1

CHAP 1
Preliminary Concepts and
Linear Finite Elements

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

Web: http://www.mae.ufl.edu/nkim/

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 METHODS

1.2

VECTOR AND TENSOR

CALCULUS
4

Vector and Tensor

Vector: Collection of scalars
Cartesian vector: Euclidean vector defined using Cartesian
coordinates
2D, 3D Cartesian vectors

u1
u
u2

u1

, or u u2
u
3

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

u u1 e1 u2e2 u3e3

Index Notation and Summation Rule

Index notation: Any vector or matrix can be expressed in
terms of its indices
v1
v [vi ] v2
v3

A11 A12 A 13
A [Aij ] A21 A22 A 23
A31 A32 A 33

Einstein summation convention

3

akbk

k 1

akbk

In this case, k is a dummy variable (can be j or i)

The same index cannot appear more than twice

Basis representation of a vector

Let ek be the basis of vector space V
Then, any vector in V can be represented by

wk ek

k 1

wk ek

Index Notation and Summation Rule cont.

Examples
Matrix multiplication:

C A B

Trace operator:

tr(A) A11 A22 A33 Akk

u v u1v1 u2v2 u3v3 uk vk

Dot product:

u v ujvk ( ej ek ) eijkujvk ei

Cross product:

Permutation
symbol

Cij AikBkj

eijk

1
1

unless i, j,k are distinct

if (i, j,k) is an even permutation
if (i, j,k) is an odd permutation

Contraction: double dot product

J A:B

AijBij AijBij
i1 j1

Cartesian Vector
Cartesian Vectors
u u1 e1 u2 e2 u3e3 uiei

X3
v

v vjej

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

e3
e1

e2

X2

X1

uivj ij uivi

1 if i j
ij
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
8

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

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
Rank 2 tensor A: Aij
Rank 4 tensor C: Cijkl

10

Tensor Operations
Basic rules for tensors
( TS)R T(SR)
T(S R) TS TR
( TS) ( T )S T(S)
1T T1 T

Different notations

TS T S

Identity tensor

1 [ij ]

Tensor (dyadic) product: increase rank

A u v uivjei ej
Aij uivj
( u v ) w u( v w)
w ( u v ) v ( w u)
( u v )( w x) ( v w)u x
uv vu

11

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 )

1 (T TT )
W

Note: W has zero diagonal

2 components and W ij = - Wji

Properties Let A be a symmetric tensor

A:W 0
A:T A:S
12

Example
Displacement gradient can be considered a tensor (rank 2)

u1
X1

u1
X2

u2
X1

u2
X2

u3
X1

u3
X2

u1
X3
u2
X3
u3
X3

u1
X1

sym(u)

1 ( u1
2 X2

1 ( u1
2 X3

1 ( u1
2 X2

u2
)
X1

u3
)
X1

u2
)
X1

u2
X2
1 ( u2
2 X3

u2
)
X1
u3
)
X1

1 ( u1
2 X3

1 ( u2
2 X3

u3
)
X2

1 ( u1
2 X2

skew(u) 21 ( X1
2
1 u1
2 ( X3

u3
)
X2

u2
)
X1

u3
)
X2

Strain tensor

u3
X3

0
u
21 ( X2
3

u3
)
X1

1 ( u1
2 X3

1 ( u2
2 X3

u3
)
X1

u3
)
X2

Rotation
tensor
13

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
14

Orthogonal Tensor
In two different coord.
u uiei uj* ej*
Direction cosines
ei*

ei*

ej ij

ijej

Change basis
u ujej ui* ei*
ui*ijej

e1

e3*

e2

e1

e2*

We can also show

ej ijei*

u* u

u T u* T ( u) (T )u
1 T

uj ijui*
u T u*

e3

T T 1 det() 1
Orthogonal tensor
Rank-2 tensor transformation

T * T T ,

Tij* ikTkljl

15

Permutation
The permutation symbol has three indices, but it is not a
tensor

eijk

1
0

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

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

the permutation is zero when any of two indices have the

same value
Identity
eijk elmk il jm im jl
vector product
u v eieijkujvk
16

Dual Vector
For any skew tensor W and a vector u
u Wu u WT u u
Wu

Wu and u are orthogonal

Let Wij eijk wk

W W12
W13

W12
0

W23

W13
W23
0

W 23

w W13

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
17

Vector and Tensor Calculus

Gradient

( X)

ei
X
Xi
Gradient is considered a vector

We will often use a simplified notation: vi , j

Laplace operator

ei

i
2

ej

X
j

vi

X j

X j X j

Gradient of a scalar field (X): vector

( X) ei
Xi
18

Vector and Tensor Calculus

Gradient of a Tensor Field (increase rank by 1)
j

ei
j ej
ei ej
Xi
Xi
Divergence (decrease rank by 1)
i

ei
j ej

Xi
Xi

Ex)

Curl

jk , j ek

v ei eijkvk , j

19

Integral Theorems
Divergence Theorem

A d n

A d

n: unit outward normal vector

Gradient Theorem

A d n A d

Stokes Theorem

n ( v ) d
c r

v dc

Reynolds Transport Theorem

d
Ad

dt

A
t d (n v)A d
20

Integration-by-Parts
u(x) and v(x) are continuously differentiable functions
1D
b

a u(x)v(x) dx

u(x)v(x)

b
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

Greens identity

n) d u vd

2
u v d u v n d u
vd

21

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

22

1.3

STRESS AND STRAIN

23

Surface Traction (Stress)

Surface traction (Stress)

f1

The entire body is in equilibrium with

external forces (f1 ~ f6)
The imaginary cut body is in equilibrium due
to external forces (f1, f2, f3) and internal
forces

n
P

f3

Internal force acting at a point P

on a plane whose unit normal is n:

(n)

f6
f1

F
lim
A 0 A

The surface traction depends on the unit

normal direction n.

t (n) t1 e1 t2 e2 t3e3

f5

f4

f2
z

Surface traction will change as n changes.

unit = force per unit area (pressure)

f2

f3
y

x
24

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
t
t

(x)

(x)

t1(x) e1

t2(x) e2

t3(x) e3

11 e1 12 e2 + 13e3
Normal
stress

z
13
F

11

Shear
stress

12
y
25

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

ijei ej

33

Sign convention

31

32

13

23

sgn(11 ) sgn(n) sgn( Fx )

sgn(12 ) sgn(n) sgn( Fy )

12

11
x

21

22

26

Symmetry of Stress Tensor

Stress tensor should be symmetric
9 components
6 components
Equilibrium of the angular moment

M l(12 21 ) 0
12 21

Similarly for all three directions:

12 21 ,

23 32 ,

13 31

11

22
Lets use vector notation:
Cartesian components
33
{

of stress tensor
12
23

13

21

12 l
C

21

11
[ij ] 12
13

12

12

22
23

13
23
33
27

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.
y
Surface area (ABC = A)
B

PAB An3 ; PBC An1 ; PAC An2

The surface traction

13
11

t ( n ) t1( n ) e1 t2( n) e2 t3(n ) e3

Force balance

31 33

n
t(n)

32

12 P
21

23

A x

22

(n)
F

t
1 1 A 11An1 21An2 31An3 0

t1( n) 11n1 21n2 31n3

28

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) 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)

29

Projected Stresses
Normal stress (n) t n n n n
Shear stress

(n)

n 2

Principal stresses t n / /n

2 (n)

ijnn
i j
(n) t n n(n)

1 , 2, 3

Mean stress (hydrostatic pressure)

1
1
p m tr( ) (11 22 33 )
3
3
Stress deviator
s m 1 Idev :
11 m
s
12

13

12

11 m
23

13
23
11 m

Which stresses are frame indifferent?

Iijkl 31 (ik jl il jk )
Rank-4 identity tensor

Idev I 31 1 1
Rank-4 deviatoric identity tensor
30

Strain
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 = (u, v, w)
Displacement of Q & R
u
u
uR u
y
uQ u
x
y
x
v
v
vQ v
x
vR v
y
y
x
w
w
wQ w
x
wR w
y
x
y

R'

Q'
R

P'(x+u,y+v,z+w)

x
P(x,y,z)
Q
x

31

Strain
Strain is defined as the elongation per unit length
uy
y
P

ux

ux ux
lim

Textbook has different, but

Tensile11(normal)
strains
x 0 x inxx- and y-directions
more rigorous derivations
uy uy
22 lim

y 0 y
y

32

Shear Strain
Shear strain is the tangent of the change in angle between two
originally perpendicular axes

1 ~ tan 1

ux

uy

x
ux
2 ~ tan 2
y

y
P

y
Shear
strain
(change
of
angle)
lim

12

12

x 0

/2 12
1

uy

ux uy ux
lim

y 0 y
x
y

1
1 uy ux
12

2
2 x
y

Positive when the angle between two positive (or two negative)

33

Strain Tensor
Strain Tensor
ijei ej

11
Cartesian Components [ij ] 12

13
Vector notation
11
11

22
22
33
33
{ }

12
212
223
23

13
213

12
22

23

13
23
33

34

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
e 31 V 1
e Idev :

x3

eij ij 31 V ij

e22

e33
1
x2

1
e11

x1

Exercise: Write Idev in matrix-vector notation

35

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
stress

Fracture

Yield stress
Proportional
limit

Youngs
modulus
Strain
Necking
hardening

36

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)
37

Generalized Hookes Law cont.

Stress-strain relation
ij Dijklkl [ijkl (ik jl il jk )]kl kk ij 2ij
Volumetric strain:
Off-diagonal part:

kk 11 22 33 v
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
m ( 23 )kk K v

Substitute

so that we can separate

Bulkvolumetric
modulus part

K 23
Total deform. = volumetric + deviatoric deform.
38

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 :
Kv 1 2e
m 1 s

Deviatoric part
Volumetric part

e Idev :

Deviatoric strain

s Idev :

Deviatoric stress

Important for plasticity; plastic deformation only occurs in deviatoric part

volumetric part is always elastic
39

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)

u1,1

u2,2

11

22

33

212
223

13

u3,3

u1,2 u2,1

u
3,2
2,3
u u
3,1
1,3

12 + 21 = 212

11

22
33

D
12
23

13
You dont need 2 here
40

3D Solid Element cont.

Elasticity matrix
D K1 1 2Idev
Relation b/w
Lames constants
and Youngs modulus
and Poissons ratio

(3 2)
,
E
2( )

E
E

,
(1 )(1 2)
2(1 )

1
1

1
1
0
0

0

2
3
31
31

Idev

0
0

31 31 0 0 0

2
1
3 0 0 0
3
31 23 0 0 0

0
0 21 0 0

1
0
0 0 2 0

0
0 0 0 21

2
D
0
0
0

0
0
0

0
0
0

0
0
0

0
0

0
0
0
0

0
0
0

0
0

41

Plane Stress
thin platelike components parallel to the xyplane
the plate is subjected to forces in its plane only

13 = 23 = 33 = 0

11

{ } 22

12

1
E

1
2
1
0 0

0
0

1 (1 )

11

22

12

13 = 23 = 0, but 33 0

42

Plane Strain
the strains with a z subscript are all equal to zero
deformation in the zdirection is constrained, (i.e., u3 = 0)

13 = 23 = 33 = 0
can be used if the structure is infinitely long in the z
direction

11

{ } 22

12

1
0

(1 )(1 2)

13 = 23 = 0
33

1
2

11

22

12

11 22

(1 )(1 2)
43

1.4

MECHANICS OF
CONTINUOUS BODIES
44

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):

ijei ej

ij

Surface traction

11
21
31

12

22

32

13
23
33

t n n
Cauchys Lemma

X3

t n t n
t n n

t n n

e1
X1

e3
e2

X2
tn

45

Balance of Linear Momentum cont

Balance of linear momentum

( fb a) d n d d

[ ( fb a)] d 0

Divergence Thm

( fb a) 0
For a static problem

fb 0

ij,i fjb 0

Balance of angular momentum

x f b d x t n d

x a d

ij ji
46

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
and

e1

u0

on h

essential BC

(2)

t n

on s

natural BC

(3)

fb

X3

(1)

fb 0

X1

e3
e2

X2
t

Solution space

DA u [C2 ()]3 | u 0 on x h , n t on x s

47

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

48

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
2

( u) D : ( u)
Linear elastic material

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

displacement u (independent of path)
49

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

x3

Potential energy

fb
u1

u3
u2

h
fs

x2

x1
( u) U( u) W( u)
1
( u) : ( u) d u fb d s u t d
.

2
50

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
51

Example Uniaxial Bar

Strong form
EAu 0
u0
EAu(L) F

x [0, L]
x0

F
L

xL

Integrate twice: EAu(x) c x c

1
2
Fx
Apply two BCs: u(x)
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

(U W) EALc1 FL 0
dc1 dc1

c1

F
EA

u(x)

Fx
EA
52

Virtual Displacement Field

Virtual displacement (Space Z)
Small arbitrary perturbation (variation) of real displacement

1
d
u lim [( u ) ( u)]
( u )
0
d

u.

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

Space Z only includes homogeneous

essential BCs

Property of variation

In the literature, u is often used instead of

d(u)
du

dx
dx

53

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
d

( u) ( u )

x
D :

(u ) : D : (u) ( u) : D : (u )
( u ) : D : ( u) d

U( u; u )

1
2

a( u, u )

for all u Z

Variation of strain energy

d u u
u

x
d

Energy bilinear form

54

PMPE As a Variation cont.

Variation of work done by applied loads
W( u; u )

b
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 ?
55

Example Uniaxial Bar

Assumed displacement u(x) = cx

virtual displacement is in the same space with u(x):

Variation of strain energy

d 1 L
U
EA (u u)

0
d 2

dx

u(x) cx

1 L
2EA(u u)u dx
2 0

0 EAuu dx EALcc

Variation of applied load

d
W
F u(L) u(L)

d
PMPE

Fu(L) FLc

U W c(EALc FL) 0

Fx
u(x) cx
EA
56

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

( fb ) u d

If the above virtual work becomes zero for arbitrary , then it

satisfies the original equilibrium equation in a weak sense
57

Principle of Virtual Work cont

PVW

(ij,i fjb )uj d 0

ij,iuj d fjbuj d

u Z

Integration-by-parts

Divergence Thm

(ijuj ),i ijuj,i d

n u d
u d
Theboundary
i ij jis decomposed
ij j,iby

fjbuj d

fjbuj d

h s
uj 0 on h and niij tj on s

b
t
u
d

u
d

f
S j j
ij j,i
j uj d

58

Principle of Virtual Work cont

Since ij is symmetric
ijuj,i ijsym(uj,i ) ij ij
Weak Form of BVP

ij ij d

fjbuj d

1 ui uj
sym(ui,j )

ij

2 Xj Xi

tjuj d

u Z

Internal virtual work = external virtual work

Starting point of FEM

Symbolic expression
a( u, u ) l ( u )
Energy form:
Load form:

a( u, u )

l (u )

[K]{ d} {F }

u Z

FE equation

: d

u fb d s u t d

59

Example Heat Transfer Problem

Steady-State Differential Equation

T = T0

T


T
ky Q 0
kx

y
x
y
x

domain
A

Boundary conditions

T T0 on ST

dT
dT

n
k

n
k
x x
y y
n
dx
dy

qn

on Sq

Sq

ST

n = {nx,
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
k

k
x x x y y y d

TQ d S Tqn dSq,
q

T Z
60

Example Beam Problem

Governing DE
EI

d4 v
dx

f(x),

f(x)

x [0, L]

Boundary conditions for cantilevered beam

dv
d2 v
d3 v
v(0)
(0)
(L)
(L) 0
2
3
dx
dx
dx

Space of kinematically admissible displacement

dv

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

Integrate-by-part twice, and apply BCs

L

d2v d2v

0 EI dx2 dx2 dx 0 fv dx,

v Z
61

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

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

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 solution
F

62

1.5

FINITE ELEMENT METHOD

63

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

Piecewiselinear
approximation

Approximate
solution
x

Finite elements
Exact solution

64

Finite Elements
Types of finite elements

1D

2D

3D

Variational equation is imposed on each element.

1

dx

0.1

dx

0.2

0.1

dx L

0.9

dx

One element

65

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

u(x) a a x,

x xx

0 is composed
1
i of two nodes:
i 1
i-th element
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).

66

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

u(x)

xi1 x
(e)

14L2 43

ui

N1 (x)

x xi
(e)

L 3
12

ui1

N2 (x)

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

Solution1 for Element
e:

N1(x) and N2(x):

xi x xi1

Shape Function or Interpolation Function 67

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)

xi

N2(x)

xi+1

The solution is approximated by piecewise linear polynomial and its

gradient is constant within an element.

68

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

du
(1) 0

dx

Use PVW

Boundary conditions

d2u

0 dx2 p u dx 0

Space of kinematically
admissible displacements

Z u H(1) [0,1] u(0) 0

Integration-by-parts
du
u
dx

1
0

1
du du

dx pu dx
0 dx dx
0
1

This variational equation also satisfies at individual element level

69

1D Interpolation Functions
Finite element approximation for one element (e) at a time
(e)
(e)
u(e) (x) uN
(x)

u
N
(x)

d
i 1
i1 2

(e)

ui

ui1

N(e) N1 N2

Satisfies interpolation condition

u(e) (xi ) ui

u(e) (xi1 ) ui1

Interpolation of displacement variation (same with u)

(e)
(e)
u (e) (x) uN
(x)

u
N
(x)

d
i 1
i1 2
Derivative of u(x)

du(e) dN1

dx
dx

dN2 ui

dx ui1

(e)
L

1
(e)
L

ui

u i1

B(e) d(e)
70

Element-Level Variational Equation

Approximate variational equation for element (e)
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

xi

du

(x
)

dx i

du

(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

xj

B(e)T B(e) dx d(e)

2x2 matrix

du

(x
)
i

xj (e)T
dx
N
p(x) dx

xi
du

(xi1 )
dx

2x1 vector

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

du

(x
)
i

dx

du

(xi1 )
dx

71

Assembly
Need to derive the element-level equation for all elements
Consider Elements 1 and 2 (connected at Node 2)
k11

k21

k11

k21

k12

k22

k12

k22

(1)

(2)

(1)

u1 f1

u
f
2 2

u2

u3

Assembly

(1)
k11
(1)
k21

(1)
k12
(1)
(2)
k22
k11
(2)
k21

(2)

f2

f
3

du

(x
)
dx 1

du

(x2 )
dx

du

(x
)
dx 2

du

(x3 )
dx

0 u1

(2)
k12 u2

(2) u
k22
3

f1(1)
(2)
(1)
f2 f 2

f3(2)

du

(x
)
dx 1

du

(x )
dx 3

Vanished
unknown term

72

Assembly cont.

Assembly of NE elements (ND = NE + 1)

(1)
k11
(1)
k21

(1)
k12
(1)
(2)
k22
k11
(2)
k221

(2)
k12
(2)
(2)
k22
k11

0
ND ND

L
O
NE

k21

0 u
1

0 u2

0 u3

M M
(N )
k22E uN
ND 1

du

(x
)
dx 1

M
M

(N )
du

fN E

(x
)
dx N
ND 1
ND 1

f1(1)

(1)
(2)
f2 f2
(3)
(2)
f3 f 3

[K]{ q} {F }

Coefficient matrix [K] is singular; it will become nonsingular after applying boundary conditions

73

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
3 3
1 1 1
(e)
k

22 L(e) 1 1 3 3 , (e 1,2,3)

RHS (p(x) = x)
{f

(e)

N1 (x)
1 x i1 x(xi1 x)
} p(x)
dx
dx (e) x

xi
i
x(x x i )
L
N2 (x)

xi xi1

6
(e) 3
L
, (e 1,2,3)
x
x
i i1
6
3
xi1

74

Example cont.
RHS cont.

f1(1)
(1)
f2

1

54

(2)
1
1 4
f 2
,
(2)
,
54
2
5
f3

Assembly

3 - 3
0
0 u1

- 3 3 + 3 - 3
0 u2

=
0
- 3
3 +3
- 3 u3

0
- 3
3
0
u4

Apply boundary conditions

1 du
(0)
54 dx

2
4
+

54 54

7
5
+

54 54
8
du
+ (1)

54 dx

(3)
f3
(3)

f
4

1 7

8
54

Element 1
Element 2
Element 3

Deleting 1st and 4th rows and columns

6 3 u2
3 6 u

1

9

1
2

u2
u3

4
81
5
81
75

EXAMPLE cont.
Approximate solution
4
x,

27
1
1
4
u(x)

x ,

3
81 27
5
5
2

x ,

81 27
3

1
3
1
2
x
3
3
2
x1
3

0x

Exact solution
1
u(x) x 1 x2
6

Three element solutions are poor

Need more elements

76

3D Solid Element
Isoparametric mapping

Build interpolation functions on the reference element

Jacobian: mapping relation between physical and reference elem.

Interpolation and mapping

u( )

NI ()uI

x( )

I 1

NI ()

I 1

1
(1 I )(1 I )(1 I )
8
x8

x7

x5

(1,
1,1)
(1, 1,1)

x6
x

x3
x1

Same for mapping

and interpolation

NI ()xI

x4

x3
x2

x2
(a) Finite Element

(1, 1,
1)

(
1,1,1)
(1,1,1
)

(1,1,
1)
(1,1,1)

(b) Reference
Element

77

3D Solid Element cont.

Jacobian matrix
8
dNI ()
dx
J33
xI
d I 1
d
Derivatives of shape functions
NI

NI

NI

NI NI

NI
x1

NI
x2

J : Jacobian

NI

x3

x1

x2

x3

x1

x2

x3

x1

x2

x3

NI NI

J 1
x

Jacobian should not be zero anywhere in the element

Zero or negative Jacobian: mapping is invalid (bad element shape)
78

3D Solid Element cont.

Displacement-strain relation
( u)

NI,1

BI uI

I 1

( u )

BI uI

I 1

BI

NI,2

NI,2
0

N
I,3

NI,1
NI,3

NI,i

0
NI,3

NI,2
NI,1

Ni

xi

79

3D Solid Element cont.

Transformation of integration domain
1

d 1 1 1 J ddd
Energy form
a( u, u )

I 1 J 1

uIT

u { d }T [k]{ d}
T
B
DB
J
d

1 1 1 I J
J

Load form
l (u )

T 1 1 1
uI
NI ()fb
1 1 1

I 1

J ddd { d }T { f }

Discrete variational equation

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

{ d } Zh
80

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
1

NG

if( i )
1 f() d
i1

NG: No. of integ. points; i: integ. point; i: integ. weight

i and i are chosen so that the integration is exact

for (2NG 1)-order polynomial

Works well for smooth function

Integration domain is [-1, 1]
81

Numerical Integration cont.

Multi-dimensions
1

NG NG

ijf(i, j )
1 1 f(, ) dd
i1 j1
1

NG NG NG

ijk f(i, j, k )
1 1 1 f(, , ) ddd
i1 j1 k 1
NG
1
2
3
4
5

Integration
Points (i)
0.0
.5773502692
.7745966692
0.0
.8611363116
.3399810436
.9061798459
.5384693101
0.0

Weights (i)
2.0
1.0
.5555555556
.8888888889
.3478546451
.6521451549
.2369268851
.4786286705
.5688888889

(a)
11

(b)
22

(c) 33

82

ELAST3D.m
A module to solve linear elastic problem using NLFEA.m
Input variables for ELAST3D.m
Variable
ETAN
UPDATE
LTAN

Array size
(6,6)
Logical variable
Logical variable

Meaning
Elastic stiffness matrix Eq. (1.81)
If true, save stress values
If true, calculate the global stiffness matrix

NE
NDOF
XYZ
LE

Integer
Integer
(3,NNODE)
(8,NE)

Total number of elements

Dimension of problem (3)
Coordinates of all nodes
Element connectivity

83

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;

84

% 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

85

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(81): shape functions,
SF=zeros(8,1);
DSF=zeros(3,8);
for I=1:8
GDSF (38): shape functions derivatives
XP = XNODE(1,I);
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

86