Chapter 3 - English

FINITE ELEMENT METHOD
Chapter 3. FEM FOR 2D AND AXISYMMETRIC SOLIDS
CHAPTER 3. FEM FOR 2D SOLIDS
Assoc. Prof. Dr. Tich Thien TRUONG 1 HCMUT 2007

CONTENTS
3.1 FEM FOR 2D SOLIDS
 INTRODUCTION
 LINEAR TRIANGULAR ELEMENTS
– Field variable interpolation
– Shape functions construction
– Using area coordinates
– Strain matrix
– Element matrices
 LINEAR RECTANGULAR ELEMENTS
– Shape functions construction
– Strain matrix
– Gauss integration
– Evaluation of me
 LINEAR QUADRILATERAL ELEMENTS

– Coordinate mapping
– Strain matrix
– Remarks
 HIGHER ORDER ELEMENTS
 COMMENTS (GAUSS INTEGRATION)
3.2. FEM FOR AXISYMMETRIC PROBLEMS
Structures of Revolution
─ Introduction
─ The problem variables
─ Governing equations

─ Variational formulation
─ Treating plane strain as a limit case
Axisymmetric Iso-P Elements

─ Introduction
─ Isoparametric definition
─ The Element Stiffness Matrix
─ Consistent Node Forces for Body Loads

3.1 FEM FOR 2D SOLIDS
INTRODUCTION
 2D solid elements are applicable for the
analysis of plane strain and plane stress
problems.
 A 2D solid element can have a triangular,
rectangular or quadrilateral shape with
straight or curved edges.
 A 2D solid element can deform only in the
plane of the 2D solid.
 At any point, there are two components in the
x and y directions for the displacement as well
as forces.
INTRODUCTION
 For plane strain problems, the thickness of the
element is unit, but for plane stress problems,
the actual thickness must be used.
 In this course, it is assumed that the element
has a uniform thickness h.
 Formulating 2D elements with a given variation
of thickness is also straightforward, as the
procedure is the same as that for a uniform
element.

2D solids – plane stress and plane strain

z
y fy
y fx
fx
x
Plane stress Plane strain

LINEAR TRIANGULAR ELEMENTS

 Less accurate than quadrilateral elements
 Used by most mesh generators for complex
geometry
 A linear triangular element:
y, v 3 (x3, y3)
(u3, v3)
fsy
fsx
A
2 (x2, y2)
(u2, v2)
1 (x1, y1)
(u1, v1) x, u

Field variable interpolation U h ( x, y )  N ( x, y )d e

 u1  
v   displacements at node 1
 1 
u2  
where de     displacements at node 2
 v2  
u3  
   displacements at node 3
 v3   y, v 3 (x3, y3)
(u3, v3)
fsy
 N1 0 N2 0 N3 0 fsx
N A
0 N1 0 N2 0 N3 
  
   2 (x2, y2)
Node 1 Node 2 Node 3
(u2, v2)
(Matrix of Shape 1 (x1, y1)
functions) (u1, v1) x, u

Shape functions construction
Assume,
N1  a1  b1 x  c1 y N i  ai  bi x  ci y i= 1, 2, 3
N 2  a2  b2 x  c2 y ai 
 
or Ni  
1 x y  bi   pT 
  
N 3  a3  b3 x  c3 y pT ci 


Delta function property:

N1 ( x1 , y1 )  1
1 for i  j
Ni ( x j , y j )   N1 ( x2 , y2 )  0
0 for i  j
N1 ( x3 , y3 )  0
Therefore, N1 ( x1 , y1 )  a1  b1 x1  c1 y1  1
N1 ( x2 , y2 )  a1  b1 x2  c1 y2  0
N1 ( x3 , y3 )  a1  b1 x3  c1 y3  0
x2 y3  x3 y2 y2  y3 x3  x2
Solving, a1  , b1  , c1 
2 Ae 2 Ae 2 Ae


1 x1 y1
1 1 1
Ae  P  1 x2 y2  [( x2 y3  x3 y2 )  ( y2  y3 ) x1  ( x3  x2 ) y1 ]
2 2 2
1 x3 y3
Area of triangle Moment matrix

Substitute a1, b1 and c1 back into N1 = a1 + b1x + c1y:
1
N1  [( y2  y3 )( x  x2 )  ( x3  x2 )( y  y2 )]
2 Ae


Similarly,
1
N 2 ( x1 , y1 )  0 N2  [( x3 y1  x1 y3 )  ( y3  y1 ) x  ( x1  x3 ) y ]
2 Ae
N 2 ( x2 , y2 )  1
1
N 2 ( x3 , y3 )  0  [( y3  y1 )( x  x3 )  ( x1  x3 )( y  y3 )]
2 Ae
1
N 3 ( x1 , y1 )  0 N3  [( x1 y2  x1 y1 )  ( y1  y2 ) x  ( x2  x1 ) y ]
2 Ae
N 3 ( x2 , y2 )  0
1
N 3 ( x3 , y3 )  1  [( y1  y2 )( x  x1 )  ( x2  x1 )( y  y1 )]
2 Ae

N i  ai  bi x  ci y
1
where ai  ( x j yk  xk y j ) , i = 1, 2, 3
2 Ae
1 j, k determined from cyclic permutation
bi  ( y j  yk )
2 Ae
i = 1, 2, 3
1 i
ci  ( xk  x j )
2 Ae
k j
k = 3, 1, 2 j = 2, 3, 1

Using area coordinates
 Alternative method of constructing shape functions

1 x y
1 1
 2-3-P: A1  1 x2 y2  [( x2 y3  x3 y2 )  ( y2  y3 ) x  ( x3  x2 ) y ]
y 2 2
k, 3
1 x3 y3
A1
L1 
Ae
A2
A1
Similarly,  3-1-P A2 L2 
P
Ae
A3
i, 1  1-2-P A3 L3 
j, 2 Ae
x

Using area coordinates

Partitions of unity: L1  L2  L3  1
A2 A2 A3 A2  A2  A3
L1  L2  L3     1
Ae Ae Ae Ae
Delta function property: e.g. L1 = 0 at if P at nodes 2 or 3
Therefore, N1  L1 , N 2  L2 , N 3  L3
U h ( x , y )  N ( x , y )d e

 
Strain matrix  xx 
u
 0
 x  x 
v  
 yy    LU where L 0
 y
 y 
u v  
 xy  
 y  x  
 y x 
  LU  LNd e  Bd e
 
 0 b1 0 b2 0 b3 0 
 x 
   B   0 c1 0 c2 0 c3 
B  LN   0  
N c1 b1 c2 b2 c3 b3 
 y 
 
  (constant strain element)
 y x 
Element matrices
h
k e   B cBdV   (  dz )BT cBdA   hBT cBdA
T
0
Ve Ae Ae
Constant matrix
k
e  hAe BT cB
With c is the matrix of elastic constants.
For plane stress problem of isotropic material:
1  0 
E  1 0 
c  
1  
2
1  
 0 0 2 
Element matrices
For plane strain problem of isotropic material:
  
 1 1 
0 
 
c
1 E  
1 0 
1   1  2  1   

 0 1  2 
0
 2 1    

Element matrices
After computing, we get:
 k11 k12 k13 k14 k15 k16 

 k k k24 k25 k26 
 22 23
C1h  k33 k34 k35 k36 

ke   
4 Ae  k44 k45 k46 
 k55 k56 
 
 sym k66 
With C1 is the elastic constant depending on the stress
state.
Assoc. Prof. Dr. Tich Thien TRUONG HCMUT 2007
Element matrices
Plane stress:
E
C1 
1  2
Plane strain:
C1 
1   E
1  1  2 
and the components kij are computed as following:

Element matrices
k14 = C2xikyjk + yikxjk

Element matrices
With C2 and  are also the elastic constants depending
on the stress state.
1 
Plane stress: C2   ;  
2
 1  2
Plane strain: C  ; 
1  2 1  
2
xij  xi  x j
y jk  y j  yk
Element matrices
h
m e    N NdV 
T
 dx  N T NdA   h  N T NdA
0
Ve Ae Ae
For elements with uniform density and thickness,
 N1 N1 0 N1 N 2 0 N1 N 3 0 
 0 N1 N1 0 N1 N 2 0 N1 N 3 

 N 2 N1 0 N2 N2 0 N 2 N3 0 
me  h    dA
Ae  0 N 2 N1 0 N2 N2 0 N 2 N3 
 N 3 N1 0 N3 N 2 0 N3 N3 0 
 
 0 N 3 N1 0 N3 N 2 0 N 3 N 3 
Eisenberg and Malvern (1973):

m !n ! p !
A L1 L2 L3 dA  (m  n  p  2)! 2 A
m n p

Element matrices
2 0 1 0 1 0 y, v 3 (x3, y3)
 2 0 1 0 1  (u3, v3)
  fsy
fsx
 hA  2 0 1 0 A
me   
12  2 0 1 
2 (x2, y2)
(u2, v2)
 sy. 2 0 1 (x1, y1)
  (u1, v1) x, u
 2
 f sx  0
fe   [N]T
  dl 0
 f sy 
2 3
l  
1   fx 

Uniform distributed load on side 23: fe  l23  
2  fy 
 fx 
 
 fy 
 
Element matrices
If uniform distributed load are applied on three sides, we
use superposition principle:
 f sx12 .l12  f sx31.l31 

 12 
 f sy .l12 
31
f sy .l31 

1  sx .l12 
f 12
f sx23 .l23 
fe   12 
2  f sy .l12  23
f sy .l23 
 f sx31.l31  f sx .l23 
23
 31 
 f sy .l31 
23
f sy .l23 

Load vector of element inferred volume load {g} = {gx gy}T:
 N1 g x   gx 
N g  g 
 1 y  y
T  x
g  N 2 g x  Ae h  g x 
fe   [N]   dV     hdA   
gy  N g 3 g y 
Ve
A 2 y
 N3 g x   gx 
   
 N3 g y   g y 

Stress vector of element:

 xx 
  e   yy    D     c  B  d e   S  e d e
 
 xy 
With stress computing matrix: S e  c  B
with:
 y jk 0  yik 0 yij 0 
1  
 B   0  x jk 0 xik 0  xij 
2A
  x jk y jk xik  yik  xij yij 
Assoc. Prof. Dr. Tich Thien TRUONG HCMUT

28 2007
1 C2 0   y jk 0  yik 0 yij 0 
C1    
  S e  C2 1 0  0  x jk 0 xik 0  xij 
2A  
 0 0     x jk y jk xik  yik  xij yij 
 y jk C2 x jk  yik C2 xik yij C2 xij 

C1  
  S e   C2 y jk  x jk C2 yik xik C2 yij  xij 
2A
  x jk  y jk  xik  yik  xij  yij 

29 2007
LINEAR RECTANGULAR
ELEMENTS
 Non-constant strain matrix
 More accurate representation of stress and strain
 Regular shape makes formulation easy


Consider a rectangular element
 u1   y, v
v   displacements at node 1 
 1  4 (x4, y4) 3 (x3, y3)
u2   (u4, v4) (u3, v3)
   displacements at node 2 
 v2   fsy
de    2b fsx
u3  
 v3   displacements at node 3 2a
   2 (x2, y2)
1 (x1, y1)
u4   (u1, v1) (u2, v2)
v   displacements at node 4
 4 
x, u


y, v 4 (1, +1) 3 (1, +1)
 (u4, v4) (u3, v3)
4 (x4, y4) 3 (x3, y3)
(u4, v4) (u3, v3)

 fsy
2b fsx
2
2a 
1 (x1, y1) 2 (x2, y2)
(u1, v1) (u2, v2) 2
1 (1, 1) 2 (1, 1)
x, u (u1, v1) (u2, v2)
x  ( x1  a) y  ( y1  b) ,  and  are the natural coordinates

 , 
a b
 N1 0 N 2 0 N3 0 N 4 0 
U ( x, y )  N ( x, y )d e where
h
N   
0 N1
 0 N 2 0 N3 0 N 4 
 

  
 
(Interpolation) Node 1 Node 2 Node 3 Node 4


N1  14 (1   )(1   )
N 2  14 (1   )(1   )
N 3  14 (1   )(1   )
N j  14 (1   j )(1   j )
4 (1, +1) 3 (1, +1)
N 4  14 (1   )(1   ) (u4, v4) (u3, v3)
N 3 at node 1  14 (1   )(1   )  1  0

 1 
N 3 at node 2  14 (1   )(1   )  1  0
 1 Delta function 2
N 3 at node 3  14 (1   )(1   )  1  1 property 
 1
N 3 at node 4  14 (1   )(1   )  1  0 2

 1 2 (1, 1)
1 (1, 1)
4 (u1, v1) (u2, v2)
 N i  N1  N 2  N 3  N 4
i 1 Partition of
 14 [(1   )(1   )  (1   )(1   )  (1   )(1   )  (1   )(1   )] unity
 14 [2(1   )  2(1   )]  1

 
 





The chain rule yields:

f f x f y f f x f y
  ,  
 x  y   x  y 
Using Cramer’s rule, which involves the determinants of
matrices, we can obtain:
f y x f
   
f y x f
f   f  
 , 
x x y y x y
   
x y x y
   
The determinant in the denominator is the determinant of the

Jacobian matrix [J].
x y
  a 0
J     ab
x y 0 b
 
We now want to express the element strains as:

  LU  LNd e  Bd e
Where B must now be expressed as a function of  and .

   y  y  
 0       0 
 x   
  1  x  x  
B  LN   0 N   0   N
 y  J       
   x  x  y  y  
    
 y x           

Strain matrix
  1a 0 1
a 0 1
a 0 1
 a 0 
1 1 1 1 1 
B  LN   0  b 0  b 0 b 0 b 
4 1
 b 1
 a
1
 b
1 1 1 1 1 
 a
 a b a b 
Note: No longer a constant matrix!

Element matrices
x  ( x1  a) y  ( y1  b)
 ,   dxdy =Jdd = abdd
a b 1 1
Therefore, k e   hB cBdA  
T
 hB cB  J  d d
T
1 1
A
h 1 1
m e   N NdV    dx N NdA   h N NdA  
T T T
 abh NT Nd d
0 1 1
V A A
With c is the matrix of elastic constants.
For plane stress problem of isotropic material:
1  0 
E  1 0 
c  
1  
2
1  
 0 0 2 
Element matrices
For plane strain problem of isotropic material:
  
 1 1 
0 
 
c
1 E  
1 0 
1   1  2  1   

 0 1  2 
0
 2 1    

Element matrices
With C2 and  are also the elastic constants depending
on the stress state.
1  C2 1  
Plane stress: C2   ;   
2 2
 1  C2 1  2
Plane strain: C2  ;  
1  2 2 1  

Gauss integration
 For evaluation of integrals in ke and me (in practice)
1 m
In 1 direction: I   f ( )d   w j f ( j )
1
j 1
m Gauss points gives exact solution of polynomial

integrand of n = 2m - 1
nx ny
1 1
In 2 directions: I    f ( , )d d   wi w j f (i , j )
1 1
i 1 j 1

Gauss integration
m j , j wj Accuracy n
1 0 2 1
2 -1/3, 1/3 1, 1 3
3 -0.6, 0, 0.6 5/9, 8/9, 5/9 5
4 -0.861136, -0.339981, 0.347855, 0.652145, 7

0.339981, 0.861136 0.652145, 0.347855
5 -0.906180, -0.538469, 0.236927, 0.478629, 9
0, 0.568889, 0.478629,
0.538469, 0.906180 0.236927
6 -0.932470, -0.661209, 0.171324, 0.360762, 11
-0.238619, 0.238619, 0.467914, 0.467914,
0.661209, 0.932470 0.360762, 0.171324

Ta sử dụng tích phân Gauss để tính ke

Với trường hợp lấy 4
1 1
k e   hB cBdA  
T
 abhB T cBd d điểm Gauss như
1 1 đang xét thì nx = 2 và
A
ny = 2
nx ny nx ny
 abh w i w j B T cB  abh wi w j f (i , j ) 3 (1, +1)

4 (1, +1)
i 1 j 1 i 1 j 1
(u4, v4) (u3, v3)
Ta sử dụng 4 điểm gauss g(i,j) như trong hình

g(1,1) : ξ1 = -0,57735 , η1 = -0,57735 g(1,2) g(2,2)
g(2,1) : ξ2 = 0,57735 , η1 = -0,57735 2

g(2,2) : ξ2 = 0,57735 , η2 = 0,57735
g(1,1) 2 g(2,1)
g(1,2) : ξ1 = -0,57735 , η2 = 0,57735
1 (1, 1) 2 (1, 1)
Với các trọng số tương ứng tại mỗi điểm (u1, v1) (u2, v2)
w1 = 1, w2 = 1
 
k e1,1  abh w1w1B T 1,1cB1,1  abh  w1w1 f (1 ,1 )   abhf (1 ,1 )
  abh  w w B  cB     abh  w w f ( , )   abhf ( , )

T
k e1,2 1 2 1,2 1,2 1 2 1 2 1 2

T
k e 2,1 2 1 2,1 2,1 2 1 2 1 2 1

T
k e 2,2 2 2 2,2 2,2 2 2 2 2 2 2
Finally, the stiffness matrix:
k e  k e1,1  k e1,2  k e 2,1  k e 2,2

 abh  f 1 ,1   f 1 , 2   f  2 ,1   f  2 , 2  

ke(1,1)
   1
 
C2 
2,488.b2 a2 2,488. ab .
2 2,488

3b2 a2
    
0,5285
ab
C2 33  33
2 1  
  
3a2 b2
2C 
 2 
3 ab 
2
3a2
2,488

b2

0,5285
ab
    
C2 33  33 

 
 


2,488.        
  1  0,5285
a2 b2 ab
C2 33  33
2
3b2
2,488

a2
2C 
 2 
3 ab 
2 1  
  
3a2 b2
0,5285
ab
    
C2 33  33
2
3a2
2,488
1
b2


 
 


 
2,488.  
1 
b2 a2
2C
 2 

3 ab 
 2
3a2
2,488

b2
0
ab
    
,5285
C2 33  33
2 1
  
3a2 b2
 
2,488. 2
C
.
ab



 



  1
2,488.  
a2 b2
     
0,5285
ab
C2 33  33
2
3a2
2,488
1
b2
2,488.
C2 
.
ab
2 1  
  
3a2 b2



Ke(1,1)  


  1
2,488.  
b2 a2
2,488. 2
C 
.
ab
2 2,488

3b2 a2
0,5285
ab
    
C2 33  33 

 
 


  1
2,488.  
a2 b2

0,5285
ab
    
C2 33  33
2
3b2
2,488

a2


 
   1 2C2  
 2,488.     
 a2 b2 3 ab  
 
   1 
 2,488.   
 a b 
2 2

 

46 2007
ke(2,1)
0,1786 2,488
 2  2
 b a
2C2
3 ab
2 2,488
2
3b a
 2    
0,5285
ab
C2 33 33
2 1 
  2  2
3a b
2,488C20,1786
ab
2
3a2

0,1786 2
b

0,5285
ab
    
C2 33 33 


 



a
   
1 0,5285
2,488. 2 0,1786 2 
b ab
C2 33 33
2
3b2

2,488 2
a
 
2C
 2 
3 ab 
2 1 
  2  2
3a b  ab
   
0,5285
C2 33 33
2
3a 2
0,1786 2
b
1


 


0,1786 2,488
b 2
 2
a
2,488C20,1786
ab
2
3a2
b

0,1786 2
0,5285

 ab
   
C 2 33 33




2 1 

 2 2
3a
 b
2C2
3 ab


 
ke(2,1) 


a

b
   
1 0,14
2,488. 2 0,1786 2  C2 33 33
ab
2
3a2
b
1
0,1786 2
2C2
3 ab
2 1 
  2  2
3a b



 


0,1786 2,488
b2
 2
a
2C2
3 ab
2 2,488

3b2 a2
0,5285
ab
   
C 2 3 3 33 

 

 a
 1
2,488. 2 0,1786 2
b
   
0,14
 C2 33 33
ab
2
3b2
2,488

a2


 0,1786 2,488 2,488C20,1786 
  
 b2 a2 ab 
  1 
 sym 2,488. 2 0,1786 2 
 a b 

47 2007
ke(2,2)


  1
 
C2 
0,1786.b2 a2  0,1786. a.b
2 0,1786
3b 2
 2
a
    
0,14
ab
C2 33  33
2 1  
  2  2
3a b 
2C 
 2 
3 ab 
2
3a2

0,1786 2
b
0,14
ab
    
C2 33  33 


 


    
  1  0,14
0,1786. 2  2 
a b  ab
C2 33  33
2
3b2

0,1786 2
a
2C 
 2 
3 ab 
2 1  
  
3a2 b2  ab
    
0,14
C2 33  33
2
3a 2
0,1786 2
b
1


 


  1
0,1786. 2  2 
b a 
2C 
 2 
3 ab 
2
3a2

0,1786 2
b ab
    
0,14
C2 33  33
2 1  
  2  2
3a b 
0,1786. 2
C 
a.b


 


  1
0,1786. 2  2 
a b 
    
0,14
ab
C2 33  33
3
2
a2
0,1786 2
b
1
0,1786. 2
C 
a.b
2 1  

3a2 b2 


ke(2,2)  


  1
0,1786. 2  2 
b a 
0,1786. 2
C 
a.b
2 0,1786
3b2

a2
0,14
ab
C     
2 33  33 

 


  1
0,1786. 2  2 
a b  ab
    
0,14
C2 33  33
2
3b2
0,1786 2
a
 

 
   1 2C2  
 0,1786. 2  2    
 b a  3 ab  
   1 
 sym 0,1786. 2  2  
 a b  

48 2007
ke(1,2)
2,488 0,1786
 b2  a2
2C2
3 ab
2 0,1786
2
3b a
 2     
0,14
 C2 33  33
ab
2 1 
  2  2
3a b 
1,786C22,488
ab
2
3a 2
2,488 2

b
0,5285
ab
    
C2 33  33 

 
 



0,1786. 2 2,488 2
a b ab
    
1 0,5285
C2 33  33
3b
2
2

0,1786 2
a
2C2

3 ab 
2 1 
  2  2
3a b 
0,5285
ab
    
C2 33  33
2
3a 2
2,488 2
b
1


 


0,1786 2,488
b2

a2
1,786C22,488
ab
2
3a2
2,488

b2
0,5285
 ab 2

C    
3 3  33




2 1 

3a2 b2
2C2
3 ab


 

ke(1,2) 

2,488. 2 0,1786 2
a
1
b
0
,5285
ab
    
C2 33  33
2
3a2
2,488
1
b2
2 C
3 ab
2 

2  1


3a2 b2
 

 


0,1786 2,488
b2

a2
2C2
3 ab
2 0,1786

3b2 a2
0,5285
ab
    
C 2 3 3  33 

 



2,488. 2 0,1786 2
a
1
b
0,5285
ab
    
C2 33  33
2
3b2
2,488

a2


 0,1786 2,488 1,786C22,488 
  2 
 b2 a ab 
  1 
 sym 2,488. 2 0,1786 2 
 a b 

49 2007
[K]e =[K]e(1,1) + [K]e(2,1) + [K]e(2,2) + [K]e(2,1)
 k11 k12 k13 k14 k15 k16 k17 k18 

 k22 k23 k24 k25 k26 k27 k28 
 
 k33 k34 k35 k36 k37 k38 
 
C1h  k44 k45 k46 k47 k48 
 K e 
ab  k55 k56 k57 k58 
 
 k66 k67 k69 
 k77 k78 
 
 sym k88 

50 2007
   1  0,1786 2,488   1  2,488 0,1786  C1

k11  abh  2,488. 2  2     0,1786.  2  2  
  b a  b 2
a 2
 b a  b 2
a2  16
 C   2 C2   C   2 C2    C1
k12  abh  2,488. 2   0,1786. 2  
 a .b 3 ab a .b 3 ab  16
 2 2,488 2 2,488 2 0,1786 2 0,1786  C1
k13  abh  2  2  2  2  2  2
 2 
 3b a 3b a 3b a 3b a2  16
 ab  
 0,5285
C2 3  
 3   3  3 

0,5285
ab
 
C2 3  3  
 3  3  

C
k14  abh   1

 ab
  2    ab
 
2  
 0,14 C 3  3   3  3   0,14 C 3  3   3  3
  16



51 2007
 2 1   2 1   2 1   2  1   C
k15  abh    2  2     2  2     2  2     2  2   1
 3 a b  3 a b  3 a b  3  a b   16
 2  C    2,488C2  0,1786 2  C2    1,786C2  2,488  C1

k16  abh    2      16
 3  ab  ab 3  ab  ab 
 2  2  2  2  C
k17  abh  2  2,488 2  2  0,1786 2  2  0,1786 2  2  2,488 2  1
 3a b 3a b 3a b 3a b  16
 0,5285
 
ab
 C2 3   
3   3  3 

0,14
ab
 
C 2 3  3  
 3  3 

C 
k18  abh   1

 ab
  2    ab
 
2  
 0,14 C 3  3   3  3  0,5285 C 3  3   3  3
 16



52 2007
  1  1  1  1 C
k22  abh 2,488. 2  2   2,488. 2  0,1786 2  0,1786. 2  2   0,1786. 2  2,488 2  1
 a b  a b a b  a b  16
 2  2  2  2  C
k24  abh  2  2,488 2  2  2,488 2  2  0,1786 2  2  0,1786 2  1
 3b a 3b a 3b a 3b a  16
 2 1   2  1   2  1   2  1   C
k26  abh    2  2     2  2    2  2    2  2   1
 3 a b  3  a b  3  a b  3  a b   16
 2 1 2 1 2 1 2 1 C
k28  abh  2  2,488 2  2  0,1786 2  2  0,1786 2  2  2,488 2  1
 3a b 3a b 3a b 3a b  16

53 2007
Finite Element Method
1 1
K e  abh  B
T
.c . B .d  .d 
1 1
We consider:
G11 G12 G13 G14 G15 G16 G17 G18 
 G22 G23 G24 G25 G26 G27 G28 
 
 G33 G34 G35 G36 G37 G38 
 
G44 G45 G46 G47 G48 
BT c B  
 G55 G56 G57 G58 
 
 G66 G67 G68 
 đx G77 G78 
 
 G88 
Assoc. Prof. Dr. Tich Thien TRUONG

54 HCMUT 2007
Component Gij of above matrix are shown as the following

form:
C1    1 C1    1 C1    1 C1    1   1
2 2 2
G11   G17  
16a2 16b2 16b2 16a2
C1  C2      1   1 C1    1   C2  C2  
G12  G18   G23
16ab 16ab
C1    1 C1    1
2 2
C1    1 C1    1   1
2
G13   G22  
16a 2
16b 2 16b2 16a2
C1    1 C2    C2    C1    1   1 C1    1
2
G14  G24  2

16ab 16b 16a2
C    1   1 C1    1   1 C1    1   1 C1    1   1
G15  1  G 26  2

16a 2
16b 2
16b 16a2
C1C2    1   1 C1    1   1 C1    1 C1    1   1
2
G16   G28  
16ab 16ab 16b2 16a2

55 HCMUT 2007
We also get the following relations:
G38  G47  G56  G12 G33  G55  G77  G11

G44  G22  G88 G57  G13
G45  G67  G18 G36  G58  G14
G46  G28 G68  G24
G48  G26 G35  G17
G34  G78  G16 G36  G14
G44  G22  G88 G37  G15
We calculate 12 components: G11 G12 G13 G14 G15 G16 G17 G18
G22 G24 G26 G28. Other components are calculated from
above relations.

56 HCMUT 2007
Double integrals of components in the matrix
1 1
f (,) 
1 1
f (,).d.d
c1    1
2
C1    1
2
 1  
G11   C1  2  2 
16a2 16b2  3a 3b 
C1  C2      1   1 C1  C2   
G12 
16ab 4ab
C1    1
2
C1    1   1
 1  
G13  
C1  2  2 
16a 2
16b2  3a 6b 

57 2007
k38  k47  k56  k12 k33  k55  k77  k11 k35  k17
k44  k22  k66  k88 k57  k13 k36  k14
k45  k67  k18 k36  k58  k14 k37  k15
k46  k28 k68  k24
k48  k26 k34  k78  k16

58 2007
Double integrals of components in the matrix
1 1
f (,) 
1 1
f (,).d.d
C1    1 C2    C2    C1  C2   

G14   
16ab ab  4 
C1    1   1 C1    1   1   1 
G15   C1  2  2 
16a2 16b2  6b 6a 
C1C2    1   1 C1    1   1 C1  C2   
G16  
16ab 16ab 4ab

59 HCMUT 2007
Tích phân hai lớp từng phần tử trong ma trận G
1 1
f (,) 
1 1
f (,).d.d
C1    1
2
C1    1   1  1  
G17   C1  2  2 
 6a 3b 
2
16b 16a2
C1    1   C2  C2   C1  C2   

G18 
16ab

4ab
C1    1
2
C1    1
2  1  
G22  
C1  2  2 
16b2 16a2  3b 3a 

60 2007
Tích phân hai lớp từng phần tử trong ma trận G
1 1
f (,) 
1 1
f (,).d.d
C1    1   1 C1    1

2
 1  
G24   C1  2  2 
 6b 3a 
2
16b 16a2
C1    1   1 C1    1   1  1  

G26   C1  2  2 
16b2 16a2  6b 6a 
C1    1
2
C1    1   1
  1 
G28  
C1  2  2 
16b2 16a2  6a 3b 

61 2007
Element matrices
After computing, we get:
 k11 k12 k13 k14 k15 k16 k17 k18 
 k22 k23 k24 k25 k26 k27 k28 

 k33 k34 k35 k36 k37 k38 
 
C1h  k44 k45 k46 k47 k48 
ke 
4ab  k55 k56 k57 k58 
 
 k66 k67 k68 
 k77 k78 
 
 sym k88 
With C1 is the elastic constant depending on the stress
state.
b2   a 2   C2 2 b 2  a 2
k11  k16   ab k24  
3 4 6
C2   2  a 2
 b 2
C2  
k12  k
ab 17   k25   ab
4 6 4
 a 2  2b 2 C2   a 2  b 2
k13  k18   ab k26  
6 4 6
C2   a 2  b 2 C2  
k14  ab k22  k27  ab
4 3 4
 a 2  b2 C2    b 2  2a 2
k15   k23   ab k28 
6 4 6

63 2007
4 b2   a 2  2   a 2  2b 2 
k11  ; k12  ab    C2  ; k13 
3 3
2 b2   a 2 
k14   C2    ab; k15   ; k16  ab    C2 
3
2  b 2  2 a 2  4  a 2  b 2 
k17  ; k18  ab    C2  ; k22 
3 3
2  a 2  2 b 2  2  a 2  b 2 
k23  k18 ; k24  ; k25  k16 ; k26  
3 3

2   b 2  2a 2 
k27  k14 ; k28  ; k33  k11 ; k34  k16 ; k35  k17 ;
3
k36  k14 ; k37  k15 ; k38  k12 ; k44  k22 ; k45  k18 ; k46  k28
k47  k12 ; k48  k26 ; k55  k11 ; k56  k12 ; k57  k13 ; k58  k14
k66  k22 ; k67  k18 ; k68  k24 ; k77  k11 ; k78  k16 ; k88  k22

Evaluation of me
1 1
mij   hab   N i N j d d
1 1
 hab 1 1

16 
1
(1  i )(1   j )d  (1  i )(1   j )d
1
 hab
 (1  13 i j )(1  13 i j )
4
 hab  hab
E.g. m33  (1  3  1 1)(1  3  1 1)  4 
1 1
4 9
Note: In practice, Gauss integration is often used

Evaluation of me
4 0 2 0 1 0 2 0
 4 0 2 0 1 0 2 
 
 4 0 2 0 1 0
 
 hab  4 0 2 0 1
me 
9  4 0 2 0
 
 4 0 2
 sy. 4 0
 
 4 

Element matrices
Load vector of element inferred by uniform volume load {g} =
{gx gy}T: N g  g 
1 x x
N g  g 
 1 y  y
 N2 g x   gx 
1 1    
T  x
g  N2 g y  gy 
fe   [N]   dV      habd d  hab  
g y  Ng  gx 
Ve
1 1  3 x 
 N3 g y  gy 
   
 N4 g x   gx 
 N4 g y  gy 
   
Element matrices
For uniform distributed load on the side 23:
 f sx 
fe   [N]T
  dl
l 2 3 f
 sy 
y, v
0 0 
4 (x4, y4) 3 (x3, y3)
0 0 (u4, v4) (u3, v3)
     fsy
 f sx   f sx  2b fsx
   
1  f sy   f sy  2a
 l23    b   1 (x1, y1) 2 (x2, y2)
2  f sx   f sx  (u2, v2)
(u1, v1)
 f sy   f sy 
    x, u
0 0
0 0
   
Element matrices
If uniform distributed load are applied on four sides, we
use superposition principle:
 f sx12 .l12  f sx41.l41   f sx12 .a  f sx41.b 
 12   12 41 
 f sy .l12  f sy .l41   f sy .a 
41
f sy .b 
 f sx12 .l12  f sx23 .l23   f sx12 .a  f sx23 .b 
 12   12 23 
1  f sy .l12  f sy .l23   f sy .a 
23
f sy .b 
fe   23    23 34 
2  f sx .l23  f sx .l34   f sx .b 
34
f sx .a 
 f sy23 .l23  f sy34 .l34   f sy23 .b  f sy34 .a 
 34   34 41 
 f sx .l34  f sx .l41   f sx .a 
41
f sx .b 
 f 34 .l  f sy41.l41   f sy34 .a  f sy41.b 
 sy 34
Stress vector of element:

 xx 
  e   yy    D     c  B  d e   S  e d e
 
 xy 
With stress computing matrix:  S  e  c  B
  y4 C2 x2 y4 C2 x1  y1 C2 x1 y1 C2 x2 
C1 
 e
S  C 2 y 4  x2 C2 y 4 x1 C2 y1  x1 C2 y1 x2 
4ab  
  x2  y4  x1  y4  x1  y1  x2  y1 
with : xi  xi  x; yi  yi  y;

71 2007
LINEAR QUADRILATERAL
ELEMENTS
 Rectangular elements have limited application
 Quadrilateral elements with unparallel edges are
more useful
 Irregular shape requires coordinate mapping
before using Gauss integration

Coordinate mapping
y 4 (x4, y4)
3 (x3, y3)

4 (1, +1) 3 (1, +1)

2 (x2, y2)
1 (x1, y1) 1 (1, 1) 2 (1, 1)
x
Physical coordinates Natural coordinates
U h ( , )  N ( , )d e (Interpolation of displacements)
X( , )  N( , )x e (Interpolation of coordinates)

Coordinate mapping
 x1  
y   coordinate at node 1
X( , )  N( , )x e  1 
 x2  
x    coordinate at node 2
where X    ,  y2  
xe   
 y  x3  
 coordinate at node 3
 y3  
 
 x4  
N1  (1   )(1   )
1
y 
 4
 coordinate at node 4
4

N 2  14 (1   )(1   ) 4
N 3  14 (1   )(1   ) x   N i ( , ) xi
i 1
N 4  14 (1   )(1   )
4
y   N i ( , ) yi
i 1

Coordinate mapping
4
On the side 23, substitute  = 1 into x   N i ( , ) xi
i 1
x  12 (1   ) x2  12 (1   ) x3 x  12 ( x2  x3 )  12  ( x3  x2 )
or
y  12 (1   ) y2  12 (1   ) y3 y  12 ( y2  y3 )  12  ( y3  y2 )
( x3  x2 )
Eliminating , y  {x  12 ( x2  x3 )}  12 ( y2  y3 )
( y3  y2 )
y 4 (x4, y4)
3 (x3, y3)

4 (1, +1)
3 (1, +1)

2 (x2, y2)
1 (x1, y1) 1 (1, 1) 2 (1, 1)
x

Strain matrix
N i N i x N i y  N i   N i 
      x 
 x  y 
or   J 
N i N i x N i y  N i   N i 
      y 
 x  y   
 x y 
   
where J    (Jacobian matrix)
 x y 
    N 4   x1 y1 
  N1 N 2 N3
   y2 
     x2
Since X ( , )  N ( ,  ) x e
, J  
 N1 N 2 N3 N 4   x3 y3 
   
      x4 y4 

Strain matrix
 N i   N i  (Relationship between differentials of
 x    
Therefore,  N   J  
1 shape functions w.r.t. physical
 i  N i  coordinates and differentials w.r.t.
  
 y    natural coordinates)
Replace differentials of Ni w.r.t. x and y with

differentials of Ni w.r.t.  and 
w.r.t. means with respect to

  x 0 
B  LN   0  y  N
 y  x 

Element matrices
Murnaghan (1951): dA = |J|dd
1 1
ke    hB cB J d d
T
1 1
h
m e   N NdV    dx NT NdA   hNT NdA
T
0
V A A
1 1
  h NT N J d d
1 1

Remarks
 Shape functions used for interpolating the coordinates
are the same as the shape functions used for
interpolation of the displacement field. Therefore, the
element is called an isoparametric element.
 Note that the shape functions for coordinate
interpolation and displacement interpolation do not have
to be the same.
 Using the different shape functions for coordinate
interpolation and displacement interpolation,
respectively, will lead to the development of so-called
subparametric or superparametric elements.

HIGHER ORDER ELEMENTS

 Higher order triangular elements
(0,0,p)
nd = (p+1)(p+2)/2
(1,0,p1) (0,1,p1)
Node i, IJ K  p
(2,0,p2)
Argyris, 1968 :
L1
L2
i (I,J,K)
N i  lII ( L1 )lJJ ( L2 )lKK ( L3 )
L3 (0,p1,1)  ( L  L 0 )( L  L 1 ) ( L  L (  1) )
l ( L ) 
( L I  L 0 )( L I  L 1 ) ( L I  L (  1) )
(p,0,0) (p1,1,0) (0,p,0)


 Higher order triangular elements (Cont’d)
y, v 3 N1  N 2  N 2  (2 L1  1) L1
N 4  N 5  N 6  4 L1 L2
5 1
6 y, v 3 N1  N 2  N 3  (3L1  1)(3L1  2) L1
2
7 9
2 8 N 4  N9  L1 L2 (3L1  1)
4 6 2
1 9 N10  27 L1 L2 L3
10
x, u 2
5
1 4
Cubic element x, u
Quadratic element


 Higher order rectangular elements
Lagrange type:
 [Zienkiewicz et al., 2000]
(0,m) (n,m)
N i  N I1D N 1J D  lIn ( )lJm ( )

i(n,m)
(   0 )(  1 ) (   k 1 )(   k 1 ) (   n )
 lk ( ) 
n
( k   0 )( k  1 ) ( k   k 1 )( k   k 1 ) ( k   n )
0
(0,0) (n,0)


 Higher order rectangular elements (Cont’d)
4 7 3

6
(nine node quadratic element)
8 9
1
N 5  N 31D ( ) N11D ( )   (1   )(1   )(1   )
2
1
1 5 2 N 6  N 21D ( ) N 31D ( )   (1   )(1   )(1   )
1 2
N1  N11D ( ) N11D ( )   (1   ) (1   )
4 1
1 N 7  N 31D ( ) N 21D ( )  (1   )(1   )(1   )
N 2  N 21D ( ) N11D ( )    (1   ) (1   ) 2
4
1
1
N3  N 21D ( ) N 21D ( )   (1   )(1   )
N8  N11D ( ) N11D ( )    (1   )(1   )
4 2
1 N 9  N 31D ( ) N 31D ( )  (1   2 )(1   2 )
N 4  N11D ( ) N 21D ( )    (1   )(1   )
4

 Serendipity type:
=1 
4 7 3
 
N j  14 (1   j )(1   j )( j   j  1), j  1, 2, 3, 4
6
8 0 N j  12 (1   2 )(1   j ) , j  5, 7

N j  12 (1   j )(1   2 ) , j  6,8
1 5 2
=1
(eight node quadratic element)

 N j  321 (1   j )(1   j )(9 2  9 2  10)

4 10 9 3
for corner nodes j  1, 2, 3, 4
11 8 N j  329 (1   j )(1   2 )(1  9 j )

for side nodes j  7, 8, 11, 12 where  j  1 and  j   13
12 7
N j  329 (1   j )(1   2 )(1  9 j )
1 5 6 2 for side nodes j  5, 6, 9, 10 where  j   13 and  j  1
(twelve node cubic element)

ELEMENT WITH CURVED EDGES

3 3
5
5
6
6
2
4 2 4
1
1
4 7 3
3
4 7
8 6 6
8
1 2
5
1 5 2

COMMENTS (GAUSS INTEGRATION)

 When the Gauss integration scheme is used, one has to
decide how many Gauss points should be used.
 Theoretically, for a one-dimensional integral, using m
points can give the exact solution for the integral of a
polynomial integrand of up to an order of (2m-1).
 As a general rule of thumb, more points should be used
for a higher order of elements.

COMMENTS (GAUSS INTEGRATION)

 Using a smaller number of Gauss points tends to
counteract the over-stiff behaviour associated with the
displacement-based method.
 Displacement in an element is assumed using shape
functions. This implies that the deformation of the
element is somehow prescribed in a fashion of the
shape function. This prescription gives a constraint to
the element. The so-constrained element behaves
stiffer than it should. It is often observed that higher
order elements are usually softer than lower order
ones. This is because using higher order elements
gives fewer constraint to the elements.

COMMENTS ON GAUSS INTEGRATION
 Two Gauss points for linear elements, and two or three

points for quadratic elements in each direction should
be sufficient for most cases.
 Most of the explicit FEM codes based on explicit
formulation tend to use one-point integration to achieve
the best performance in saving CPU time.

CASE STUDY
 Side drive micro-motor

CASE STUDY
10N/m
10N/m Elastic Properties of Polysilicon
Young’s Modulus, E 169GPa
Poisson’s ratio,  0.262

10N/m
Density,  2300kgm-3

CASE STUDY
Analysis no. 1: von Mises stress

distribution using 24 bilinear
quadrilateral elements (41 nodes)
CASE STUDY

CASE STUDY

CASE STUDY

distribution using 24 eight-nodal,
quadratic elements (105 nodes)
CASE STUDY

distribution using 192 three-nodal,
triangular elements (129 nodes)
Total number Maximum

Analysis Number / type
of nodes in von Mises
no. of elements
model Stress (GPa)
24 bilinear,
1 quadrilateral
41 0.0139
CASE 2
96 bilinear,
quadrilateral
129 0.0180
STUDY 144 bilinear,

3 quadrilateral
185 0.0197
24 quadratic,
4 quadrilateral
105 0.0191
5 192 linear, 129 0.0167

triangular
3.2. Axisymmetric Element

3.2.1. Structures of Revolution
3.2.1.1. Introduction
In the Introduction to Finite Element Methods (IFEM) course
two-dimensional problems were emphasized. The
axisymmetric problem considered in this and following two
Chapters of this course provides a “bridge” to the treatment of
three-dimensional elasticity. Besides its instructional value, the
treatment of axisymmetric structures has considerable
practical interest in aerospace, civil, mechanical and nuclear
engineering.

3.2.1.1.1. The Axisymmetric Problem

The axisymmetric problem deals with the analysis of
structures of revolution under axisymmetric loading. A
structure of revolution or SOR is generated by a generating
cross section that rotates 3600 about an axis of revolution, as
illustrated in Figure 2.51. Such structures are said to be
rotationally symmetric.
The technical importance of SOR’s is considerable because of
the following practical considerations:
1. Fabrication: axisymmetric bodies are usually easier to
manufacture than bodies with more complex geometries.
Think for example of pipes, piles, axles, wheels, bottles, cans,
cups, nails.
3.2.1.

2. Strength: axisymmetric configurations are often optimal in

terms of strength to weight ratio because of the favorable
distribution of the structural material. (Recall that the strongest
columns and shafts, if wall buckling is ignored, have annular
cross sections).
3. Multipurpose: hollow axisymmetric bodies can assume a
dual purpose as both structure and shelter, as in containers,
vessels, tanks, rockets, etc.
Perhaps the most important application of SORs is
containment and transport of liquid and gasses. Specific
examples of such structures are pressure vessels,
containment vessels, pipes, cooling towers, and rotating
machinery (turbines, generators, shafts, etc.).

3.2.2.

But a SOR by itself does not necessarily define an

axisymmetric problem. It is also necessary that the loading, as
well as the support boundary conditions, be rotationally
symmetric. This is illustrated in Figure 2.52 for loads.
If these two conditions are met:

axisymmetric geometry and axisymmetric loading
the response of the structure is axisymmetric (also called
radially symmetric). By this is meant that all quantities of
interest in structural analysis: displacement, strains, and
stresses, are independent of the circumferential coordinate
defined below.

Remark 3.2.1.1
A SOR under non-axisymmetric loading can be treated by a Fourier

decomposition method. This involves decomposing the load into a
Fourier series in the circumferential direction, calculating the
response of the structure to each harmonic term retained in the series,
and superposing the results. The axisymmetric problem considered
here may be viewed as computing the response to the zero-th
harmonic. This superposition technique, however, is limited to linear
problems.

Figure 3.2.3. Axisymmetric FE analysis of a typical rocket nozzle (carried out

by E. L. Wilson at Aerojet Corporation, circa 1963). Figure from
paper cited in footnote 1.

3.2.1.1.2. Some SOR Examples

Rocket Analysis. The analysis of axisymmetric structures by the
Finite Element Method (FEM) has a long history that may be
traced back to the early 1960s. Recall that the FEM originated in
the aircraft industry in the mid 1950s. Aircraft are not SORs, but
several structures of interest in aerospace are, notably rockets.
As the FEM began to disseminate throughout the aerospace
industry, interest in application to rocket analysis prompted the
development of the first axisymmetric finite elements during the
period 1960-1965. These elements were of shell and solid type.
The first archival-journal paper on axisymmetric solid elements,
by E. L.Wilson, appeared in 1965. Figure 2.53 shows a realistic
application to a rocket nozzle presented in that first paper.

SOR Members as Major Structural Components. Often

important structural components have axisymmetric geometry
such as pipes, but the entire structure is not SOR. Two
examples taken from the field of petroleum engineering are
shown in Figures 2.54 and 2.55. These are two recent designs
of oil-drilling platforms intended for water depths of 300 to
400m. As can be observed the main structural members are
axisymmetric (reinforced concrete cylindrical shells). This kind
of structure is often analyzed by global-local techniques. In the
global analysis such members are treated with beam or
simplified shell models. Forces computed from the global
analysis are then applied to individual members for a more
detailed 3D analysis that may take advantage of axisymmetry.

Figure 3.2.4. The Draugen oil-

drilling platform (artist’s sketch).
The first monotower concrete
platform built by Norwegian
Contractors. The concrete
structure is 295m high. First
deployed in 1993. The seven
cells at the bottom of the sea form
a reservoir system that can store
up to 1.4 M barrels of oil.

Figure 3.2.5. The Troll oil-drilling

platform (artist sketch). The
tallest concrete platform built to
date. It is 386 m tall and has
220,000 m3 of concrete. The
foundation consists of 36 m tall
concrete skirts that penetrate
into the soft seabed.

Quasi-axisymmetric structures. There is an important class of

structures that may be termed “quasiaxisymmetric,” for which
the axisymmetric geometry is locally perturbed by non-
axisymmetric features. Typical of such features are access
openings, foundations and nonstructural attachments.
Important examples are cooling towers, container vehicles and

jet engines. Such structures may benefit from a global-local
analysis if the axisymmetric characteristics dominate. In this
case the global analysis is axisymmetric but the local analyses
are not.

3.2.1.2. THE PROBLEM VARIABLES

3.2.1.2.1. Global Coordinate System
To simplify the governing equations of the axisymmetric
problem it is natural to use a global cylindrical coordinate
system (r, z, θ) where
• r the radial coordinate: distance from the axis of revolution;
always r ≥ 0.
• z the axial coordinate: directed along the axis of revolution.
• θ the circumferential coordinate, also called the longitude.
The global coordinate system is sketched in Figure 3.2.6.

Figure 3.2.6. Cylindrical coordinate system (r ,

z, θ) for axisymmetric structural analysis.

Remark 3.2.1.2
Note that (r , z) form a Cartesian coordinate system on the θ =
const planes, whereas (r , θ) form a polar coordinate system
on the z = const planes.
3.2.1.2.2. Displacement, Strains, Stresses
The displacement field is a function of r and z only, defined by
two components:
 u r (r, z) 
u(r, z)    (3.2.1)
u
 z (r, z) 

ur is called the radial displacement and uz is the axial

displacement. The circumferential displacement component,
u , is zero on account of rotational symmetry.
The infinitesimal strain tensor in cylindrical coordinates is
represented by the symmetric matrix:
 e rr e rz e r 
[e]   e zr e zz e z  (3.2.2)
er ez e 
Because of the assumed axisymmetric state, er and ez
vanish, leaving only four distinct components:

 e rr e rz 0
[e]  e zr e zz 0  (3.2.3)
 0 0 e 
Each of these non-vanishing components is a function of r and
z only. As usual in preparation for finite element work, the non-
vanishing components are arranged as a 4 × 1 strain vector:
 e rr 
e 
e   zz  (3.2.4)
e 
 
  rz 

in which rz = erz + ezr = 2erz. This differs from the plane stress
case considered in the introductory course in the appearance
of e , the “hoop” or circumferential strain.
The stress tensor in cylindrical coordinates is represented by

the symmetric matrix
  rr rz r 
[]   zr zz  z  (3.2.5)
r z  

Figure 3.2.7. Axisymmetric strains and stresses.

Again because of axisymmetry the components σr and σz

vanish, leaving four nontrivial components:
 rr rz 0 
[]  zr zz 0  (3.2.6)
 0 0  
Each of the non-vanishing components is a function of r and z.
Collecting these four components into a stress vector:

 rr 
 
   zz  (3.2.7)
 
 
 rz 
where rz ≡ σrz. The difference with respect to the plane stress
problem is again the appearance of the “hoop” or
circumferential stress σ .
The stresses and strains over an infinitesimal volume are

depicted in Figure 3.2.7.

3.2.1.3. GOVERNING EQUATIONS

The elasticity equations for the axisymmetric problem are the
field equations: strain-displacement, stress-strain, and stress
equilibrium equations, complemented by displacement and
stress boundary conditions.
3.2.1.3.1. Kinematic Equations
The strain-displacement equations for the axisymmetric
problem are:
u r u z ur
e rr  ; e zz  ; e  ;
r z r
(3.2.8)
u r u z
 rz    e rz  e zr  2e rz
z r
In matrix form:
 e rr   e rr    r 0 
e   e   0    u
 zz   zz   z   r
e      Du (3.2.9)
e   e   1 r 0  u z 
     
  rz   2e rz   z  r 

where D is the 4×2 strain-displacement (symmetric-gradient)

operator. The noteworthy difference with respect to the plane
stress case is the appearance of the hoop strain e = ur /r.
Thus a uniform radial displacement is no longer a rigid body
motion, but produces a circumferential strain. The physical
reason behind this phenomenon is illustrated in Figure 2.58.
The length of the original circumference is 2πr, which grows to
2π(r + ur), inducing a strain 2πur /2πr = ur /r.

3.2.1.3.2. Constitutive Equations

For a linear hyperelastic material, and ignoring thermal and
prestress effects, the most general constitutive equation
consistent with axisymmetry takes the form:
 rr   E11 E12 E13 E14   e rr 

   E E 22 E 23 E 24   e zz 
  zz 
  21
 Ee (3.2.10)
   E 31 E 31 E 33 0  e 
    
 rz   E 41 E 42 0 E 44    rz 

Figure 3.2.8. Showing how a uniform radial

displacement ur induces a circumferential strain ur /r .

To retain axisymmetry, the cross-coupling between the shear

strain and hoop stress must vanish. Consequently E34 = E43 =
0.
For an isotropic material of elastic modulus E and Poisson’s
ratio ν,
1     0 
  1   0 
E   (3.2.11)
E
(1  ) 1  2     1  0 
 
 0 0 0 2 1  2  
1

Remark 3.2.1.3
The coefficients of E go to infinity if  → ½, which

characterizes an incompressible material. This behavior is a
consequence of the “confinement” effect in solids and appears
also in general 3D analysis. On the other hand, the plane
stress constitutive matrix remains finite for  = ½, a behavior
that is characteristic of thin bodies such as plates and shells.
Physically, the small transverse dimension of bodies in plane
stress (plates) allows the material to freely expand or contract
in the z direction.

3.2.1.3.3. Equilibrium Equations
The general (three dimensional) differential equations of

equilibrium in cylindrical coordinates are
(3.2.12)

where br , bz , b are the components of the body force field in

the r , z and θ directions, respectively. For the axisymmetric
problem these equations reduce to
(3.2.13)
The third equation in (2.213) is identically satified if b = 0,

because r = z = 0 and σ is independent of θ. If b  0 the
problem cannot be treated as axisymmetric.

3.2.1.3.4. Boundary Conditions

As usual boundary conditions can be of displacement (PBC)
or of stress or traction (FBC) type. They are specified on
portions Su and St of the boundary, respectively. The reduction
of the stress BCs to two dimension is further discussed in
3.2.1.4.2. and 3.2.4.3.

3.2.1.4. VARIATIONAL FORMULATION

The variational form of the axisymmetric problem is illustrated
with the widely used Total Potential Energy (TPE) form. The
delicate part of the formulation is the dimensionality reduction
step.
3.2.1.4.1. The TPE Functional
The Total Potential Energy (TPE) functional contains only
displacements as master field:
[u] = U[u] − W[u] (3.2.14)
Here the strain energy functional is

(3.2.15)
In (3.2.15) the strains are a slave field which are derived from
displacements.
The external work potential is the sum of contributions due to
body force and prescribed surface tractions:

(3.2.16)
Here b is the body force vector and t̂ the vector of surface

tractions.

3.2.1.4.2. Dimensionality Reduction

The element of volume dV that appears in U and Wb can be
expressed as the “ring element”
dV  2rdA (3.2.17)
where dA is the element of area in the generating cross
section. Insertion in (3.2.15) and the second of (3.2.16)
reduces U and Wb to area integrals:
1
U  2  reT EedA (3.2.18)
2 A

W  2  rbT udA (3.2.19)

A
Notice the appearance of r in the integrand.
Similarly, the element of surface dS in Wt can be expressed as
dS  2rds (3.2.20)
where ds is an arc length element. Inserting in the last of

(3.2.16) reduces Wt to a one-dimensional (line) integral
Wt  2  rt uds
T
(3.2.21)
St

The common factor 2π in these integrals is (usually)

suppressed in the finite element implementation. This should
not cause difficulties except for the case of a concentrated
load, as discussed in the following subsection.
We summarize the outcome of this dimensionality reduction by

saying that the original three dimensional problem has been
reduced to a two-dimensional one.

3.2.1.4.3. Line and Point Forces

Body forces (e.g. gravity or centrifugal forces) and distributed
surface forces (e.g. pressure) are handled like in plane
elasticity case explained in LFEM, but concentrated loads
require more careful treatment. There are two possibilities: a
line load and an actual concentrated load.
A line load is actually a “ring” load (see Figure 3.2.2) acting on
a circle described by a point of the generating cross section. If
the global components of this load are Fr and Fz, the
appropriate energy contribution to the loads potential W is
WF  2r(Fr u r  Fz u z ) (3.2.22)

where (ur , uz) are the displacements of the “ring” point. Thus
the ubiquitous 2π term can be suppressed
A concentrated or point load F, however, can only act along
the z direction at points on the axis of revolution as illustrated
in Figure 3.2.2. The corresponding work term is
WF  Fu z (3.2.23)
so the factor 2π is missing. To render this compatible with the
other energy terms the load is divided by 2π, so the
contribution to the external loads potential is
 F 
WF  2   u z (3.2.24)
 2 

This device can be visualized by regarding F as the limit of a

z-directed ring load Fz as r → 0.
Remark 3.2.1.4
What the last equation means in practice is that if a
concentrated force of, say, 1000 lb acts on the z axis, it has to
be divided by 2π (that is, 1000/2π) before giving it to a SOR
finite element program if the factor of 2π has been
suppressed. (It is important to read the users’ manual to see if
that is the case).
3.2.1.4.4. Other Variational Forms
The Hellinger-Reissner (HR) functional and the equilibrium-
stress hybrid functionals are derived in the Exercises.

3.2.1.5. TREATING PLANE STRAIN AS A LIMIT CASE

The problem of plane strain may be viewed as the limit of the
axisymmetric case in which the axis of revolution is moved to
infinity so that r →∞, and a “slice” of unit thickness is taken.
Thus a finite element program that handles the axisymmetric

problem may be used to solve problems of plane strain with
acceptable approximation.

Homework Exercises for 3.2.1

Structures of Revolution
EXERCISE 3.2.1.1
Derive the HR functional for the axisymmetric problem. Is
there any difference in the treatment of loads with respect to
the TPE form?
EXERCISE 3.2.2.2
Derive the equilibrium-stress hybrid functional for the
axisymmetric problem. Is there any difference in the
treatment of loads with respect to the TPE form?

3.2.2. Axisymmetric Iso-P Elements

3.2.2.1. Introduction
In this section we consider the finite element discretization of
structures of revolution (SOR). The focus in this Chapter will
be on isoparametric elements, so as to tie with the
introductory course.
The dimensionality reduction process described in section
3.2.1 “folds” the problem into integrals taken over the
generating cross section and its boundaries. The finite
element discretization can be therefore confined to the r–z
plane and the circumferential (θ) dimension conceptually
disappears.

The resulting finite elements are called axisymmetric solid

elements, SOR elements, or “ring” elements in the literature.
They are defined completely by the geometry of their cross
section in the
(r, z) plane, as illustrated in Figure 3.2.9. Because this cross
section is plane, the element geometry definition is two-
dimensional. It follows that the two-dimensional element
configurations studied in IFEM: 3 and 6-node triangles, 4-, 8-,
and 9-node quadrilaterals, etc, can be used.
The key difference with respect to the plane stress case is
the appearance of the “hoop” strain and stress, which
together contribute a term.

1
  (3.2.25)
2
to the strain energy density. This in turn introduces some
additional terms in the structure of the strain-displacement
matrix B, which translate into additional contributions in the
formation of the element stiffness matrix and consistent node
force vector.

3.2.2.2. Isoparametric definition

Axisymmetric finite elements are developed in this section
using the displacement-based isoparametric formulation.
That is, the element cross section geometry and
displacement field are interpolated by the same shape
functions. For an isoparametric element with n nodes:
(3.2.26)

Here ur(r, z) and uz(r, z) are finite element approximations to

the actual displacement field components, while Ni(e) are the
usual element shape functions in element natural coordinates
defined in the Introduction to Finite Element Methods (IFEM)
course. The same continuity and completeness requirements
apply with minor differences (see remarks below).
Remark 3.2.2.1
Displacement based elements are based on the TPE
functional. The variational index for the only master field:
displacements, is one. Consequently the interelement
continuity required is C0. That is, the displacement
components ur and uz must be continuous between adjacent
elements. Element boundaries lying on the axis of
revolution are special: at such points the radial displacement

ur must vanish if the structure is continuous there (that is, a
tiny hole is precluded) although there is no need to make that
condition explicit in the element formulation.
Remark 3.2.2.2
The completeness criterion demands that all rigid body modes

and strain states be exactly represented. This is met if the
element shape functions can represent any displacement field
that is linear in r and z. (In fact, a more detailed analysis
shows that this is a slight overkill, but will do for now).

Generating cross-section
Figure 3.2.9. FEM

discretization of a SOR with
axisymmetric “ring” elements.
Generatic “ring” finite element

3.2.2.3. The Element Stiffness Matrix
3.2.2.3.1. The Strain-Displacement Matrix
The strain-displacement matrix that appears in the

computation of the element stiffness matrix is given by
B  DN (3.2.27)
where D is the strain-displacement gradient matrix introduced

in the previous Chapter,

(3.2.28)
The strain-displacement matrix that appears in the

computation of the element stiffness matrix is given by
(3.2.29)

Here we have assumed that the same shape functions are

used in both r and z directions.
Comparing these expressions with those in IFEM, we see

that N is the same, and D has an extra row. Multiplying
(3.2.27) through we obtain
(3.2.30)
where the n-component row vectors qr , qz , q are defined

by

(3.2.31)
3.2.2.3.2. The Stiffness Matrix Integrand
Let the constitutive (stress-strain) matrix be

(3.2.32)
in which E34 = E43 = 0 to maintain axisymmetry conditions.

First we form the product
(3.2.33)

Premultiplying by BT gives
(3.2.34)
in which
(3.2.35)

3.2.2.3.3. Numerical Integration
We shall assume here that the entries of the element

stiffness matrix are computed by the standard numerical
integration techniques covered in IFEM. For example, the
stiffness matrix of a quadrilateral isoparametric element
integrated by a p-point Gauss rule along each isoparametric
coordinate is evaluated as
(3.2.36)
in which as usual ξk and ηℓ denote abcissae of Gauss points

whereas wk and wℓ denote integration weights.

Notice the appearance of the radius r at the Gauss point; this

quantity also appears inside B in the form 1/r. This radius is
obtained by the isoparametric interpolation, namely, the first
row of (3.2.30):
n
r(k ,  )   ri N i(e) (k ,  ) (3.2.37)
i 1
Other quantities such as the Jacobian determinant J, are
calculated as explained in IFEM. In particular the shape
function subroutines can be reused, as explained in more
detail in the following Chapter.
The stiffness of triangular elements can also be evaluated by
numerical quadrature but using integration rules appropriate
to the triangular geometry.

3.2.2.4. Consistent Node Forces for Body Loads

Body forces (also called volume forces) arise frequently in
SOR analysis. The most important loads of this type are
1. Gravity (own weight). This effect is important in massive SOR,
as encountered in civil, geophysical and nuclear applications.
2. Centrifugal forces in rotating structures. These are important in
aerospace and mechanical structures (for examble, high-
speed rotating machinery such as turbines).
3. Thermal, shrinkage and pre-stress effects. These may be
important depending on fabrication techniques, material, and
the environment to which the structure will be exposed. In the
present course thermal effects are covered briefly.

3.2.2.4.1. Body Forces

An applied body force field b is defined by two components in
the r and z directions, each of which is a function of those two
directions only:
 b r (r, z) 
b(r, z)    (3.2.38)
 b z (r, z) 
The consistent node forces f are given by
f (e)   N T brdA (3.2.39)

A

Where N is the matrix of element shape functions given in

(3.2.29). This can be conveniently processed by numerical
integration. Considering again the quadrilateral element with
a p-point Gauss rule in each direction,
(3.2.40)
3.2.2.4.2. Thermal Effects

Suppose the temperature of the structure changes
axisymmetrically by T (r, z) from a reference temperature.
The constitutive equations become

  E(e  T) (3.2.41)

where  is an array of dilatation coefficients that provide
thermal strains in a mechanically unconstrained body. (A
linear relation between strains and temperature changes is
assumed.) For isotropic materials
 T      0 (3.2.42)
where  is the usual coefficient of dilatation of the material.

On inserting (3.2.42) into the potential energy formulation for
a generic element yields element stiffness equations that
account for thermal effects:

K (e) u (e)  f (e)  f M(e)  f T(e) (3.2.43)

where fM are the mechanical forces considered so far, and fT,
called the thermal forces or equivalent thermal loads, account
for contribution from the temperature changes:
f T(e)   BT TrdA (3.2.44)
A
This term can be evaluated by numerical integration.
The sum of mechanical and thermal node forces is
sometimes called the effective node force vector or simply
effective forces in the FEM literature.

Chapter 3 - English

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

Chapter 3 - English

Uploaded by

Copyright:

Available Formats

FINITE ELEMENT METHOD

Chapter 3. FEM FOR 2D AND AXISYMMETRIC SOLIDS

CHAPTER 3. FEM FOR 2D SOLIDS

Assoc. Prof. Dr. Tich Thien TRUONG 1 HCMUT 2007

 LINEAR QUADRILATERAL ELEMENTS

Assoc. Prof. Dr. Tich Thien TRUONG 3 HCMUT 2007

Axisymmetric Iso-P Elements

Assoc. Prof. Dr. Tich Thien TRUONG 4 HCMUT 2007

Assoc. Prof. Dr. Tich Thien TRUONG 6 HCMUT 2007

2D solids – plane stress and plane strain

Plane stress Plane strain

Assoc. Prof. Dr. Tich Thien TRUONG 7 HCMUT 2007

LINEAR TRIANGULAR ELEMENTS

Assoc. Prof. Dr. Tich Thien TRUONG 8 HCMUT 2007

Field variable interpolation U h ( x, y )  N ( x, y )d e

Assoc. Prof. Dr. Tich Thien TRUONG 9 HCMUT 2007

Shape functions construction

Assoc. Prof. Dr. Tich Thien TRUONG 10 HCMUT 2007

Shape functions construction

Delta function property:

Assoc. Prof. Dr. Tich Thien TRUONG 11 HCMUT 2007

Shape functions construction

Area of triangle Moment matrix

Assoc. Prof. Dr. Tich Thien TRUONG 12 HCMUT 2007

Shape functions construction

Assoc. Prof. Dr. Tich Thien TRUONG 13 HCMUT 2007

Shape functions construction

Assoc. Prof. Dr. Tich Thien TRUONG 14 HCMUT 2007

Using area coordinates

 Alternative method of constructing shape functions

Assoc. Prof. Dr. Tich Thien TRUONG 15 HCMUT 2007

Using area coordinates

Delta function property: e.g. L1 = 0 at if P at nodes 2 or 3

Assoc. Prof. Dr. Tich Thien TRUONG 16 HCMUT 2007

Assoc. Prof. Dr. Tich Thien TRUONG 19 HCMUT 2007

 k11 k12 k13 k14 k15 k16 

C1h  k33 k34 k35 k36 

Assoc. Prof. Dr. Tich Thien TRUONG HCMUT 2007

k14 = C2xikyjk + yikxjk

Assoc. Prof. Dr. Tich Thien TRUONG HCMUT 2007

Eisenberg and Malvern (1973):

Assoc. Prof. Dr. Tich Thien TRUONG 24 HCMUT 2007

 f sx12 .l12  f sx31.l31 

Assoc. Prof. Dr. Tich Thien TRUONG HCMUT 2007

Load vector of element inferred volume load {g} = {gx gy}T:

Assoc. Prof. Dr. Tich Thien TRUONG HCMUT 2007

Stress vector of element:

Assoc. Prof. Dr. Tich Thien TRUONG HCMUT

 y jk C2 x jk  yik C2 xik yij C2 xij 

Assoc. Prof. Dr. Tich Thien TRUONG HCMUT

Assoc. Prof. Dr. Tich Thien TRUONG 30 HCMUT 2007

Shape functions construction

Assoc. Prof. Dr. Tich Thien TRUONG 31 HCMUT 2007

Shape functions construction

x  ( x1  a) y  ( y1  b) ,  and  are the natural coordinates

Assoc. Prof. Dr. Tich Thien TRUONG 32 HCMUT 2007

Shape functions construction

N 3 at node 1  14 (1   )(1   )  1  0

N 3 at node 4  14 (1   )(1   )  1  0 2

Assoc. Prof. Dr. Tich Thien TRUONG 33 HCMUT 2007

Assoc. Prof. Dr. Tich Thien TRUONG HCMUT 2007

The chain rule yields:

The determinant in the denominator is the determinant of the