dimensional case of contact:
For each contactor node k and the
corresponding target nodes k1 , k2 , it is
defined as:
k (nk + k sk )
t+t k
Rc = (1 k )k (nk + k sk )
k k (nk + k sk )
Fkc for each contactor node k can be written as:

"
#
w(gk , k )
t+t k
Rc =
v(u k , k )
whereas the constraint vector
t+t
13
Special Considerations-Geometric Stiffness

An alternate approach to the Large Displacement Problem for practical
considerations (namely truss, beam problems)
A cable, when subjected to a large tension force, has an increased lateral stiffness.
If a long rod is subjected to a large compressive force, and is on the verge of
buckling, we know that the lateral stiffness of the rod has been reduced
significantly and a small lateral load may cause the rod to buckle. This general
type of behavior is caused by a change in the geometric stiffness of the
structure. This stiffness is a function of the load in the structural member and
can be either positive or negative.
14

Cable Element
The fundamental equations for the geometric stiffness for a rod or a cable are
very simple to derive. Consider the horizontal cable, of length L with an initial
tension T. If the cable is subjected to lateral displacements, vi and vj , at both
DYNAMIC ANALYSIS OF STRUCTURES
ends, as shown, then additional forces, Fi and Fj , must be developed for the
cable element to be in equilibrium in its displaced position.
Fi
T
Deformed Position
Fj
vi
T
vj
Note that we haveFigure

assumed
all forces
andActing
displacements
are positive
11.1.
Forces
on a Cable
Elementin the up
direction. We have also made the assumption that the displacements are small
Taking
moments
point j in
in the
the cable.
deformed position, the following equilibrium
and do not
changeabout
the tension
of Structural Engineering
equation can be Institute
written:
15

Taking moments about point j in the deformed position, the following
equilibrium equation can be written:
Fi =
T
(vi vj )
L
And from vertical equilibrium the following equation is apparent:

Fi = Fj
Combining the above the lateral forces can be expressed in terms of the lateral
displacements by the following matrix equation:

T
1
1
vi
Fi
=
or symbolically, FG = KG v
1
vj
Fj
L 1
Note that the 2 2 geometric stiffness, KG , matrix is not a function of the mechanical
properties of the cable and is only a function of the elements length and the force in the
element. Hence, the term geometric or stress stiffness matrix is introduced as opposed
to the mechanical stiffness matrix which is based on the physical properties. The
geometric stiffness exists in all structures; however, it only becomes important if it is
large compared to the mechanical stiffness of the structural system.
16

Beam Element
In the case of a beam element with bending properties in which the deformed
shape is assumed to be a cubic function due to the rotations i and j at the
ends, additional moments Mi and Mj are developed. The force-displacement
relationship is given by the following equation:
Fi
36
3L
36
3L
vi
2
2
Mi
T
4L
3L L
3L
i or FG = KG v
Fj = 30L 36 3L
36
3L vj
Mj
3L L2 3L 4L2
j
The well-known elastic force deformation relationship, for a prismatic beam
without shearing deformations, is
Fi
12
6L
12
6L
vi
2
2
Mi
EI
4L
6L 2L
6L
i or FE = KE v
Fj = L3 12 6L
12
6L vj
Mj
6L 2L2 6L
4L2
j
Therefore, the total forces acting on the beam element will be:
FT = FE + FG = [KE + KG ]v = KT v
17
Conclusion
So in the case of Constant dead loads where T is usually constant
the calculation of the large displacement effect for cable and beam
elements is performed by the addition in the FEM code of an
appropriate extra geometric stiffness term.
18
Introduction to Dynamic Analysis
I
Introduction
dSwiss Federal
i
Institute of Technology
Page 3
The very basics - Newtons 2nd law of motion:

The
k Unstretched p
position
I very
Introduction
t d basics
ti
to
t Dynamic
D
i Analysis
A l i
k
Newtons 2nd law of motion:
mx = w k ( + x)
mx The
w +very
k ( +basics
x) = 0
Stretched position
w = mg
Newtons 2nd law of motion:
mx = w k ( x)
mx k + k ( + x) = mx + kx = 0
k
(2 +
x) = 0

x + mxx=w
x++k
nx =0
m
mx k k
x mx kx
I
Introduction
d i
m
w = mg
x x ,
x
m
n = 2 =
= 2
k
n
fn =
1
2
(static equilibrium)
Unstretched
position
p
Stretched position
(static equilibrium)
x x ,
x
k
m
0
x = A=sinAsin
nt t nx t=
x sin
n=
t + x0 x0 cos
nnt t + x0 cosn t
k n t + B
x(t)
tcos
+
x(t)
sin
2 Bcos
n
2
m

x
+
x
=
x
+
x
=
0
= 2
=
= 2
n
n
ethod of Finite Elements II
2
n =
=
where
T
k
m
fn =
1
2
k
m
x = A sin nt + B cos nt x = x sin nt + 0 x cos nt
19

We have previously considered the equilibrium equations governing
the linear dynamic response of a system of finite elements:
+ CU
+ KU = R
MU
FI (t) + FD (t) + FE (t) = R(t)

where:
where:
M:
Mass Matrix
FI (t) = MU
Inertial Force
K:
Stiffness Matrix
FD (t) = CU
Damping Force
U:
Displacements
FE (t) = KU
Internal Force
:
U
Velocities
:
U
Accelerations
20
When is Dynamic Analysis required?

Whether a dynamic analysis is needed or not is generally up to
engineering judgment
requires understanding of the interaction between loading and
structural response!
In general if the loading varies over time with frequencies higher
than the Eigen-frequencies of the structure 7 dynamic analysis will
be required
21

Objective
Solve the dynamic Equation of motion (numerically)
+ CU
+ KU = R
MU
In principle the equilibrium equations may be solved by any standard
numerical integration scheme BUT!
Efficiency numerical efforts must be considered and it is worthwhile
to look at special techniques of integration which are especially
suited for the analysis of finite element assemblies.
*The section on Direct Integration Methods is based on
Prof. M. Fabers notes of the FEM II course - Fall 2009
22
Direct Integration Methods
Direct means: The equations are solved in their original form

Two ideas are utilized
1. The equilibrium equations are satisfied only at time steps, i.e. at
discrete times with intervals t
2. A particular variation of displacements, velocities and
accelerations within each time interval is assumed
The accuracy depends on these assumptions as well as the choice of
time intervals!
23
Assumptions
The displacements, velocities and accelerations
U0 :
Displacement vector at time t
0:
U
Velocity vector at time t
0 :
U
Acceleration vector at time t
are assumed to be known and we aim to establish the solution of the

equilibrium equations for the period 0 - T .
For this purpose we sub-divide T into n intervals of length
t = T /n and establish solutions for the times t, 2t, 3t,..., T
24

We distinguish principally between Implicit and Explicit methods
Explicit methods:
Solution is based on the equilibrium equations at time t
1st order Example: The Forward Euler Method
Assuming we want to approximate the solution of the initial
value problem
y = f (t, y(t)),
y(t0 ) = y0
by using the first two terms of the Taylor expansion of y, which

represents the linear approximation around the point (t0 , y(t0 ))
we obtain the forward Euler integration rule:
yn+1 = yn + hf (tn , yn )
25
Implicit methods:
Solution is based on the equilibrium equations at time t + t
1st order Example: The Backward Euler Method
Assuming we want to approximate the solution of the same
initial value problem the backward Euler integration rule is
obtained as:
yn+1 = yn + hf (tn+1 , yn+1 )
Since f (t, y) is, in general, a non-linear function of y, iteration
is required to solve this last equation for yn+1 .
26
Most Commonly used Direct Integration Methods

(for the case of the Dynamic Equation of Motion)
The central difference method
The Houbolt method
The Wilson method
The Newmark method
Coupling of integration operators
27
The Central Difference Method

Approximate the velocity (first derivative) as:
t
= 1 (t+t U t-t U)
U
2t
(1)
Approximate the acceleration (second derivative) as:

t
= 1 (t+t U 2t U + t-t U)
U
t2
(2)
+ CU
+ KU = R:
Substitute Eqns (1),(2) into MU

1
1
M+
C
t2
t2
t+t

2
1
1
t
t-t
U = tR K
M
U
C
U
t2
t2
2t
(a0 M + a1 C) t+t U = t R (K a2 M) t U (a0 M a1 C) t-t U

28
We see that we do not need to factorize the stiffness matrix

(explicit method)
We also see that in order to calculate the displacements at time
t we need to know the displacements at time 0 and t
are known and we can use Eqns (1),(2)
In general, 0 U , 0 U , 0 U
t
to obtain
U:
-t
2
+ t 0 U
U = 0 U t0 U
2
29

Solution Procedure
A. Initial Calculations
30
Di Central
Direct
t Integration
I t Difference
ti Methods
M Method
th d
The
Solution Procedure
Th Central
The
C
l difference
diff
method:
h d Solution
S l i
procedure:
d
B.
B: For
Foreach
eachtime
timestep
step
1) Calculate effective loads at time t:

t
= t R ( K a M ) t U ( a M a C ) t t U
R
2
0
1
2) Solve for the displacements U at time t + t
LDLT t +t U = t R
3) If required, solve for the corresponding velocities and
accelerations
t
= a ( t t U 2 t U + t +t U)
U
0
U = a1 (
t t
U + t +t U)

31
Di Central
Direct
t Integration
I t Difference
ti Methods
M thMethod
d
The
The
The effectiveness
Central difference
methoddifference method depends on the
of the central
efficiency of the time step solution since generally a small
discretization
is required.
The effectiveness
of the central difference method depends on the
efficiency of the time step solution because we need a lot of
For
this reason the method is usually only applied when a lumped
them
(diagonal)
mass the
matrix
assumed
and
when
F this
For
thi reason
th
method
thcan
d is
ibeusually
ll only
l applied
li dthe
when
hvelocity
a lumped
l
d
mass matrix
can be(C)
assumed
when thei.e.:
velocity dependent
dependent
damping
can beand
neglected,
damping can be neglected, i.e.:
1
M t +t U = t R
t 2
2
t
R = t R K 2 M t U 2 M t t U
1
t
R = t R t F 2 M ( t t U 2 t U)
t
i
(i )
t
U i = t Ri (
), mii > 0
mii
2
t +t
K t U = K (i ) t U = t F
i
(i )
32
Di t Integration
Direct
I t
ti Methods
M th d
Example - 2 DOF system
k1 = 4
m1 = 2
U1 ,U1 ,U1
k2 = 2
m2 = 1
U 2 ,U 2 ,U2
R1 = 0
2 0 U1 6 2 U1 0
0 1 + 2 4 U = 10
U 2
2
R2 = 10
k3 = 2
For this system the natural periods are T1 = 4.45, T2 = 2.8

33

34
Di t IIntegration
Direct
t
ti Methods
M th d
2 0 U1 6
0 1 + 2
U 2
The Central difference method: Example:

a0 =
1
,
t 2
a1 =
1
,
2t
a2 = 2a0 ,
a3 =
1
a2
1
1
= 12.8, a1 =
= 1.79,
(0.28) 2
2 0.28
1
1
a2 = 2
= 25.5, a3 =
= 0.0392
2
(0.28)
25.5
a0 =
For t = 0.28
0
0
0 0
Ui = 0.28
0 28 + 00.0392
0392 =
0
0
10 0.0392
2 0
0 0 25.5 0
= 12.8
M
12 8
+
1.79
1
79
0 0 = 0 12.8
0 1
2 t
25.5 0 t t
= 0 + 45.0
R
U
10 2
U
21.5

0 12.8

35

Page 18
Di t IIntegration
Direct
t
ti Methods
M th d
2 0 U1 6 2 U1 0
0 1 + 2 4 U = 10
U 2
2
The Central difference method: Example:
The equation which must be solved for each time step is:
The equation which must be solved for each time step is:
25.5 0 U1 t
0 12.8 t +t = R
U 2
t +t
45 0
2 tU1 25
25.5
5 0 U1
= 0 + 45.0
R
t
10 2
21.5 U 2 0 12.8 t tU

2
t t
36

The results are:
6
CentralDifference_U1
4
3
2
1
0
0
0.5
1.5
2.5
3.5
37
TheSwiss
Houbolt
Method
Federal Institute of Technology
Page 20
Di t Integration
Direct
I Method
t
tiDerivative
Methods
M th Approximation
d
Houbolt
+ CU
+ KU = R
MU
= 1 (2 t +t U 5 t U + 4 t t U t 2 t U)
U
t 2
1
t +t
(11t +t U 18 t U + 9 t t U 2 t 2 t U)
U=
6t
t +t
a
b
a and b inserted in c
12
2
C + K t +t U =
2 M+
6t
t
3
3 t t
1 t 2 t
5
4
1
t +t
R + 2 M + C t U 2 M +
C U + 2 M +
C
U
t
t
t
t
2
t
t
3
38
The
Method
Di Houbolt
Direct
t IIntegration
t
ti Methods
M th d
We
Th will
The
Houbolt
H not
b lconsider
method
h d
the Houbolt in more detail however it is noted
that it is necessary to factorize the stiffness matrix
We will not consider the Houbolt in more detail however it is
(implicit
method)
noted that
it is necessary to factorize the stiffness matrix (implicit
method)
Furthermore, if the mass and damping terms are neglected, the

Furthermore,
if the
mass
andstatic
damping
terms
are neglected, the
Houbolt
method
results
in the
analysis
equations:
Houbolt method results in the static analysis equations
12
2
C + K t +t U =
2 M+
6t
t
3
3 t t
1 t 2 t
5
4
1
t +t
R + 2 M + C t U 2 M +
C U + 2 M +
C
U
t
t
t
t
t
2
3
39
Di t Integration
Direct
I t
ti Methods
M th d
The
Wilson
Method
Th Wilson
The
Wil
method
h d
In this method the acceleration is assumed to vary linearly from
time t to t + t
t +
= t U
+ ( t +t U
t U
)
U
t
By integration we obtain
t +
t +
2
= tU
+ tU
+ ( t +t U
t U
)
U
2t
+ 1 tU
2 + 1 3 ( t +t U
t U
)
U = tU+ tU
2
6t
40
The Wilson
Th Wilson
The
Wil Method
method
h d
Setting = t we get
t +t
t +t
= tU
+ t ( t +t U
+ t U
)
U
2
t + 1 (t )2 ( t +t U
+ 2 t U
)
U = tU+ tU
6
from
which we can solve
Page 24
t +t
t +t
t
) d6 t t U 2 t U
Di U t= Integration
Direct
I(tt )2 ( ti U Methods
M Uth
3 t +t
t +t
=
( Uh tdU ) 2 t U 2 t t U
Th UWilson
The
Wil
t method
t +t
t +t
=
U
(t )
t +t
U t U)
2t U

U
= 3 ( t +t U t U ) 2 t U
t t U

U
2
t
Method of We
Finitenow
Elements
II
solve
for the displacements, velocities and accelerations

by inserting into the dynamic equilibrium equation
+ C t +t U
+ K t +t U =
M t +t U
t +t
t +t
R = t R + ( t +t R t R )
41
The Wilson Method

Solution Procedure
A. Initial Calculations
Implicit Procedure!
42
The
Di Wilson
Direct
t Integration
I t Method
ti Methods
M th d
Solution Procedure
Th Wilson
The
Wil
method
h d : Solution
S l i
procedure:
d
B. For each time step

B: For each time step
1) Calculate effective loads at time t+t:

t +t
= t R + ( t +t R t R ) + M (a t U + a t U
+ 2t U
)
R
0
2
t
t
t
+C(a U + 2 U + a U)
1

LDLT t +t U =
t +t
3) Solve for the corresponding velocities and accelerations

= a ( t +t U t U ) + a t U
+a tU

U
4
5
6
t +t
t
t +t
t
U = U + a7 ( U + U)
t +t
+ a ( t +t U
+ 2 t U
)
U = t U + t t U
8
t +t
43
The Newmark Method
Th Newmark
The
N
k method
h d
The Newmark method may be seen as an extension of the Wilson
This
method
method may be seen as an extension of the Wilson method
t +t
t +t
= tU
+ (1 ) t U
+ t +t U
t
U
t + ( 1 ) t U
+ t +t U
t 2
U = tU+ tU
2
and are parameters which may be adjusted to achieve

accuracy and stability
and a are parameters which may be adjusted to achieve
=0.5,
= 1/6and
corresponds
accuracy
stability to the linear acceleration method which
also correspond to the Wilson method with = 1
= 0.5, a = 1/6 corresponds to the linear acceleration method

Newmark
Ne
ma k also
originally
o iginall
proposed
p oposed
=0.5,
05
=method
1/4 which
hich
results
es
lts1 in an
which
correspond
to the Wilson
with
=
unconditionally stable scheme (the trapetzoidal rule)
Newmark originally proposed =0.5, a = 1/4 which results in an

unconditionally stable scheme (the trapezoidal rule)
44
The Newmark Method

We now solve for the displacements, velocities and accelerations by
inserting the above into the dynamic equilibrium equation:
where
45
The Newmark Method

Solution Procedure
A. For each time step
46
Di Newmark
Direct
t IIntegration
t
tiMethod
Methods
M th d
The
Solution Procedure
The Newmark method : Solution procedure:
B. For each time step

B: For each time step
1) Calculate effective loads at time t:
t +t
+a tU
)
R = t +t R + M (a0 t U + a2 t U
3
+a tU
)
+C(a t U + a t U
1
LDLT t +t U = t +t R
3) Solve for the corresponding velocities and accelerations
= a ( t +t U t U) a t U
a tU

U
0
2
3
t +
t
t
t +
t
t
U = U+a
U+a U
t +t
47
The Newmark Method
Example
6
Newmark_U1
Newmark_U2
k1 = 4
m1 = 2
U 1 ,U1 ,U1
k2 = 2
m2 = 1
2
U 2 ,U 2 ,U2
R1 = 0
R2 = 10
k3 = 2
0
0
0.5
1.5
2.5
3.5
48
Di t Integration
Direct
I t
ti Methods
M th d
The Newmark
Method
Example
The Newmark method
m1 = 2
U 1 ,U1 ,U1
U 2 ,U 2 ,U2
Newmark_U2
R2 = 1
k3 = 2
t = 28
Newmark_U1
_
k2 = 2
m2 = 1
: Example:
R1 = 0
4
3
= 0
U
10

2
1
0
0
05
0.5
15
1.5
25
2.5
35
3.5
3.5
Newmark_U1
t = 28
Newmark_U2
2.5
2
1.5
= 0
U
0

1
0.5
0
0 Elements
0.5
Method of Finite
II
1.5
2.5
3.5
49
Coupling of integration operators

For some problems it may be an advantage to combine the different
types of integration schemes e.g. if a structure is subjected to
dynamic load effect from hydrodynamic loading then the analysis of
the hydrodynamic forces may be assessed using an explicit scheme
and the structural response by using an implicit scheme.
The best choice of strategy will depend on the problem in regard
stability and accuracy!
50

State Space Equation Formulation

2dof Mass Spring System
F1 ( t )
k1
c1
k2 ( x2 x1 )
m1
m2
c2 ( x2 x1 )
c1 x1
c1
x1 ( t )
F2 ( t )
k2 ( x2 x1 )
k1 x1
m2
m1
F1 ( t )
FBD
F2 ( t )
k2
c2 ( x2 x1 )
(Lumped Mass System)
x2 ( t )
We introduce the augmented state vector:

T
x = x1 x2 x 1 x 2
(controllable form equivalent). Then,
0
0
1
0
x1
0
x2
0
0
1
x =

x 1

m1 k
m1 c
x 2
0 0

0 0 F1
+
1 F2
m

2dof Mass Spring System
x = Ax + Bu
where u =
F1 F2
T
Assume you would like to monitor the displacement x1 , x2 . Then

the observation vector is:
1
0
y=
0
0
0
1
0
0
0
0
0
0
x1
0
0 x2
0 x3
0
x4
+ O42 u(t)
y = Cx + Du

Using the state space representation we have converted a 2nd order ODE
into an equivalent 1st order ODE system. We can now use any 1st order
ODE integration method to convert the continuous system into a
discrete one and obtain an approximate solution:
1st order ODE Integration Methods
dy
Assume
= f (t, y(t)), y(t0 ) = 0
dt
Forward Euler Method
yn+1 = yn + hf (tn , yn )
where h is the integration time step. This explicit expression is
obtained from the truncated Taylor Expansion of y(tn + h)
Backward Euler Method
yn+1 = yn + hf (tn+1, yn+1 )
This implicit expression (since yn+1 is on the right hand side) is
obtained from the truncated Taylor Expansion of y(tn+1 h)

2nd Order Runge Kutta (RK2)
1
1
k2 = hf (tn + h, yn + k1 )
2
2
= yn + k2 + O(h3 )
k1 = hf (tn , yn ),
yn+1
4th Order Runge Kutta (RK4)

1
1
k2 = hf (tn + h, yn + k1 )
2
2
1
1
k3 = hf (tn + h, yn + k2 ), k4 = hf (tn + h, yn + k3 )
2
2
1
1
1
1
yn+1 = yn + k1 + k2 + k3 + k4 + O(h5 )
6
3
3
6
k1 = hf (tn , yn ),
Mode Superposition Method

Assuming diagonal mass matrix and no damping, implicit methods
require on average 2nmk calculations per time step
where n is the the dimension of the stiffness matrix and mk the
half-bandwidth
The central difference method (explicit scheme) requires significantly
fewer operations per time step however has other drawbacks (requires
T < Tcr more time steps!)
The total number of operations in a Direct Integration Method are
directly proportional to the number of load steps only effective for
short durations (small number of time steps)
Mode Superposition Methods
These methods work by transforming the equilibrium equations into a
form in which step by step solution is less costly
In particular we aim at a reduction of mk which would proportionally
decrease the step by step solution

Equilibrium Equations Transformation - Change of Basis
The following transformation is used on the n finite element nodal
point displacements U:
U(t) = PX(t)
where P is an n n square matrix and X(t) is the generalized
displacement vector (time dependent)
The dynamic equation of motion then can be written as:
X(t)
X(t)
M
+C
+ KX(t)
= R(t)
= PT MP C
= PT CP K
= PT KP R
= PT R
where M
are the transformed matrices corresponding to a smaller bandwidth
than the original system.

Choice of Transformation Matrix P
P is established using the displacement solution for Undamped, Free
vibrations:
+ KU = 0
MU
(1)
The solution of this 2nd Order ODE is of the form:

U = sin(t t0 )
where is a vector of order n, t is the time variable, t0 is a time
constant corresponding to the phase and is the vibration frequency
of vector , then
= 2 sin(t t0 )
U
and (1) becomes:
2 M + K = 0
(eigenproblem)

Matrix Form
Defining a matrix whose columns are the eigenvectors i and a
diagonal matrix2 which stores the eigenvalues i2 on its diagonal,
i.e:

2
;
= 1 , 2 , ... n ; 2 =
...
n
we can write the n solutions to the eigenproblem as:
M2 = K
Properties of Eigenvectors
Solution of the Eigenproblem
2 M + K = 0
(eigenproblem)
The above homogeneous linear system of equations can only have a

solution if the determinant is equal to 0:
2

M + K = 0
The solution of the above equation will yield:
n eigenvalues i2 , i = 1, ..., n
with 0 1 ...n (the eigenfrequencies) and
The solution of 2 M = K (eigenproblem) will yield:
n eigenvectors (or modal vectors) i
10
M-Orthonormality
From Matrix Properties we know that (AB)T = B T AT ,
Thus, for two eigenvectors n , r we obtain:
(Tn Kr )T = Tr KT n
Since the Stiffness matrix is symmetric, K = KT , hence
(Tn Kr )T = Tr Kn
The same applies for the Mass matrix, M = MT , yielding:
(Tn Mr )T = Tr Mn
11
M-Orthonormality
The eigenproblem for vector n then can be written as:
r2 Tn Mr = Tn Kr
transpose
r2 Tr Mn = Tr Kn
and the eigenproblem for vector n is: The eigenproblem for vector n
then can be written as:
n2 Tr Mn = Tr Kn
By subtracting the two previous formulas we obtain:
(r2 n2 )Tr Mn = 0
We conclude that for n 6= r
Tr Mn = 0,
Tr Mn = 0,
12
M-Orthonormality
We choose n , such that:
T M = I
therefore from M2 = K we have that:
T K = 2
The principle of M-orthonormality can

1,
T
r Mn =
0,
2
n ,
Tr Kn =
0,
then be written as:

n=r
n 6= r
n=r
n 6= r
13

It is now obvious that we can use as the transformation matrix P.
Using:
U(t) = X(t)
we obtain the transformed equilibrium equation:
X(t)
+ T CX(t)
+ 2 X(t) = T R(t)
where using the property of M-orthonormality, the initial conditions
will be:
0
X = T M0 U;
= T M0 U
X
14
Dynamic Response with Damping Neglected

If we neglect the velocity dependent damping effects the equilibrium
equation reduces to:
+ 2 X(t) = T R(t)
X(t)
i.e n individual equations of the form (since 2 is diagonal):

x
i (t) + i2 xi (t) = ri (t)
i = 1, 2, ..., n
ri (t) = Ti R(t)
with
0
xi = Ti M0 U;
x i = Ti M0 U
15

SDOF Response
Each equation of the previous system describes a single degree of freedom
system with unit mass and stiffness i2 . The solution to this equation for a
random input excitation can either be obtained by using the direct
integration methods or by using the Duhamel Integral:
Z t
1
ri ( )sini (t ) d + i sini t + i cosi t
xi (t) =
i 0
where i , i are determined from the initial conditions. Therefore, the
SDOF response id owed to two contributions
A dynamic (steady - state) response obtained by multiplying the static
response by a dynamic load factor (this is the particular solution of
the governing differential equation), and
An additional dynamic response called the transient response
16
Complete Response
The solution of all n SDOF equations are calculated and the finite
element nodal point displacements are obtained by superposition of
the response in each mode:
U(t) = X(t) U(t) =
n
X
i xi (t)
i=1
Therefore the solution scheme is:

Solve for the eigenvalues and eigenvectors of problem
Solve for the response of the decoupled SDOF equations
Use superposition to find the total response.
17
Superposition Principle
An alternative View
Superposition
AdynamicloadcanbedesignedasFourierseries
A dynamic
load can be designed as
Fourier series of harmonic sine and cosine
contributions
Thetotalsolutionofsuchaproblemisequalto
The total
solution of such a problem is
thesuperpositionofsolutionoftheFourierterms.
equal to the superposition of solution of
the Fourier terms.
U(t) = X(t) U(t) =
n
X
i xi (t)
i=1
13.11.2009
ModeSuperposition
11
18
The effect of Damping

The presence of damping reduces the dynamic load factor (which then
cannot be infinite) and damps out the transient response
Problemswithneglecteddampi
The response in the
modes with
large is
negligible
For
close to zero the
system follows the loads

statically
Effectively only the first p modes need to be used p n, in order to obtain
a good approximate solution Advantage over Direct Integration Methods
13.11.2009
ModeSuperposition
19
Pag
Di t Integration
Direct
I t
ti Methods
M th d
Calculate the displacement response of the system
k1 = 4
m1 = 2
U1 ,U1 ,U1
k2 = 2
m2 = 1
U 2 ,U 2 ,U2
R1 = 0
2 0 U1 6 2 U1 0
0 1 + 2 4 U = 10
U 2
2
R2 = 10
k3 = 2
20

Eigenproblem Setup
n2 Mn
= Kn
n2
2, 0
0 1

=
2
4
6,
2

n
Eigenvalue Calculation

2

K n2 M = 0 6 2n ,

2
196 160
=
= 2,
4
Eigenvector Calculation
12
14

2
4
2

2 = 2n 14n + 20 = 0
4 1n

K 12 M 1 = 0
2,
2
2
2

22
11
12
14 +
196 160
=5
4

=0
11
12

=
1
1
21

Use of M - orthonormality
1
3
2 3
2
T1 M1 = 1 2 (m1 + m2 ) = 1 =
Hence,
3
1 = 1
Similarly, 2 =
Transformed Equilibrium Equation
X(t)
+ 2 X(t) = T R(t)
1

2, 0
3
X(t)
+
X(t) =
2
0 5
2 3
1

3 0
2 10
22

Decoupled SDOF equations
10
x
1 + 2x1 =
3
r
x
2 + 5x2 = 10
2
3
Initial Conditions
U(0) = 0
U(0)
=0
x1 (0) = 0,
and
xi (0) = Ti MU(0)
x i (0) = T MU(0)
i
x 1 (0) = 0,
x2 (0) = 0,
x 2 (0) = 0
Then the exact solutions to the ODEs are:

r
5
2
x1 = (1 cos 2t) x2 = 2
(1 + cos 5t)
3
3
And the total solution is:
U(t) =
n
X
i xi (t)
i=1
23

Comparison of Direct Integration and Modal Superposition
Methods
24
Note
In this case we dealt with a 2DOF example and used an analytical
solution to solve the two decoupled SDOF system equations
In practice for multi degree of freedom systems, higher modes are
neglected and instead of an analytical solution we again use a Direct
Integration scheme to solve for each SDOF ODE
In essence the finally achieved accuracy might be the same between
the two approaches, however teh mode superposition method usually
requires for less computational cost
25
Analysis Including Damping

Transformed Equilibrium Equation
X(t)
+ T CX(t)
+ 2 X(t) = T R(t)
Proportional Damping Assumption
Ti Cj = 2i i ij
(2)
where i is a modal damping parameter and ij is the Kronecker delta

(ij = 1 for i = j, ij = 0 for i 6= j)
26
Therefore, we still end up with Decoupled SDOF equations for

each xi :
x
i (t) + 2i i x i (t) + i2 xi (t) = ri (t)
with the Duhamel Integral now being:
Z
1 tn
xi (t) =
ri ( ) ei i (t ) sin
i (t ) d
i 0
o
+ ei i t (i sin
i t + i cos
i t)
where
i = i
q
1 i2
27

Rayleigh Damping
If there are only two different damping ratios i ,
Damping can be used:
i = 1, 2 Rayleigh
C = M + K
(3)
Eqns (2), (3) now yield:

Ti (M + K)i = 2i i
+ i2 = 2i i
The 2 2 system can be solved to obtain , .
In actual analysis it may well be that the damping ratios are known
for many more than two frequencies. In that case two average values
say 1 , 2 are used to evaluate , .
28
Measuring Damping
Logarithmic decrement method can be used to measure damping in time
domain. In this method, the free vibration displacement amplitude history of a
system to an impulse is recorded. Logarithmic decrement is the natural
logarithmic value of the ratio of two adjacent peak values of displacement in free
decay vibration.
29
Example - Damping for an MDOF system

Assume that the approximate damping to be specified for a multiple degree of
freedom system is as follows:
30
Example - Damping for an MDOF system

Damping as function of frequency
A problem with Rayleigh

damping is that higher
modes are much more
damped than lower modes.
However, in general
Rayleigh damping provides
a good approximation
31

Cours

Uploaded by

Document Information

Original Description:

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

Cours

Uploaded by

Copyright:

Available Formats

The Finite Element Method for the Analysis of

Non-Linear and Dynamic Systems

Institute of Structural Engineering

Method of Finite Elements II

Method of Finite Elements II

Overview of Solution Methods

Dynamic Finite Element Calculations

Method of Finite Elements II

Performance Evaluation - Homeworks (100%)

Institute of Structural Engineering

Method of Finite Elements II

Review of the Finite Element Method (FEM)

FEM: Numerical Technique for approximating the solution of continuous

Method of Finite Elements II

Review of the Finite Element Method (FEM)

Axial bar Example

Review of the Finite Element Method (FEM)

Boundary Condition Classification

Essential (Dirichlet): u(x0 , y0 ) = u0

Review of the Finite Element Method (FEM)

Institute of Structural Engineering

Method of Finite Elements II

Strong Form - 1D FEM

Physical Problem (1D)

One dimensional Heat

Axially Loaded Bar

Institute of Structural Engineering

Method of Finite Elements II

Weak Form - 1D FEM

Method of Finite Elements II

Weak Form - 1D FEM

From Strong Form to Weak form

Institute of Structural Engineering

Method of Finite Elements II

Weak Form - 1D FEM

Method of Finite Elements II

Weak Form - 1D FEM

Method of Finite Elements II

Weak Form - 1D FEM

S = {u|u C 0 , u(l) = 0}, S 0 = {w |w C 0 , w (l) = 0}

w (x)[(c(x)u 0 (x))0 + f (x)]dx = 0

Institute of Structural Engineering

Method of Finite Elements II

Weak Form - 1D FEM

The weak form is then reduced to the following problem. Also, in

Method of Finite Elements II

Natural (Neumann) boundary conditions, are imposed on the

Essential (Dirichlet) or geometric boundary conditions, are imposed

Method of Finite Elements II

Institute of Structural Engineering

Method of Finite Elements II

1D FE formulation: Galerkins Method

Method of Finite Elements II

1D FE formulation: Galerkins Method

Bounded and Continuous

x, things become simplified when

Method of Finite Elements II

1D FE formulation: Galerkins Method

i.e. for 2 nodes:

Alternatively we could have a Petrov-Galerkin formulation, where w (x) is

Institute of Structural Engineering

Method of Finite Elements II

1D FE formulation: Galerkins Method