1

CHAP 1
Preliminary Concepts and
Linear Finite Elements

Instructor: Nam-Ho Kim (nkim@ufl.edu)

Web: http://www2.mae.ufl.edu/nkim/INFEM/
2
 Table of Contents
1.1.INTRODUCTION

1.2. VECTOR AND TENSOR CALCULUS

1.3. STRESS AND STRAIN

1.4. MECHANICS OF CONTINUOUS BODIES

1.5. FINITE ELEMENT METHOD

3
1.1

INTRODUCTION

4
 Background
Finite Element Method (FEM):
a powerful tool for solving partial differential equations and
integro-differential equations
Linear FEM:
methods of modeling and solution procedure are well established
Nonlinear FEM:
different modeling and solution procedures based on the
characteristics of the problems complicated
many textbooks in the nonlinear FEMs emphasize complicated
theoretical parts or advanced topics
This book:
to simply introduce the nonlinear finite element analysis procedure
and to clearly explain the solution procedure
detailed theories, solution procedures, and implementation using
MATLAB for only representative problems

5
 Chapter Outline
2. Vector and Tensor Calculus
Preliminary to understand mathematical derivations in other
chapters
3. Stress and Strain
Review of mechanics of materials and elasticity
4. Mechanics of Continuous Bodies
Energy principles for structural equilibrium (principle of minimum
potential energy)
Principle of virtual work for more general non-potential problems
5. Finite Element Method
Discretization of continuum equations and approximation of
solution using piecewise polynomials
Introduction to MATLAB program ELAST3D

6
1.2

VECTOR AND TENSOR

CALCULUS

7
 Vector and Tensor
Vector: Collection of scalars
Cartesian vector: Euclidean vector defined using Cartesian
coordinates y
2D, 3D Cartesian vectors u2

u1
u1
u , or u u2 e2
e1 x
u2 u
3 e3 u1
u3
z

Using basis vectors: e1 = {1, 0, 0}T, e2 = {0, 1, 0}T, e3 = {0, 0, 1}T

u u1 e1 u2e2 u3e3

8
 Index Notation and Summation Rule
Index notation: Any vector or matrix can be expressed in
terms of its indices
v1 A11 A12 A 13
v [vi ] v2 A [Aij ] A21 A22 A 23
v3 A31 A32 A 33
Einstein summation convention
3
akbk akbk Repeated indices mean summation!!
k 1
In this case, k is a dummy variable (can be j or i) akbk ajbj
The same index cannot appear more than twice
Basis representation of a vector
Let ek be the basis of vector space V
N
Then, any vector in V can be represented by w wk ek wk ek
k 1 9
 Index Notation and Summation Rule cont.
Examples
Matrix multiplication: C A B Cij AikBkj
Trace operator: tr(A) A11 A22 A33 Akk
Dot product: u v u1v1 u2v2 u3v3 uk vk
Cross product: u v ujvk ( ej ek ) eijkujvk ei

0 unless i, j,k are distinct

Permutation
eijk 1 if (i, j,k) is an even permutation
symbol
1 if (i, j,k) is an odd permutation

Contraction: double dot product

3 3
J A:B AijBij AijBij
i1 j1

10
 Cartesian Vector
X3
Cartesian Vectors u
u u1 e1 u2 e2 u3e3 uiei v e3
v vjej e1 e2
X2

X1
Dot product
u v (uiei ) (vjej ) uv
i j ( ei e
j ) uivj ij uivi
Kronecker delta function

1 if i j
ij jj 11 22 33 3
0 if i j
Equivalent to change index j to i, or vice versa
How to obtain Cartesian components of a vector

ei v ei (vjej ) vj ij vi Projection
Magnitude of a vector (norm):
v v v
11
 Notation Used Here

Direct tensor notation Tensor component notation Matrix notation

a b ab
i i aT b
A ab Aij ab
i j A abT
b A a bi Aijaj b Aa
b a A bj aA
i ij bT aT A

12
 Tensor and Rank
Tensor
A tensor is an extension of scalar, vector, and matrix
(multidimensional array in a given basis)
A tensor is independent of any chosen frame of reference
Tensor field: a tensor-valued function associated with each point
in geometric space
Rank of Tensor
No. of indices required to write down the components of tensor
Scalar (rank 0), vector (rank 1), matrix (rank 2), etc
Every tensor can be expressed as a linear combination of rank 1
tensors
Rank 1 tensor v: vi 11 12 13
Rank-2
Rank 2 tensor A: Aij
[ij ] 21 22 23 stress
tensor
31 32 33
Rank 4 tensor C: Cijkl 13
 Tensor Operations
Basic rules for tensors
Different notations
( TS)R T(SR) TS T S
T(S R) TS TR
Identity tensor
( TS) ( T )S T(S) 1 [ij ]
1T T1 T
Tensor (dyadic) product: increase rank
A u v uivjei ej Aij uivj AT Ajiei ej
( u v ) w u( v w)
w ( u v ) v ( w u)
( u v )( w x) ( v w)u x uv vu

Rank-4 tensor: D Dijklei ej ek el

14
 Tensor Operations cont.
Symmetric and skew tensors
Symmetric S ST
Skew W WT
Every tensor can be uniquely decomposed by symmetric and
skew tensors
T SW
S 21 ( T T T )
W 1 (T TT )
Note: W has zero diagonal
2 components and Wij = - Wji

Properties Let A be a symmetric tensor

A:W 0
A:T A:S
15
 Example
Displacement gradient can be considered a tensor (rank 2)
u1 u1 u1
X1 X2 X3
u u2 u2 u2
u X3
X X1 X2
u3 u3 u3
X1 X2 X3

u1 1 ( u1 u2 1 ( u1 u3
X1 2 X2
X1
) 2 X3
)
X1

sym(u) 1 ( u1 u2 u2 1 ( u2 u3 Strain tensor
2 X2
X1
) X2 2 X3
)
X2
1 ( u1 u3 1 ( u2 u3 u3
2 X3
X1
) 2 X3
X2
) X3
1 ( u1 u2 1 ( u1 u3
0 2 X2
X1
) 2 X3
)
X1

skew(u) 21 ( X1
u u2 1 ( u2 u3
X1
) 0 2 X3
)
X2
Spin
2
1 u1 tensor
u3 u u3
2 ( X3 X1
) 21 ( X2 X2
) 0
3
16
 Contraction and Trace
Contraction of rank-2 tensors
a : b aijbij a11b11 a12b12 K a32b32 a33b33
contraction operator reduces four ranks from the sum of ranks of
two tensors
magnitude (or, norm) of a rank-2 tensor
a a:a
Constitutive relation between stress and strain
D : , ij Dijklkl
Trace: part of contraction
tr(A) Aii A11 A22 A33
In tensor notation

tr(A) A : 1 1 : A
17
 Orthogonal Tensor
In two different coord.
e3 e3*
e1 *
u uiei uj* ej*
e2
Direction cosines e1
e2*
[ij ] [ ei* ej ] ei* ijej
We can also show
Change basis ej ijei* u* u
u ujej ui* ei* u T u* T ( u) (T )u
ui*ijej
1 T
uj ijui* T T 1 det() 1
Orthogonal tensor
u T u* Rank-2 tensor transformation
T * T T , Tij* ikTkljl
18
 Permutation
The permutation symbol has three indices, but it is not a
tensor

1 if ijk are an even permutation : 123, 231, 312

eijk 1 if ijk are an odd permutation : 132, 213, 321
0 otherwise

the permutation is zero when any of two indices have the
same value: e112 = e121 = e111 = 0
Identity

eijk elmk il jm im jl

vector product

u v eieijkujvk
19
 Dual Vector
For any skew tensor W and a vector u
u Wu u WT u u
Wu
0

Wu and u are orthogonal
0 W12 W13 W 23
Let Wij eijk wk
W W12 0 W23 w W13
W13 W23 0
W 12

Then, W u e w u e w u
ij j ijk k j ikj k j

Wu w u

Dual vector of skew tensor W

wi 21 eijkWjk

20
 Vector and Tensor Calculus
Gradient
( X)

ei
X Xi
Gradient is considered a vector
vi
We will often use a simplified notation: vi , j
X j
Laplace operator
2

ei ej
Xi X Xj Xj
j
Gradient of a scalar field (X): vector
2 2 2

( X) ei X12 X22 X32
Xi

21
 Vector and Tensor Calculus
Gradient of a Tensor Field (increase rank by 1)
i
iei ej ei ej
Xj Xj
Divergence (decrease rank by 2)
i
ei
Xi
jej
Xi

Ex)
jk,jek
Curl

v eieijk vk,j

22
 Integral Theorems
Divergence Theorem

A d n A d n: unit outward normal vector

Gradient Theorem

A d n A d c

Stokes Theorem

n ( v ) d
c r v dc
r
Reynolds Transport Theorem
d A

dt
Ad t d (n v)A d
23
 Integration-by-Parts
u(x) and v(x) are continuously differentiable functions
1D
b b b
a u(x)v(x) dx u(x)v(x) a
u(x)v(x) dx
a

2D, 3D
u v
xi v d uvni d u xi d
For a vector field v(x)

u v d u( v n) d u vd

Greens identity

2
u v d u v n d u

vd

24
 Example: Divergence Theorem
S: unit sphere (x2 + y2 + z2 = 1), F = 2xi + y2j + z2k
Integrate S F n dS
S F n dS F d
2 (1 y z) d

2 d 2 y d 2 z d

2 d

8

3

25
 Quiz
A symmetric rank four tensor is defined by

D 1 1 2I
where 1 = [ij] is a unit tensor of rank-two and
I 21 [ik jl il jk ] is a symmetric unit tensor of rank-
four. When E is an arbitrary symmetric rank-two tensor,
calculate S = D:E in terms of E.

26
1.3

STRESS AND STRAIN

27
 Surface Traction (Stress)
F
Surface traction (Stress) f1

The entire body is in equilibrium with n

external forces (f1 ~ f6) P

A
The imaginary cut body is in equilibrium due f2

to external forces (f1, f2, f3) and internal f3

forces
Internal force acting at a point P
on a plane whose unit normal is n: f6
f1
(n) F f5
t lim
A 0 A
The surface traction depends on the unit f2 f4
normal direction n. z f3
Surface traction will change as n changes.
y
unit = force per unit area (pressure) x

t (n) t1 e1 t2 e2 t3e3
28
 Cartesian Stress Components
Surface traction changes according to the direction of
the surface.
Impossible to store stress information for all directions.
Lets store surface traction parallel to the three
coordinate directions.
Surface traction in other directions can be calculated
from them.
Consider the x-face of an infinitesimal cube
x
t (x)
t1(x) e1 t2(x) e2 + t3(x) e3
z
t (x)
11 e1 12 e2 + 13e3 13
z F

Normal
11 12
Shear
stress stress y
x y 29
 Stress Tensor
First index is the face and the second index is its direction
When two indices are the same, normal stress, otherwise shear
stress.
Continuation for other surfaces.
Total nine components
Same stress components are defined for the negative planes.

Rank-2 Stress Tensor

33
ijei ej x

31 32
Sign convention z
13 23
z
sgn(11 ) sgn(n) sgn( Fx ) 21
11 12 22
sgn(12 ) sgn(n) sgn( Fy ) y
x y

30
 Symmetry of Stress Tensor
Stress tensor should be symmetric
9 components 6 components 21
Equilibrium of the angular moment A B
y
M l(12 21 ) 0 12 l O x 12
12 21
l
Similarly for all three directions: C D
12 21 , 23 32 , 13 31 21

11

22 11 12 13
Lets use vector notation:
Cartesian components 33 [ij ] 12 22 23
of stress tensor { }
12 13 23 33
23

13 31
 Stress in Arbitrary Plane
If Cartesian stress components are known, it is possible to
determine the surface traction acting on any plane.
Consider a plane whose normal is n.
Surface area (ABC = A) y
B
PAB An3 ; PBC An1 ; PAC An2
n
31 33 t(n)
The surface traction 13
11 32
t ( n ) t1( n ) e1 t2( n) e2 t3(n ) e3
12 P 23 A x
21
22
Force balance z C
1 1 A 11An1 21An2 31An3 0
F t (n)

t1( n) 11n1 21n2 31n3

32
 Cauchys Lemma
All three-directions
t1( n) 11n1 21n2 31n3
t2( n) 12n1 22n2 32n3
t3(n ) 13n1 23n2 33n3

Tensor
( n ) notation
t n t ( n) n

stress tensor; completely characterize the state of stress at a

point
Cauchys Lemma
the surface tractions acting on opposite sides of the same surface
are equal in magnitude and opposite in direction
t (n) t ( n)
33
 Projected Stresses
Normal stress (n) t n n n n ijnn
i j
n 2
Shear stress (n) t 2 (n), (n) t n n(n)
Principal stresses tn P n 1 , 2 , 3
Mean stress (hydrostatic pressure)
1 1
p m tr( ) (11 22 33 )
3 3
Stress deviator I 21 ( ik jl il jk )
ijkl
s m 1 Idev : Rank-4 identity tensor
11 m 12 13 Idev I 31 1 1
s 12 11 m 23
Rank-4 deviatoric identity tensor
13 23 11 m
Idev : 1 0, Idev : s s
34
 Principal Stresses
Normal & shear stress change as n changes
Is there a plane on which the normal (or shear)stress becomes the
maximum?
There are at least three mutually perpendicular planes
on which the normal stress attains an extremum
Shear stresses are zero on these planes Principal directions
Traction t(n) is parallel to surface normal n

t ( n) nn n n n

Eigenvalue problem Principal Principal

stress direction
[ n 1] n 0
11 n 12 13 n1 0


12 22 n 23 n2 0
13 23 33 n n
3
0
35
 Eigenvalue Problem for Principal Stresses
The eigenvalue problem has non-trivial solution if and only
if the determinant is zero:
11 n 12 13
12 22 n 23 0
13 23 33 n
The above equation becomes a cubic equation:
n3 I1n2 I2n I3 0
I1 11 22 33
2 2 2
I2 1122 2233 3311 12 23 13
2 2 2
I3 112233 2122313 1123 2213 3312
Three roots are principal stresses
1 2 3
36
 Principal Directions
Stress Invariants: I1, I2, I3
independent of the coordinate system
Principal directions
Substitute each principal stress to the eigenvalue problem to get n
Since the determinant is zero, an infinite number of solutions
exist
Among them,
2 choose the one with a unit magnitude
ni (n1i )2 (n2i )2 (n3i )2 1, i 1,2,3

Principali directions
n n j 0, i are
j mutually perpendicular

37
 Principal Directions
There are three cases for principal directions:
1. 1, 2, and 3 are distinct principal directions are three unique
mutually orthogonal unit vectors.
2. 1 = 2 and 3 are distinct n3 is a unique principal direction, and
any two orthogonal directions on the plane that is perpendicular
to n3 are principal directions.
3. 1 = 2 = 3 any three orthogonal directions are principal
directions. This state of stress corresponds to a hydrostatic
pressure.

n3

38
 Strains (Simple Version)
Strain is defined as the elongation per unit length

u2

x2

P x1 P
u1

u1 u1
11 lim
Tensile (normal) 0 x1 inx
x1 strains x11- and x2-directions
Textbook has different, but
more rigorous derivations
u2 u2
22 lim
x2 0 x2 x2

39
 Shear Strain
Shear strain is the tangent of the change in angle between two
originally perpendicular axes
u1
u2
1 ~ tan 1
x1 2

u1 x2
/2 12
2 ~ tan 2
x2 P
1 u2
x1

u
2 u1 u2 u1
12 strain
Shear 2 lim
1 (change of angle) lim
x1 0 x1 x2 0 x2 x1 x2
1 1 u2 u1
12 12
2 2 x1 x2

Positive when the angle between two positive (or two negative) 40
 Strains (Rigorous Version)
Strain: a measure of deformation
Normal strain: change in length of a line segment
Shear strain: change in angle between two perpendicular line
segments
Displacement of P = (u1, u2, u3)
Displacement of Q & R
Q u1 u1
u1 u1 x1 u1 u1
R
x2
x1 x2
u2
R'
Q u
u2 u2 x1 u2R u2 2 x2
x1 x2 Q'

Q u3 u3 R P'(x1+u1, x2+u2, x3+u3)

u3 u3 x1 u3 u3
R
x2 x2
x1 x2 x 2 x1
P(x1,x2,x3) Q

x1

x3 41
 Displacement Field
Coordinates of P, Q, and R before and after deformation
P : (x1 , x2, x3 )
Q : (x1 x1, x2, x3 )
R : (x1, x1 x2, x3 )
P : (x1 u1P , x2 u2P , x3 u3P ) (x1 u1, x2 u2, x3 u3 )
Q : (x1 x1 u1Q , x2 u2Q , x3 u3Q )
u1 u u
(x1 x1 u1 x1, x2 u2 2 x1, x3 u3 3 x1 )
x1 x1 x1
R : (x1 u1R , x2 x2 u2R , x3 u3R )
u1 u u
(x1 u1 x2 , x2 x2 u2 2 x2 , x3 u3 3 x2 )
x2 x2 x2

Length of the line segment P'Q'

2 2 2
PQ x1P x1Q x2P x2Q x3P x3Q
42
 Deformation Field
Length of the line segment P'Q'
2 2 2
u1 u 2 u3
PQ x1 1
x1

x

1 x1
1/2
2 2 2
u u1 u 2 u3
x1 1 2 1
x1 x1
x
1 x1

u 1 u1
2
1 u 2
2
1 u3
2
u1
x1 1 1 x 1
x1
x1 2 x1

2 x 1 2 x1

Linear Nonlinear Ignore H.O.T. when displacement
gradients are small
Linear normal strain
PQ PQ u1
11
PQ x1
u2 u3
22 , 33
x2 x3 43
 Deformation Field
Shear strain xy
change in angle between two lines originally parallel to x and y
axes xQ xQ u xR x R u
2 2 2 1 1 1
1 2
x1 x1 x2 x2
u1 u2
12 1 2
x2 x1
u2 u3
23 Engineering shear strain
x3 x2
u3 u1
13
x1 x3 Different notations

1 u1 u2 1 ui uj
12
2 x2 x1
ij
2 xj xi


1 u2 u3
23
2 x3 x2 ij 21 (ui,j uj,i )
1 u3 u1
13
2 x1 x3

sym(u) 44
 Strain Tensor
Strain Tensor
ijei ej
11 12 13
Cartesian Components [ij ] 12 22 23

13 23 33
Vector notation
11 11

22 22
33 33
{ }
212 12
223 23

213 13

45
 Volumetric and Deviatoric Strain
Volumetric strain (from small strain assumption)
V V0
V (1 11 )(1 22 )(1 33 ) 1 11 22 33
V0
V 11 22 33 kk

Deviatoric strain x3
e 31 V 1 eij ij 31 V ij e22
e33
e Idev : 1
x2 1
1
e11 x1

Exercise: Write Idev in matrix-vector notation

46
 Stress-Strain Relationship
Applied Load shape change (strain) stress
There must be a relation between stress and strain
Linear Elasticity: Simplest and most commonly used

Ultimate
Fracture
stress
Yield stress

Proportional
limit Youngs
modulus

Strain Necking
hardening

47
 Generalized Hookes Law
Linear elastic material D : , ij Dijklkl
In general, Dijkl has 81 components
Due to symmetry in ij, Dijkl = Djikl
Due to symmetry in kl, Dijkl = Dijlk 21 independent coeff

from definition of strain energy, Dijkl = Dklij

Isotropic material (no directional dependence)

Most general 4-th order isotropic tensor
Dijkl ijkl ik jl il jk
ijkl (ik jl il jk ) D 1 1 2I

Have only two independent coefficients

(Lames constants: and )

48
 Generalized Hookes Law cont.
Stress-strain relation
ij Dijklkl [ijkl (ik jl il jk )]kl kk ij 2ij
Volumetric strain:
kk 11 22 33 v
Off-diagonal part:
12 212 12 is the shear modulus
Bulk modulus K: relation b/w volumetric stress & strain

I1 3m jj kk jj 2 jj (3 2)kk

p m ( 23 )kk K v
Substitute so that we can separate
Bulkvolumetric
modulus part
K 23
Total deform. = volumetric + deviatoric deform.

49
 Generalized Hookes Law cont.
Stress-strain relation cont.
ij (K 23 )kk ij 2ij
Kkk ij 2ij 23 kk ij
Kijklkl 2[ik jl 31 ijkl ]kl
Kijkl 2(Idev )ijkl kl

K1 1 2Idev :
Deviatoric part
e Idev :
Volumetric part Deviatoric strain
Kv 1 2e s Idev :
m 1 s Deviatoric stress

Important for plasticity; plastic deformation only occurs in deviatoric part

volumetric part is always elastic
50
 Generalized Hookes Law cont.
Vector notation
The tensor notation is not convenient for computer implementation
Thus, we use Voigt notation 2nd-order tensor vector
4th-order tensor matrix
Strain (61 vector), Stress (61 vector), and C (66 matrix)

11 u1,1 11

u2,2
22 22
33 u3,3 33
D
212 u1,2 u2,1 12
223 23
u
2,3 u3,2

213 u u
13
1,3 3,1

12 + 21 = 212 You dont need 2 here

51
 3D Solid Element cont.
Elasticity matrix 2 31 31 0 0 0
1 3

1 31 2 1
3 0 0 0
D K1 1 2Idev
3
31 31 23 0 0 0
1
1 Idev
0 0 21 0 0
0
0

Relation b/w 0 0 0 1
0 0 2 0

Lames constants 0
0 0 0 21

0 0
and Youngs modulus
and Poissons ratio 2 0 0 0
2 0 0 0
(3 2)
, E
2( ) 2 0 0 0
D
E E 0 0 0 0 0
,
(1 )(1 2) 2(1 ) 0 0 0 0 0

0 0 0 0 0

Textbook has a definition of D in terms of E and

52
 Plane Stress
Thin platelike components parallel to the xyplane
The plate is subjected to forces in its in-plane only
13 = 23 = 33 = 0

11 1 0 11
E
{ } 22 1 0 22
2
1 1 (1 )
12 0 0 2 12

13 = 23 = 0, but 33 0
33 can be calculated from the condition of 33 = 0:

33 (11 22 )
1
53
 Plane Strain
Strains with a z subscript are all zero: 13 = 23 = 33 = 0
Deformation in the zdirection is constrained, (i.e., u3 = 0)
can be used if the structure is infinitely long in the z
direction

11 1 0 11
E
{ } 22 1 0 22
(1 )(1 2) 1
12 0 0 12
2

13 = 23 = 0, but 33 0
33 can be calculated from the condition of 33 = 0:
E
33
(1 )(1 2)
11 22

54
1.4

MECHANICS OF
CONTINUOUS BODIES

55
 Governing Equations for Equilibrium
Governing differential equations for structural equilibrium
Three laws of mechanics: conservation of mass, conservation of
linear momentum and conservation of angular momentum
Boundary-valued problem: satisfied at every point in
Governing D.E. + Boundary conditions
Solutions: C2continuous for truss & solid, C4continuous for beam
Unnecessarily requirements for higher-order continuity
Energy-based method
For conservative system, structural equilibrium when the potential
energy has its minimum: Principle of minimum potential energy
If the solution of BVP exists, then that solution is the minimizing
solution of the potential energy
When no solution exists in BVP, PMPE may have a natural solution
Principle of virtual work
Equilibrium of the work done by both internal and external forces
with small arbitrary virtual displacements
56
 Balance of Linear Momentum
Balance of linear momentum

fb d t n d a d fb: body force

tn: surface traction

Stress tensor (rank 2): 0 for static problem

11 12 13
ijei ej ij
21 22 23
31 32 33

Surface traction

X
t n n
Cauchys Lemma
X3
t n t n e3
n
e1 X2
e2
t n n t n n X1 tn

57
 Balance of Linear Momentum cont
Balance of linear momentum

( fb a) d n d d

Divergence Theorem
[ ( fb a)] d 0

( fb a) 0
For a static problem

fb 0 ij,i fjb 0

Balance of angular momentum

x fb d x t n d x a d

T ij ji
58
 Boundary-Valued Problem
We want to determine the state of a body in equilibrium
The equilibrium state (solution) of the body must satisfy
local momentum balance equation
boundary conditions

Strong form of BVP

X
Given body force fb, and traction t
on the boundary, find u such that fb
X3
fb 0 (1)
e3
and e1 n
e2 X2
u0 on h essential BC (2) X1 t
t n on s natural BC (3)

Solution space

DA u [C2 ()]3 | u 0 on x h , n t on x s 59
 Boundary-Valued Problem cont.
How to solve BVP
To solve the strong form, we want to construct trial solutions that
automatically satisfy a part of BVP and find the solution that
satisfy remaining conditions.
Statically admissible stress field: satisfy (1) and (3)
Kinematically admissible displacement field: satisfy (2) and have
piecewise continuous first partial derivative
Admissible stress field is difficult to construct. Thus, admissible
displacement field is used often

60
 Principle of Minimum Potential Energy (PMPE)
Deformable bodies generate internal forces by
deformation against externally applied forces
Equilibrium: balance between internal and external forces
For elastic materials, the concept of force equilibrium can
be extended to energy balance
Strain energy: stored energy due to deformation
(corresponding to internal force)
1
U( u) ( u) : ( u) d ( u) D : ( u)
2
Linear elastic material

For elastic material, U(u) is only a function of total

displacement u (independent of path)
61
 PMPE cont.
Work done by applied loads (conservative loads)

W( u) u fb d s u t d
.

U(u) is a quadratic function of u, while W(u) is a linear

function of u. g
u3
fb
u1 u2
x3 h

Potential energy fs
x2

( u) U( u) W( u) x1

1
( u) : ( u) d u fb d s u t d
.
2

62
 PMPE cont.
PMPE: for all displacements that satisfy the boundary
conditions, known as kinematically admissible
displacements, those which satisfy the boundary-valued
problem make the total potential energy stationary on DA

But, the potential energy is well defined in the space of

kinematically admissible displacements

Z u [H1 ()]3 | u 0 on x h ,
H1: first-order derivatives are integrable

No need to satisfy traction BC (it is a part of potential)

Less requirement on continuity
The solution is called a generalized (natural) solution
63
 Example Uniaxial Bar
Strong form
EAu 0 x [0, L] x F
u0 x0
L
EAu(L) F xL
Integrate twice: EAu(x) c x c
1 2
Fx
Apply two BCs: u(x) Solution of BVP
EA
PMPE with assumed solution u(x) = c1x + c2
To satisfy KAD space, u(0) = 0, u(x) = c1x
Potential energy: U 1 L EA(u)2 dx EALc12
2 0
W Fu(L) FLc1

d d F Fx
(U W) EALc1 FL 0 c1 u(x)
dc1 dc1 EA EA
64
 Virtual Displacement
Virtual displacement is not experienced but only assumed to exist so
that various possible equilibrium positions may be compared to
determine the correct one
Let mass m and springs are in equilibrium at the current position
Then, a small arbitrary perturbation, r, can be assumed
Since r is so small, the member forces are assumed unchanged
The work done by virtual displacement is

W F1 r F2 r F3 r F
4 r (
F1 F2 F3 F
4) r
If the current position is in force equilibrium, W = 0

F2 F3

F1 F4
65
 Virtual Displacement Field
Virtual displacement (Space Z)
Small arbitrary perturbation (variation) of real displacement
1 d
u lim [( u ) ( u)] ( u ) u.
0 d 0
Let be the virtual displacement, then u +
must be kinematically
admissible, too
Then, must satisfy homogeneous displacement BC

u a u u V u Z

Z u u [H1 ()]3, u h
0 u
Space Z only includes homogeneous
essential BCs
In the literature, u is often used instead of
Property of variation
du d(u)

dx dx
66
 PMPE As a Variation
Necessary condition for minimum PE
Stationary condition <--> first variation = 0
1 d
( u; u ) lim [( u u ) ( u)] ( u u ) 0
0 d 0

for all u Z
Variation of strain energy

u d u u u

x d x 0 x
( u) ( u ) D :

U( u; u ) (u ) : D : (u) ( u) : D : (u )
1
2
d
( u ) : D : ( u) d

a( u, u ) Energy bilinear form
67
 PMPE As a Variation cont.
Variation of work done by applied loads
b
W( u; u ) u f d s u t d l ( u ) Load linear form

( u; u ) U( u; u ) W( u; u ) 0
Thus, PMPE becomes

a( u, u ) l ( u ), u Z
Load form is linear with respect to
l (u )
Energy form a(u, ) is symmetric, bilinear w.r.t. u and
Different problems have different a(u, ) and , but they share
l (u )
the same property
How can we satisfy for all requirement?
Can we test an infinite number of ?
68
 Example Uniaxial Bar
Assumed displacement u(x) = cx u(x) cx
virtual displacement is in the same space with u(x):
Variation of strain energy
d 1 L 2 1 L
U
d 2 0
EA (u u) dx
0
2EA(u u)u dx
2 0 0
L
0 EAuu dx EALcc
Variation of applied load
d
W F u(L) u(L)
Fu(L) FLc
d 0

PMPE
Fx
U W c(EALc FL) 0 u(x) cx
EA
Arbitrary u(x) arbitrary c coefficient of c must be zero 69
 Principle of Virtual Work
Instead of solving the strong form directly, we want to
solve the equation with relaxed requirement (weak form)
Virtual work Work resulting from real forces acting
through a virtual displacement
Principle of virtual work when a system is in equilibrium,
the forces applied to the system will not produce any
virtual work for arbitrary virtual displacements
Balance of linear momentum is force equilibrium fb 0
Thus, the virtual work can be obtained by multiplying the force
equilibrium equation with a virtual displacement

W ( fb ) u d
If the above virtual work becomes zero for arbitrary , then it
satisfies the original equilibrium equation in a weak sense
70
 Principle of Virtual Work cont
PVW (ij,i fjb )uj d 0
u Z

ij,iuj d fjbuj d

Integration-by-parts

(ijuj ),i ijuj,i d

fjbuj d

Divergence Thm

The
n uj ddecomposed
boundary
u byd fjbuj d
i ij is ij j,i

h s
uj 0 on h and niij tj on s

b
t u
S j j
d ij j,i

u d j uj d
f

0 u
71
 Principle of Virtual Work cont
Since ij is symmetric
1 ui uj
ijuj,i ijsym(uj,i ) ij ij sym(ui,j ) ij

2 Xj Xi
Weak Form of BVP

ij ij d fjbuj d s tjuj d u Z

Internal virtual work = external virtual work

Starting point of FEM
Symbolic expression

a( u, u ) l ( u ) u Z [K]{ d} {F }
FE equation
Energy form:
a( u, u ) : d
Load form:
l (u ) u fb d s u t d
72
 Example Heat Transfer Problem
Steady-State Differential Equation
T = T0
T T
kx ky Q 0 ST
x x y y domain
A
Boundary conditions qn Sq Q

T T0 on ST

dT dT
nq n k
x x n k
y y on Sq n = {nx,
dx dy ny}T
Space of kinematically admissible temperature

Z T H1 () T( x) 0, x ST
Multiply by virtual temperature, integrate by part, and
apply boundary conditions
T T T T
x x x y y y d
k k TQ d S Tqn dSq,
q
T Z
73
 Example Beam Problem
Governing DE f(x)

d4 v
EI 4
f(x), x [0, L] x L
dx
Boundary conditions for cantilevered beam
dv d2 v d3 v
v(0) (0) (L) (L) 0
dx dx 2
dx 3

Space of kinematically admissible displacement

2 dv
Z v H [0, L] v(0) (0) 0
dx

Integrate-by-part twice, and apply BCs

L d2v d2v L
0 EI dx2 dx2 dx 0 fv dx, v Z
74
 Difference b/w Strong and Weak Solutions
The solution of the strong form needs to be twice
differentiable k T ky T Q 0
x y
x x y
The solution of the weak form requires the first-order
derivatives are integrable bigger solution space than
that of the strong form k T T k T T d
x
x x y
y y

If the strong form has a solution, it is the solution of the
weak form
If the strong form does not have a solution, the weak
form may have a natural solution

75
1.5

FINITE ELEMENT METHOD

76
 Finite Element Approximation
Difficult to solve a variational equation analytically
Approximate solution n
Linear combination of trial functions u(x) c (x)
i i
Smoothness & accuracy depend on i1
the choice of trial functions
If the approximate solution is expressed in the entire domain, it is
difficult to satisfy kinematically admissible conditions
Finite element approximation
Approximate solution in simple sub-domains (elements)
Simple trial functions (low-order polynomials) within an element
Kinematically admissible conditions only for elements on the
boundary u(x)
Nodes Approximate
solution
Piecewise-
x
linear
approximation
Finite elements
Exact solution 77
 Finite Elements
Types of finite elements
1D 2D 3D

Linear

Quadratic

Variational equation is imposed on each element.

1 0.1 0.2 1
0 dx 0 dx
0.1
dx L
0.9
dx

One element
78
 Trial Solution
Solution within an element is approximated using simple
polynomials.

u(x) a a x,
i-th element 1 x xx
0 is composedi of two nodes:
i 1 xi and xi+1. Since two
unknowns are involved, linear polynomial can be used:

The unknown coefficients, a0 and a1, will be expressed in terms of

nodal solutions u(xi) and u(xi+1).
79
 Trial Solution cont.
Substitute two nodal values
u(xi ) ui a0 a1xi

u(xi1 ) ui1 a0 a1xi1

Express a0 and a1 in terms of ui and ui+1. Then, the solution is

approximated by
xi1 x x xi
u(x) (e)
ui (e)
ui1
14L2 43 L 3
12
N1 (x) N2 (x)

u(x) N (x)ui N2 (x)ui1 , xi x xi1

Solution1 for Element e:

N1(x) and N2(x): Shape Function or Interpolation Function 80

 Trial Solution cont.
Observations
Solution u(x) is interpolated using its nodal values ui and ui+1.
N1(x) = 1 at node xi, and =0 at node xi+1.
N1(x) N2(x)

xi xi+1

The solution is approximated by piecewise linear polynomial and its

gradient is constant within an element.

81
 1D Finite Elements
1D BVP d2u p(x) 0, 0 x 1
2
dx
u(0) 0

du Boundary conditions
(1) 0
dx Space of kinematically
admissible displacements
1 d2u
Use PVW 0 dx2 p u dx 0
Z u H(1) [0,1] u(0) 0
Integration-by-parts
1
du du du
1 1
u dx pu dx
dx 0
0 dx dx 0
This variational equation also satisfies at individual element level

xj du du xj
(1) xi dx dx
dx x i
pu dx u
82
 1D Interpolation Functions
Finite element approximation for one element (e) at a time

u(e) (x) uN
i 1 (x) u N
i1 2 (x) N (e)
d(e)

(e) ui
d N(e) N1 N2
ui1
Satisfies interpolation condition u(e) (xi ) ui
u(e) (xi1 ) ui1
Interpolation of displacement variation (same with u)
u (e) (x) uN
i 1 (x) u N
i1 2 (x) N(e)
d (e)

Derivative of u(x): differentiating interpolation functions

du(e) dN1 dN2 ui 1 1 ui

(e) (e) B(e) d(e)
dx dx dx ui1 L L u i1
83
 Element-Level Variational Equation
Approximate variational equation (1) for element (e)
du
(x )
d (e)T
xj xj
B(e)T B(e) dx d(e) d (e)T N(e)T p(x) dx d (e)T dx i
x xi

du

i
(xi1 )
dx
Must satisfied for all u (e) (x) Z
If element (e) is not on the boundary, d (e) can be arbitrary

Element-level variational equation

du
(x )
i
xj
B(e)T B(e) dx d(e)
xj (e)T dx
x
N
xi
p(x) dx
du

i
(xi1 )
dx
2x2 matrix 2x1 vector
du
(x )
i

[k (e) ]{ d(e) } f(e) dx
du

(xi1 )
dx
84
 Assembly
Need to derive the element-level equation for all elements
Consider Elements 1 and 2 (connected at Node 2)
du
k11 k12
(1) (1)
dx 1 (x )
u1 f1

k21 k22 u f
2 2 du
(x2 )
dx
du

k11 k12
(2) (2)
dx 2 (x )
u2 f2

k21 k22 u3 f
3 du
(x3 )
dx
Assembly
Vanished
du unknown term
(1)
k11 (1)
k12 0 u1 f1(1)
dx 1 (x )
(1) (1) (2)

(2) (1) (2)
k21 k22 k11 k12 u2 f2 f 2 0
(2) (2) u du
0 k21 k22 3 f3(2) (x )
dx 3
85
 Assembly cont.
Assembly of NE elements (ND = NE + 1)
du

(1)
k11 (1)
k12 0 K 0 u f1(1) dx 1 (x )
1
(1) (1) (2) (2) (1) (2)
k21 k22 k11 k12 L 0 u2 f2 f2 0
0 (2) (2) (2) (3)
k221 k22 k11 L 0 u3 (2)
f3 f 3 0

M M M O M M M M

NE (N ) du
k22E uN fN E
(N )
0 0 0 k21 (x
dx N )
ND 1 ND 1
ND ND ND 1

[K]{ d} {F }
Coefficient matrix [K] is singular; it will become non-
singular after applying boundary conditions

86
 Example
Use three equal-length elements
d2u
x 0, 0 x 1 u(0) 0, u(1) 0
2
dx
All elements have the same coefficient matrix
(e) 1 1 1 3 3
k
22 L(e) 1 1 3 3 , (e 1,2,3)

RHS (p(x) = x)

(e) xi1 N1 (x) 1 x i1 x(xi1 x)

{f } p(x) dx (e) x dx
xi
N2 (x) L i x(x x i )
xi xi1

(e) 3 6
L , (e 1,2,3)
x
i i1x
6 3

87
 Example cont.
RHS cont. f1(1) 1 1 (2)
f 2 1 4 f3
(3)
1 7
(1) , (2) , (3)

f2 54 2 f3 54 5 f
4 54
8

Assembly
1 du
- (0)
54 dx Element 1
3 - 3 0
0 u1
2 4 Element 2
- 3 3 + 3 - 3 0 u2 +
= 54 54 Element 3

0 - 3 3 +3 - 3 u3 7 5
+

0 0 - 3 3 u4 54 54
8 du
+ (1)
Apply boundary conditions 54 dx
Deleting 1st and 4th rows and columns
4
6 3 u2 1 1 u2 81
3 6 u
3 9
2 u3 5
81

88
 EXAMPLE cont.
Approximate solution
4 1
x, 0x
27 3
4 1 1 1 2
u(x) x , x
81 27 3 3 3
5 5 2 2
x , x1
81 27 3 3
Exact solution

1
u(x) x 1 x2
6

Three element solutions are poor
Need more elements

89
 3D Solid Element
Isoparametric mapping
Build interpolation functions on the reference element
Jacobian: mapping relation between physical and reference elem.
Interpolation and mapping
8 8
Same for mapping
u( ) NI ()uI x( ) NI ()xI and interpolation
I 1 I 1
1
NI () (1 I )(1 I )(1 I )
8

x8 x7 (1, (
1,1) 1,1,1)
x5 (1, 1,1)
(1,1,1
x6 )
x4 x3 (1,1,
x 1
(1, 1,
x3 1)
1) (1,1,1)
x2
x2
x1 (a) Finite Element (b) Reference
90
Element
 3D Solid Element cont.
Jacobian matrix
8
x NI () J : Jacobian
J33 xI
I 1
x1 x1 x1
Derivatives of shape functions

NI NI NI NI NI NI x2 x2 x2

x1 x2 x3

x3 x3 x3
NI NI
J
x
NI NI
J 1
x
Jacobian should not be zero anywhere in the element
Zero or negative Jacobian: mapping is invalid (bad element shape)

91
 3D Solid Element cont.
Displacement-strain relation NI,1 0 0
8
0 NI,2 0
( u) BI uI 0 0 NI,3
I 1 BI
8 NI,2 NI,1 0

( u ) BI uI 0 NI,3 NI,2
I 1 N 0 NI,1
I,3

Ni
NI,i
xi

92
 3D Solid Element cont.
Transformation of integration domain
1 1 1
d 1 1 1 J ddd
Energy form
8 8
1 1 1 u { d }T [k]{ d}
a( u, u ) uIT T
1 1 1 I J
B DB J d d d
J
I 1 J 1

Load form
8
T 1 1 1
l (u )
uI
1 1 1
NI ()fb J ddd { d }T { f }
I 1

Discrete variational equation

{ d }T [k]{ d} { d }T { f }, { d } Zh

93
 Numerical Integration
For bar and beam, analytical integration is possible
For plate and solid, analytical integration is difficult, if
not impossible
Gauss quadrature is most popular in FEM due to simplicity
and accuracy
1D Gauss quadrature
NG
1
1 f() d
i1
if( i )
NG: No. of integ. points; i: integ. point; i: integ. weight
i and i are chosen so that the integration is exact
for (2*NG 1)-order polynomial
Works well for smooth function
Integration domain is [-1, 1]
94
 Numerical Integration cont.
Multi-dimensions
NG NG
1 1
1 1 f(, ) dd
i1 j1
ijf(i, j )

NG NG NG
1 1 1
1 1 1 f(, , ) ddd ijk f(i, j, k )
i1 j1 k 1

Integration
NG Weights (i)
Points (i)
1 0.0 2.0
2 .5773502692 1.0
.7745966692 .5555555556
3 (a)
0.0 .8888888889
11
.8611363116 .3478546451
4
.3399810436 .6521451549
.9061798459 .2369268851
5 .5384693101 .4786286705
.5688888889 (b) (c) 33
0.0 95
22
 ELAST3D.m
A module to solve linear elastic problem using NLFEA.m

Input variables for ELAST3D.m

Variable Array size Meaning
ETAN (6,6) Elastic stiffness matrix Eq. (1.81)
UPDATE Logical variable If true, save stress values
LTAN Logical variable If true, calculate the global stiffness matrix
NE Integer Total number of elements
NDOF Integer Dimension of problem (3)
XYZ (3,NNODE) Coordinates of all nodes
LE (8,NE) Element connectivity

96
 How to Solve Linear Problem in Nonlinear Code
Linear matrix solver
[K]{ d} {F } {fint} = {fext} {f} = {fext} {fint} = {0}

Construct stiffness matrix and force vector

Use LU decomposition to solve for unknown displacement {d}
Nonlinear solver (iterative solver)
Assume the solution at iteration n is known, and n+1 is unknown
{ dn 1 } { dn } { d} For linear problem, {dn} = {0}

n 1 n f
{f } { f } { d} { 0 }
d
{F } [K]{ dn } [K]{ d} 0

[K]{ d} {F } Only one iteration!!

97
function ELAST3D(ETAN, UPDATE, LTAN, NE, NDOF, XYZ, LE)
%***********************************************************************
% MAIN PROGRAM COMPUTING GLOBAL STIFFNESS MATRIX AND RESIDUAL FORCE FOR
% ELASTIC MATERIAL MODELS
%***********************************************************************
%%
global DISPTD FORCE GKF SIGMA
%
% Integration points and weights (2-point integration)
XG=[-0.57735026918963D0, 0.57735026918963D0];
WGT=[1.00000000000000D0, 1.00000000000000D0];
%
% Index for history variables (each integration pt)
INTN=0;
%
%LOOP OVER ELEMENTS, THIS IS MAIN LOOP TO COMPUTE K AND F
for IE=1:NE
% Nodal coordinates and incremental displacements
ELXY=XYZ(LE(IE,:),:);
% Local to global mapping
IDOF=zeros(1,24);
for I=1:8
II=(I-1)*NDOF+1;
IDOF(II:II+2)=(LE(IE,I)-1)*NDOF+1:(LE(IE,I)-1)*NDOF+3;
end
DSP=DISPTD(IDOF);
DSP=reshape(DSP,NDOF,8);
%
%LOOP OVER INTEGRATION POINTS
for LX=1:2, for LY=1:2, for LZ=1:2
E1=XG(LX); E2=XG(LY); E3=XG(LZ);
INTN = INTN + 1;
%
% Determinant and shape function derivatives
[~, SHPD, DET] = SHAPEL([E1 E2 E3], ELXY);
FAC=WGT(LX)*WGT(LY)*WGT(LZ)*DET; 98
 % Strain
DEPS=DSP*SHPD';
DDEPS=[DEPS(1,1) DEPS(2,2) DEPS(3,3) ...
DEPS(1,2)+DEPS(2,1) DEPS(2,3)+DEPS(3,2) DEPS(1,3)+DEPS(3,1)]';
%
% Stress
STRESS = ETAN*DDEPS;
%
% Update stress
if UPDATE
SIGMA(:,INTN)=STRESS;
continue;
end
%
% Add residual force and stiffness matrix
BM=zeros(6,24);
for I=1:8
COL=(I-1)*3+1:(I-1)*3+3;
BM(:,COL)=[SHPD(1,I) 0 0;
0 SHPD(2,I) 0;
0 0 SHPD(3,I);
SHPD(2,I) SHPD(1,I) 0;
0 SHPD(3,I) SHPD(2,I);
SHPD(3,I) 0 SHPD(1,I)];
end
%
% Residual forces
FORCE(IDOF) = FORCE(IDOF) - FAC*BM'*STRESS;
%
% Tangent stiffness
if LTAN
EKF = BM'*ETAN*BM;
GKF(IDOF,IDOF)=GKF(IDOF,IDOF)+FAC*EKF;
end
end, end, end, end
end 99
function [SF, GDSF, DET] = SHAPEL(XI, ELXY)
%*************************************************************************
% Compute shape function, derivatives, and determinant of hexahedral element
%*************************************************************************
%%
XNODE=[-1 1 1 -1 -1 1 1 -1;
-1 -1 1 1 -1 -1 1 1;
-1 -1 -1 -1 1 1 1 1];
QUAR = 0.125;
SF=zeros(8,1); SF(81): shape functions,
DSF=zeros(3,8);
for I=1:8
XP = XNODE(1,I); GDSF (38): shape functions derivatives
YP = XNODE(2,I);
ZP = XNODE(3,I);
%
DET: Jacobian of the mapping
XI0 = [1+XI(1)*XP 1+XI(2)*YP 1+XI(3)*ZP];
%
SF(I) = QUAR*XI0(1)*XI0(2)*XI0(3);
DSF(1,I) = QUAR*XP*XI0(2)*XI0(3);
DSF(2,I) = QUAR*YP*XI0(1)*XI0(3);
DSF(3,I) = QUAR*ZP*XI0(1)*XI0(2);
end
GJ = DSF*ELXY;
DET = det(GJ);
GJINV=inv(GJ);
GDSF=GJINV*DSF;
end

100
 One Element Tension Example
x3
%
% One element example 10kN 10kN
%
% Nodal coordinates 10kN 5 10kN
XYZ=[0 0 0;1 0 0;1 1 0;0 1 0;0 0 1;1 0 1;1 1 1;0 1 1]; 8
%
% Element connectivity
LE=[1 2 3 4 5 6 7 8];
6 7
%
% External forces [Node, DOF, Value]
EXTFORCE=[5 3 10.0E3; 6 3 10.0E3; 7 3 10.0E3; 8 3 10.0E3]; x2
% 1 4
% Prescribed displacements [Node, DOF, Value]
SDISPT=[1 1 0;1 2 0;1 3 0;2 2 0;2 3 0;3 3 0;4 1 0;4 3 0];
% 2
% Material properties x1 3
% MID:0(Linear elastic) PROP=[LAMBDA NU]
MID=0;
PROP=[110.747E3 80.1938E3];
%
% Load increments [Start End Increment InitialFactor FinalFactor]
TIMS=[0.0 1.0 1.0 0.0 1.0]';
%
% Set program parameters
ITRA=30; ATOL=1.0E5; NTOL=6; TOL=1E-6;
%
% Calling main function
NOUT = fopen('output.txt','w');
NLFEA(ITRA, TOL, ATOL, NTOL, TIMS, NOUT, MID, PROP, EXTFORCE, SDISPT, XYZ, LE);
fclose(NOUT); 101
 One Element Output
Command line output
Time Time step Iter Residual
1.00000 1.000e+00 2 5.45697e-12

Contents in output.txt
TIME = 1.000e+00

Nodal Displacements

Node U1 U2 U3
1 0.000e+00 0.000e+00 0.000e+00
2 -5.607e-08 0.000e+00 0.000e+00
3 -5.607e-08 -5.607e-08 0.000e+00
4 0.000e+00 -5.607e-08 0.000e+00
5 -5.494e-23 1.830e-23 1.933e-07
6 -5.607e-08 4.061e-23 1.933e-07
7 -5.607e-08 -5.607e-08 1.933e-07
8 -8.032e-23 -5.607e-08 1.933e-07

Element Stress

S11 S22 S33 S12 S23 S13

Element 1
0.000e+00 1.091e-11 4.000e+04 -2.322e-13 6.633e-13 -3.317e-12
0.000e+00 0.000e+00 4.000e+04 -3.980e-13 1.327e-13 -9.287e-13
-3.638e-12 7.276e-12 4.000e+04 -1.592e-12 -2.123e-12 -3.317e-12
0.000e+00 0.000e+00 4.000e+04 2.653e-13 -2.123e-12 5.307e-13
0.000e+00 0.000e+00 4.000e+04 5.638e-13 3.449e-12 -1.327e-12
0.000e+00 0.000e+00 4.000e+04 -1.194e-12 4.776e-12 1.061e-12
0.000e+00 0.000e+00 4.000e+04 -7.960e-13 2.919e-12 -3.449e-12
3.638e-12 3.638e-12 4.000e+04 -5.307e-13 3.715e-12 1.061e-12
*** Successful end of program *** 102