CHAP 2 WEIGHTED RESIDUAL METHODS

FOR 1D PROBLEMS

1
 INTRODUCTION
• Direct stiffness method is limited for simple 1D problems
• FEM can be applied to many engineering problems that are
governed by a differential equation
• Need systematic approaches to generate FE equations
– Weighted residual method
– Energy method
• Ordinary differential equation (second-order or fourth-order)
can be solved using the weighted residual method, in
particular using Galerkin method
• Principle of minimum potential energy can be used to derive
finite element equations

2
2.1 EXACT VS. APPROXIMATE
SOLUTION

3
 EXACT VS. APPROXIMATE SOLUTION
• Exact solution
– Boundary value problem: differential equation + boundary conditions
– Displacements in a uniaxial bar subject to a distributed force p(x)
d2u
2
+ p( x ) = 0, 0 £ x £ 1
dx
u(0) = 0 üï
ïï
du ý Boundary conditions
(1) = 1ïï
dx ïþ
– Essential BC: The solution value at a point is prescribed (displacement
or kinematic BC)
– Natural BC: The derivative is given at a point (stress BC)
– Exact solution u(x): twice differential function
– In general, it is difficult to find the exact solution when the domain
and/or boundary conditions are complicated
– Sometimes the solution may not exists even if the problem is well
defined
4
 EXACT VS. APPROXIMATE SOLUTION cont.

• Approximate solution u(x)
– It satisfies the essential BC, but not natural BC
– The approximate solution may not satisfy the DE exactly
– Residual: d2u
2
+ p( x ) = R ( x )
dx
– Want to minimize the residual by multiplying with a weight W and
integrate over the domain
1
ò0 R( x )W ( x )dx = 0 Weight function

– If it satisfies for any W(x), then R(x) will approaches zero, and the
approximate solution will approach the exact solution
– Depending on the choice of W(x): least square error method,
collocation method, Petrov-Galerkin method, and Galerkin method

5
 1.4

1.2

1
u(x)

0.8

0.6

0.4
Exact solution
0.2 Approx. solution with W = 1
Approx. solution with W = x
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

6
 0.4 0.3

0.2 0.2

R(x)W(x)
R(x)W(x)

0 0.1

-0.2 0

-0.4 -0.1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x
(a) W = 1 (b) W = x

7
2.2 GALERKIN METHOD

8
 GALERKIN METHOD
• Approximate solution is a linear combination of trial functions
N
u( x ) = å ci fi ( x ) Trial function
i =1
– Accuracy depends on the choice of trial functions
– The approximate solution must satisfy the essential BC
• Galerkin method
– Use N trial functions for weight functions
1
ò0 R( x )fi ( x )dx = 0, i = 1,, N

1æ d2u
 ö÷
ò0 ççèç dx 2 + p( x ) ÷÷øfi ( x )dx = 0, i = 1,,N
1 d 2u
 1
ò0 dx 2 fi ( x )dx = -ò0 p( x )fi ( x )dx, i = 1,, N

9
 GALERKIN METHOD cont.
• Galerkin method cont.
– Integration-by-parts: reduce the order of differentiation in u(x)
1
du  dfi
1 du 1
fi -ò dx = -ò p( x )fi ( x )dx, i = 1,,N
dx 0 0 dx dx 0

– Apply natural BC and rearrange

1 df du 1 du du
ò0 dx = ò0 p( x )fi ( x )dx + (1)fi (1) - (0)fi (0), i = 1,, N
i
dx dx dx dx

– Same order of differentiation for both trial function and approx. solution
– Substitute the approximate solution

1 df
N
df j 1 du du
ò0
i
å cj
dx j =1 dx
dx = ò0 p( x )fi ( x )dx +
dx
(1)fi (1) -
dx
(0)fi (0), i = 1,, N

10
 GALERKIN METHOD cont.
• Galerkin method cont.
– Write in matrix form
N

å Kij c j = Fi , i = 1,, N [K] {c} = {F}

( N´N )( N´1) ( N´1)
j =1

1 d fi d f j
K ij = ò0 dx dx
dx

1 du du
Fi = ò0 p( x )fi ( x )dx +
dx
(1)fi (1) -
dx
(0)fi (0)

– Coefficient matrix is symmetric; Kij = Kji

– N equations with N unknown coefficients

11
 EXAMPLE1
• Differential equation Trial functions
d 2u f1( x ) = x f1¢( x ) = 1
2
+ 1 = 0,0 £ x £ 1
dx f2 ( x ) = x 2 f2¢ ( x ) = 2 x
u(0) = 0 üï
ï
ï
du ý Boundary conditions
(1) = 1ï
ï
dx ï
þ
• Approximate solution (satisfies the essential BC)
2
u( x ) = å ci fi ( x ) = c1x + c2 x 2
i =1

• Coefficient matrix and RHS vector

1 2 1 du 3
K11 = ò
0
( f1¢ ) dx = 1 F1 = ò0 f1( x )dx + f1(1) -
dx
(0)f1(0) =
2
1 1 du 4
K12 = K21 = ò ( f1¢f2¢ ) dx = 1 F2 = ò f2 ( x )dx + f2 (1) - (0)f2 (0) =
0 0 dx 3
1 2 4
K22 = ò ( f2¢ ) dx =
0 3
12
 EXAMPLE1 cont.
• Matrix equation
1ì
ï9ü ìïï 2 üïï
1 é3 3ù ï
{F} = í ý
-1
{c } = [K ] {F } = í 1 ý
[K ] = ê ú ïîï - 2 ïþï
3 êë 3 4 úû 6ï
ï8ï
î ï
þ

• Approximate solution
x2
u( x ) = 2 x -
2
– Approximate solution is also the exact solution because the linear
combination of the trial functions can represent the exact solution

13
 EXAMPLE2
• Differential equation Trial functions
d 2u f1( x ) = x f1¢( x ) = 1
2
+ x = 0, 0 £ x £1
dx f2 ( x ) = x 2 f2¢ ( x ) = 2 x
u(0) = 0 üï
ïï
du ý Boundary conditions
(1) = 1ïï
dx ïþ
1 ìï16 üï
• Coefficient matrix is same, force vector: {F} = í ý
12 ïîï15 ïþï
ì
ï 19 ü
12 ï 19 x2
{c } = [K ] {F } = í 1 ïý
-1 ï u( x ) = x-
ï
ï - 4 ïþï
î 12 4
• Exact solution
3 x3
u( x ) = x -
2 6

– The trial functions cannot express the exact solution; thus,

approximate solution is different from the exact one

14
 EXAMPLE2 cont.
• Approximation is good for u(x), but not good for du/dx

1.6

1.4

1.2

1.0
u(x), du/dx

0.8

0.6

0.4

0.2

0.0
0 0.2 0.4 x 0.6 0.8 1

u-exact u-approx. du/dx (exact) du/dx (approx.)

15
2.3 HIGHER-ORDER
EQUATIONS

16
 HIGHER-ORDER DIFFERENTIAL EQUATIONS
w (0) = 0 üï
• Fourth-order differential equation ïï
ý Essential BC
dw
(0) = 0 ïï
d 4w dx ïþ
- p( x ) = 0, 0£x£L
dx 4 d 2w üï
( L ) = M ïï
– Beam bending under pressure load dx 2 ïï
ý Natural BC
3
d w ïï
• Approximate solution 3
(L ) = -V ïï
dx ïþ
N
w ( x ) = å ci fi ( x )
i =1

• Weighted residual equation (Galerkin method)

L æ d 4w
 ö÷
çç
ò0 çè dx 4 - p( x ) ø÷÷fi ( x )dx = 0, i = 1,,N
– In order to make the order of differentiation same, integration-by-parts
must be done twice

17
 HIGHER-ORDER DE cont.
• After integration-by-parts twice
3 L 2 L
d w d w d fi L d 2w
 d 2fi L
f
3 i
- 2 +ò 2 2
dx = ò0 p( x )fi ( x )dx, i = 1,, N
dx 0 dx dx 0
0 dx dx
L L
L d 2w

d fi 2 L d w 3 2
d w d fi
ò0 2
dx dx 2
dx = ò0 p( x )fi ( x )dx - 3 fi
dx
+
dx 2 dx
, i = 1,, N
0 0

• Substitute approximate solution

L L
L
N
d 2f j d 2fi L d 3w d 2w d fi
ò0 å c j dx 2 dx 2 dx = ò0 p( x )fi ( x )dx - 3 fi
dx
+ 2
dx dx
, i = 1,, N
j =1 0 0

– Do not substitute the approx. solution in the boundary terms

• Matrix form L d 2f d 2fj
Kij = ò 2
i
2
dx
0 dx dx
[K] {c} = {F}
N´N N´1 N´1 L L
3 2
L d w d w dfi
Fi = ò p( x )fi ( x )dx - f
3 i
+
0 dx 0 dx 2 dx 0
18
 EXMAPLE
• Fourth-order DE dw
w (0) = 0 (0) = 0
dx
d4w d2w d3w
4
- 1 = 0, 0 £ x £1 (1) = 2 (1) = -1
dx dx 2
dx 3

• Two trial functions

f1 = x 2, f2 = x 3 f1¢¢ = 2, f2¢¢ = 6x

• Coefficient matrix
1 2
K11 = ò0 ( f1¢¢) dx = 4
1 é4 6 ù
K12 = K 21 = ò0 ( f1¢¢f2¢¢ ) dx = 6 [K ] = ê ú
êë 6 12 úû
1 2
K 22 = ò0 ( f2¢¢ ) dx = 12

19
 EXAMPLE cont.
• RHS
1 d3w (0) d2w (0) 16
F1 = ò0 2
x dx + V f1(1) +
dx 3
f1(0) + M f1¢(1) -
dx 2
f1¢(0) =
3
1 d3w (0) d2w (0) 29
F2 = ò0 3
x dx + V f2 (1) +
dx 3
f2 (0) + M f2¢ (1) -
dx 2
f2¢ (0) =
4

• Approximate solution
ì
ï 41 ü
ï 41 2 1 3
{c} = [K]-1 {F} = ïí 241 ï ý w ( x ) = x - x
ïîï - 4 ï
ï
þ 24 4

• Exact solution
1 4 1 3 7 2
w(x) = x - x + x
24 3 4

20
 EXAMPLE cont.

1
w'', w''

-1

-2

-3
0 0.2 0.4 x 0.6 0.8 1
w'' (exact) w'' (approx.) w''' (exact) w''' (approx.)

21
2.4 FINITE ELEMENT
APPROXIMATION

22
 FINITE ELEMENT APPROXIMATION
• Domain Discretization
– Weighted residual method is still difficult to obtain the trial functions
that satisfy the essential BC
– FEM is to divide the entire domain into a set of simple sub-domains
(finite element) and share nodes with adjacent elements

F
F

Structure

ìï ¶s
Element
ïï xx + ¶txy + b = 0
ïï ¶x
x
¶y
í
ïï ¶txy ¶syy
ïï + + by = 0
ïî ¶x ¶y
23
 FINITE ELEMENT APPROXIMATION
• Domain Discretization
– Within a finite element, the solution is approximated in a simple
polynomial form
u(x) Nodes
Approximate
solution
x

Finite Exact
elements solution
– When more number of finite elements are used, the approximated
piecewise linear solution may converge to the analytical solution

u
Exact solution
Two elements
Four elements
Eight elements

x
24
 FINITE ELEMENT METHOD cont.
• Types of finite elements

1D 2D 3D

• Variational equation is imposed on each element.

1 0.1 0.2 1
ò0 dx = ò0 dx + ò
0.1
dx +  + ò
0.9
dx

One element

25
 Galerkin vs. Finite Element Method

Finite element method Galerkin method

N
u( x ) = Ni ( x )ui + Ni +1( x )ui +1 u( x ) = å ci fi ( x )
i =1

Coefficients are nodal value Coefficients don’t have physical

meaning
Defined only within an element Defined in entire domain
Easy to satisfy BC Difficult to find weight function to satisfy
BC
Equilibrium in element by element Equilibrium in entire domain
Need assembly No need to assembly

26
 TRIAL SOLUTION
– Solution within an element is approximated using simple polynomials.

– i-th element is composed of two nodes: xi and xi+1. Since two

unknowns are involved, linear polynomial can be used:

u( x ) = a0 + a1x, xi £ x £ xi +1

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

nodal solutions u(xi) and u(xi+1).
27
 TRIAL SOLUTION cont.
– Substitute two nodal values

ì
ï
ï u( xi ) = ui = a0 + a1xi
í
ï
î u( xi +1) = ui +1 = a0 + a1xi +1
ï
– Express a0 and a1 in terms of ui and ui+1. Then, the solution is
approximated by

xi +1 - x x - xi
u( x ) = (i )
ui + ( i ) ui +1
L 
 L 

Ni ( x ) Ni +1( x )

– Solution for i-th element:

u( x ) = Ni ( x )ui + Ni +1( x )ui +1, xi £ x £ xi +1

– Ni(x) and Ni+1(x): Shape Function or Interpolation Function

28
 TRIAL SOLUTION cont.
• Observations
– Solution u(x) is interpolated using its nodal values ui and ui+1.
– Ni(x) = 1 at node xi, and =0 at node xi+1.
Ni(x) Ni+1(x)

xi xi+1
– The solution is approximated by piecewise linear polynomial and its
gradient is constant within an element.
u ui+2 du
ui dx
ui+1

xi xi+1 xi+2 xi xi+1 xi+2

– Stress and strain (derivative) are often averaged at the node.

29
 GALERKIN METHOD
• Relation between interpolation functions and trial functions
– 1D problem with linear interpolation
ì
ï 0, 0 £ x £ xi -1
ï
ï
ï
ï ( i -1) x - xi -1
ï N ( x ) = , xi -1 < x £ xi
ND ï i ( i -1)
fi ( x ) = ï
L
u( x ) = å ui fi ( x ) í
ï xi +1 - x
ï
ï
(i )
Ni ( x ) = , xi < x £ xi +1
i =1 ï (i )
ï L
ï
ï
î 0,
ï xi +1 < x £ xND
– Difference: the interpolation function does not exist in the entire
domain, but it exists only in elements connected to the node
• Derivative
ì
ï 0, 0 £ x £ xi -1 1
ï
ï
ï 1 fi (x )
ï 1/ L( i -1 )
ï , xi -1 < x £ xi
d fi ( x ) ï ( i - 1)
=ï
L
í x i -2 x i -1 xi x i +1
dx ï
ï 1
ï - , xi < x £ xi +1 -1/ L( i )
ï
ï L ( i ) d fi (x )
ï
ï dx
î 0,
ï xi +1 < x £ xND
30
 EXAMPLE
• Solve using two equal-length elements
2 u(0) = 0 üï
du ïï
+ 1 = 0, 0 £ x £ 1 du ý Boundary conditions
dx 2
(1) = 1ïï
dx ïþ

• Three nodes at x = 0, 0.5, 1.0; displ at nodes = u1, u2, u3

• Approximate solution u( x ) = u1f1( x ) + u2f2 ( x ) + u3f3 ( x )
ìï1 - 2 x, 0 £ x £ 0.5 ìï 2 x, 0 £ x £ 0.5
f1( x ) = í f2 ( x ) = í
ïïî 0, 0.5 < x £ 1 ïïî 2 - 2 x, 0.5 < x £ 1
ìï 0, 0 £ x £ 0.5
f3 ( x ) = í 1

ïïî -1 + 2 x, 0.5 < x £ 1


 0.5 


0
0 0.5 1
x

31
 EXAMPLE cont.
• Derivatives of interpolation functions
df1( x ) ìï -2, 0 £ x £ 0.5 df2 ( x ) ìï 2, 0 £ x £ 0.5
=í =í
dx ï
î 0, 0.5 < x £ 1
ï dx ïïî -2, 0.5 < x £ 1
df3 ( x ) ìï 0, 0 £ x £ 0.5
=í
dx ïïî 2, 0.5 < x £ 1

• Coefficient matrix
df1 df2
1 0.5 1
K12 = ò dx = ò (-2)(2)dx + ò (0)(-2)dx = -2
0 dx dx 0 0.5
1 d f df 0.5 1
K 22 = ò 2 2
dx = ò 4dx + ò 4dx = 4
0 dx dx 0 0.5
• RHS
0.5 1 du du du
F1 = ò 1´ (1 - 2 x )dx + ò 1´ (0)dx + (1)f1(1) - (0)f1(0) = 0.25 - (0)
0 0.5 dx dx dx
0.5 1 du du
F2 = ò0 2 xd x + ò
0.5
(2 - 2 x )dx +
dx
(1)f2 (1) -
dx
(0)f2 (0) = 0.5
32
 EXAMPLE cont.
• Matrix equation
é 2 -2 0 ù ïì u1 ïü ïì F1 ïü Consider it as unknown
ê úï ï ï ï
ê -2 4 -2 ú í u2 ý = í 0.5 ï
ï ï ï
ý
ê úï ï ï ï
êë 0 -2 2 úû ïî u3 ïþ ïî1.25 ï
ï ï ï ïþ

• Striking the 1st row and striking the 1st column (BC)
é 4 -2 ù ìï u2 üï ìï 0.5 üï
ê úí ý = í ý
êë -2 2 ûú îïï u3 ïþï îïï1.25 þïï
• Solve for u2 = 0.875, u3 = 1.5
• Approximate solution
ìï1.75 x, 0 £ x £ 0.5

u( x ) = í
ïïî 0.25 + 1.25 x, 0.5 £ x £ 1

– Piecewise linear solution

33
 EXAMPLE cont.
• Solution comparison 1.6

• Approx. solution has about 1.2

8% error

u(x)
0.8
• Derivative shows a large
discrepancy 0.4
u-exact

u-approx.
• Approx. derivative is 0
constant as the solution is 0 0.2 0.4 x 0.6 0.8 1

piecewise linear 2

du/dx 1.5

du/dx (exact)
0.5
du/dx (approx.)

0
0 0.2 0.4 x 0.6 0.8 1

34
2.5 FORMAL PROCEDURE

35
 FORMAL PROCEDURE
• Galerkin method is still not general enough for computer code
• Apply Galerkin method to one element (e) at a time
• Introduce a local coordinate
x - xi x - xi
x = xi (1 - x ) + x j x x= = (e )
x j - xi L

• Approximate solution within the element

N 1(x) N 2 (x)
u( x ) = ui N1( x ) + u j N2 ( x )

N1(x ) = (1- x ) Element e

N2 (x ) = x xi xj
x
æ L(e )
x-x
N1( x ) = çç1 - ( e ) i ÷÷ö dN1 dN1 dx
=
1
= - (e )
è L ø dx dx dx L
x - xi dN 2 dN 2 dx 1
N2 ( x ) = = = + (e )
L( e ) dx dx dx L
36
 FORMAL PROCEDURE cont.
• Interpolation property
N1( xi ) = 1, N1( x j ) = 0 u( xi ) = ui
N2 ( xi ) = 0, N2 ( x j ) = 1 u( x j ) = u j

• Derivative of approx. solution

du dN1 dN2
= ui + uj
dx dx dx
du ê dN1 dN2 ú ìï u1 ü ï 1 ê dN1 dN2 ú ìï u1 üï
=ê úí ý = ê
( e ) ê dx
úí ý
dx êë dx dx úû ï ï
ï L ë
ï u2 þ
î dx úû ï
ï u2 ïþï
î

• Apply Galerkin method in the element level

xj dNi du xj du du
òx i dx dx
dx = òx i
p( x )Ni ( x )dx +
dx
( x j )Ni ( x j ) - ( xi )Ni ( xi ), i = 1,2
dx

37
 FORMAL PROCEDURE cont.
• Change variable from x to 
1 1 d Ni ê dN1 dN 2 ú ì
ï u1 ü
ï 1

L( e ) ò0 dx
ê ú dx í ý = L ò p( x )Ni (x )dx
(e )
êë dx dx úû ï
ï u2 ï
î ï
þ 0

du du
+ ( x j )Ni (1) - ( xi )Ni (0), i = 1,2
dx dx
– Do not use approximate solution for boundary terms
• Element-level matrix equation
ìï du üï
ïï - ( xi ) ïï 1 ìï N1(x ) üï
[k( e ) ]{q( e ) } = { f ( e ) } + ïí
ïï du
dx ïý
ïï
{f (e ) (e )
}=L ò0 p( x )í ý dx
ïîï N2 (x ) ïþï
ïï + ( x j ) ïï
î dx þ
é æ dN ö2 dN1 dN2 ùú
ê ç 1 ÷÷
1 ç dx ÷ø
1ê è dx dx ú é 1 -1ù
é k(e ) ù = ê ú dx = 1 ê
ë û L( e ) ò0 ê 2 ú L ë -1 1 úû
(e ) ê
ú
2´2 ê dN2 dN1 æ dN2 ö÷ ú
çç
ê
ë dx dx è dx ø÷÷ ûú
38
 FORMAL PROCEDURE cont.
• Need to derive the element-level equation for all elements
• Consider Elements 1 and 2 (connected at Node 2)
ì
ï du ü
ï
é k11 ù(1)
ì ü ì ü (1)ï
ï - ( x 1 ) ï
ï
k12 ï 1 ï ï 1 ï
u f ï d x ï
ê ú í ý = í ý +í ý
êë k21 ú
k22 û î ï
ï 2þ
u ï
ï îï
ï 2þ
f ï
ï ï d u ïï
ï
ï + ( x2 ) ï
ï
î dx ï
þ
ì
ï du ü
ï
é k11 (2) (2) ï - ( x ) ï
k12 ù ì ü ì ü ï 2 ï
ú íï 2 ýï = íï 2 ýï + íï dx
u f ï
ê ý
êë k21 k22 úû ï ï u3 ï
î ï ï
þ ï f3 ï
î ï
þ ïï du ïï
ï + (x )
ï dx 3 ï
î ï
þ
• Assembly
Vanished
ì
ï du üï
é k (1) (1)
k12 0 úìù ü ïì f (1) ü
ï ï
ï - ( x1 ï
) ï unknown term
ê 11 ï u1ï ï
ï ïï ïï (1)
1 ï
ï ïïï d x ï
ê k (1) (1) (2) (2) ú ï (2) ï ïï
ê 21 k22 + k11 k12 ú í u2 ý = í f2 + f2 ý + í 0 ý
ê ï ï ï ï ï ïï
(2) ú ï ï ï ï ï
ïï ïï du
ê 0
ë
(2)
k21 k22 úû ï
î u3 ï
þ ïïî f3
(2)
þ ï ( x3 ) ïï
ï
ï dx
î þï
39
 FORMAL PROCEDURE cont.
• Assembly of NE elements (ND = NE + 1)
é (1) ù ì
ï du ü
ï
ê k11 (1)
k12 0  ú
0 ìï u üï ï ì f (1) ü
ï ï
ï - ( x1 ) ï
ï
ê úï 1 ï ï ï
1 ï
ï ï d x ï
ê (1) (1) (2) (2) ú ï ï ï (1) (2) ï ïï ï
ï
ê k21 k22 + k11 k12  0 ú ïï u2 ïï ï ï f2 + f2 ï ïï 0 ï
ê ï ï ï
ú ï ï = ï (2) ï ï ïï
(2) (2) (2) (3) ï+ ï 0
ê 0 k221 k22 + k11  0 ú í u3 ý í f3 + f3 ý í ý
ê ï
úï ï ï ï ï ï
ï ï
ï ï
ï
ê      ú ïï  ïï ï ï  ï
ï ï
ï  ï
ï
ê ú ïï ïï ï ï ï
ï ï
ï ï
ï
ê 0 ( NE ) ( NE ) ú ï u ï
k22 ú î N þ ï ( NE )
ï d u
êë 0 0 k21
( ND´1) ï N
î f ï ï + ( xN ) ïï
þ ï
û ( ND´1) ï dx
î ï
þ
( ND´ND ) ( ND´1)

[K s ]{Q s } = {Fs }

• Coefficient matrix [Ks] is singular; it will become non-singular

after applying boundary conditions

40
 EXAMPLE
• Use three equal-length elements
d2u
2
+ x = 0, 0 £ x £1 u(0) = 0, u(1) = 0
dx

• All elements have the same coefficient matrix

é k(e ) ù 1 é 1 -1ù é 3 -3 ù
ë û 2´2 = L( e ) êê -1 1 úú = êê -3 3 úú , (e = 1,2,3)
ë û ë û
• Change variable of p(x) = x to p(): p(x ) = xi (1 - x ) + x j x
• RHS
1 ìï N1(x ) üï 1 ìï1 - x üï
ò0 ý dx = L ò [ xi ( 1 - x ) + x j x ]í
(e ) (e ) (e )
{f }=L p( x ) í ý dx
ï
ï N2 ( x ) ï
î ï
þ 0 ï
ï x ï
î ï
þ
ìï xi x ü
ïï + j ïïï
3 6ï
= L( e ) ïí ý, (e = 1,2,3)
ïï xi x j ïï
ïï + ïï
î6 3þ
41
 EXAMPLE cont.
• RHS cont. ìï f (1) üï ìï f (2) ü ìï f (3) üï
ïí 1 ïý = 1 ì ï 1ü
í
ï
ý, ïí 2 ï ï
ý=
1 ìï 4 üï
í ý, ïí 3 ïý = 1 ìïí 7 üïý
ïï f ïï 54 ïî
(1)
ï 2ï
ï
þ ï (2) ï 54 ïîï 5 þïï ïîï f4 ïïþ 54 îïï 8 þïï
(3)
î2 þ îï f3 þï
• Assembly
ìï 1 du üï
ïï - (0) ïï
ï 54 dx ïï Element 1
é 3 -3 0 0 ù ìï u1 üï ïï 2 4 ïï
ê ú ïï ïï ïï Element 2
ê -3 3 + 3 -3 0 ú ïï u2 ïï ïï 54 + 54 ïïï
ê úí ý = í ý Element 3
ê 0 -3 3+3 ú ï
-3 ï u3 ï ï 7ï ï 5 ïï
ê úï ï ï + ïï
êë 0 0 -3 ú ï ï
3 û îï u4 þï ï ï 54 54 ï
ïï 8 du ïï ï
ïï + (1)
ïî 54 dx ïïþ
• Apply boundary conditions
– Deleting 1st and 4th rows and columns
u2 = 4
é 6 -3 ù ìï u2 üï 1 ìï 1 üï 81
ê úí ý = í ý
ëê -3 6 ûú îïï u3 ïþï 9 îïï 2 þïï u3 = 5
81

42
 EXAMPLE cont.
• Approximate solution
ìï 4 1
ïïï 27 x, 0£x£
3 0.08
ïï u-approx.
u-exact
ïï 4 1æ 1 ö÷ 1 2

u( x ) = í + ç x - ÷÷, ç £x£
0.06

ïï 81 27 è 3ø 3 3

u(x)
ïï 0.04

ïï 5 - 5 æç x - 2 ÷ö, 2
£ x £1
ïïî 81 27 çè 3 ÷÷ø
0.02
3
0

• Exact solution 0 0.2 0.4

x
0.6 0.8 1

1
u( x ) = x (1 - x 2 )
6
– Three element solutions are poor
– Need more elements

43
2.6 ENERGY METHODS

44
 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
• 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
dW = F1 ⋅ dr + F2 ⋅ dr + F3 ⋅ dr + F4 ⋅ dr = (F1 + F2 + F3 + F4 ) ⋅ dr
• If the current position is in force equilibrium, W = 0

F2 F3

r

F1 F4
45
 Principle of Virtual Work
• Powerful alternative method to obtain FE equations
• Principle of virtual work for a particle
– for a particle in equilibrium the virtual work is identically equal to zero
– Virtual work: work done by the (real) external forces through the virtual
displacements
– Virtual displacement: small arbitrary (imaginary, not real) displacement
that is consistent with the kinematic constraints of the particle
• Force equilibrium
å Fx = 0, å Fy = 0, å Fz = 0
• Virtual displacements: u, v, and w
• Virtual work
dW = du å Fx + dv å Fy + dw å Fz = 0
• If the virtual work is zero for arbitrary virtual displacements,
then the particle is in equilibrium under the applied forces
46
 PVW of 1DOF System
• Static equilibrium of a mass-sprint system
– At equilibrium, the spring force, F, is equal to applied load, mg
F (D) - mg = 0
– If the position is perturbed by *
F (D) ⋅ D* = mg ⋅ D*
– External work done by the external forces mg during the application of
a small virtual displacement is equal to the internal work done by the
spring force during the application of that small virtual displacement

47
 PRINCIPLE OF VIRTUAL WORK
• Deformable body (uniaxial bar under body force and tip force)
x
E, A(x)
F
Bx

L
ds
• Equilibrium equation: x
+ Bx = 0 This is force equilibrium
dx
• PVW L
æ ds x ö÷
ò ò çè dx + Bx ÷÷ø du( x )dAdx = 0
ç
0 A

• Integrate over the area, axial force P(x) = A(x)

L
æ dP ö÷
ò ççè dx + bx ÷÷ø du( x )dx = 0
0
48
 PVW cont.
• Integration by parts
L L
L d( du )
P du - ò P dx + ò bx du( x )dx = 0
0 d x
0 0
– At x = 0, u(0) = 0. Thus, u(0) = 0
– the virtual displacement should be consistent with the displacement
constraints of the body
– At x = L, P(L) = F
• Virtual strain de( x ) = d( du )
dx
• PVW:
L L
F du(L ) + ò bx du( x )dx = ò Pde( x )dx
0 0

dWe -dWi

dWe + dWi = 0
49
 PVW cont.
• in equilibrium, the sum of external and internal virtual work is
zero for every virtual displacement field
• 3D PVW has the same form with different expressions
• With distributed forces and concentrated forces
dWe = ò ( t x du + t y dv + tzdw ) dS + åi ( Fxi dui + Fyi dv i + Fzi dw i )
S

• Internal virtual work

dWi = -ò ( s x dex + s y dey + .... + t xy dg xy ) dV
V

50
 Variation of a Function
• Virtual displacements in the previous section can be
considered as a variation of real displacements
• Perturbation of displ u(x) by arbitrary virtual displ u(x)
ut ( x ) = u ( x ) + tdu ( x )
• Variation of displacement
dut ( x )
= du( x ) Displacement variation
dt t =0

• Variation of a function f(u)

df (ut ) df
df = = du
d t t =0 d u

• The order of variation & differentiation can be interchangeable

æ du ö÷ d(du )
dex = d çç ÷ =
è dx ø dx
51
 Variation of a Function
• Displacement Variation
– Need to satisfy homogeneous essential BC
– Both u(x) and u(x) are solution

ut ( x ) = u ( x ) + tdu ( x )

– Let assume that u(1) = 1 (essential BC)

– Then
ut (1) = u (1) + tdu (1) = 1
– Therefore,
du (1) = 0 Homogeneous (i.e., = 0) BC

u(x) can be treated as u(x)

52
 PRINCIPLE OF MINIMUM POTENTIAL ENERGY
• Strain energy density of 1D body
1 1
U0 = s x ex = E ex2
2 2

• Variation in the strain energy density by u(x)

dU0 = E ex dex = s x dex

• Variation of strain energy

L L L
dU = ò ò dU0dAdx = ò ò sx dex dAdx = ò Pdex dx
0 A 0 A 0

dU = -dWi

53
 PMPE cont.
• Potential energy of external forces
– Force F is applied at x = L with corresponding virtual displ u(L)
– Work done by the force = Fu(L)
– The potential is reduced by the amount of work

dV = -F du (L ) F is constant
dV = -d ( Fu (L ) )
virtual displacement
– With distributed forces and concentrated force
L
V = -Fu(L ) - ò bx u( x )dx dV = -dWe
0

• PVW dU + dV = 0 or d ( U + V ) = 0

– Define total potential energy P = U +V

dP = 0

54
 PMPE cont.
• Principle of minimum potential energy
Of all displacement configurations of a solid consistent with its
displacement (kinematic) constraints, the actual one that satisfies
the equilibrium equations is given by the minimum value of total
potential energy

• Variation is similar to differentiation

dP
dP = du = 0
du
Variation = 0 Derivative = 0
N
• In general, u( x ) = å ci fi ( x )  P(u ) = P(ci )
i =1
dP
dP = 0  = 0, i = 1,, N
dc i
55
 EXAMPLE: PMPE TO DISCRETE SYSTEMS
• Express U and V in terms of u2
displacements, and then F3
1 2 3
differential  w.r.t displacements
1
• k = 100 N/mm, k = 200 N/mm
(1) (2) 3 F3
k(3) = 150 N/mm, F2 = 1,000 N 2 u3
F3 = 500 N u1
• Strain energy of elements (springs)
(1) 1 2 1 é k (1) -k (1) ù ìï u1 üï
U = k (1) ( u2 - u1 ) U (1) = ëê u1 u2 ûú ê (1) úí ý
2 2 (1´2 ) êë -k k úû ïîï u2 ïþï
(1)
( 2´2 ) ( 2´1)
1 é k (2) -k (2) ù ì u1 ü
U (2)
= ëê u1 u3 ûú ê (2) ú í ýï
ï
2 êë -k k (2) úû ï
ï u3 ï
î ï
þ

1 é k (3) -k (3) ù ìï u2 üï
U (3) = êë u2 u3 úû ê (3) úí ý
2 êë -k k (3) úû ïîï u3 ïþï

56
 EXAMPLE cont.
3
• Strain energy of the system U = åU (e)
e=1

é k (1) + k (2) -k (1) -k (2) ù ìï u1 üï

1 ê úï ï
U= êë u1 u2 u3 úû êê -k (1) k (1) + k (3) -k (3) ú ïí u2 ïý
2 úï ï
ê -k (2)
êë - k (3)
k + k úû ïî u3 ïïþ
(2) (3) ú ï

1
U = {Q } T [K ]{Q }
2 {Q} = {u1, u2, u3 }T

• Potential energy of applied forces

ìï F1 üï
ïï ïï
V = -( F1u1 + F2u2 + F3u3 ) = - êë u1 u2 u3 úû í F2 ý = -{Q} T {F}
ïï ïï
ïî F3 ïþ
• Total potential energy
1
P = U +V = {Q } T [K ]{Q } - {Q } T {F}
2
57
 EXAMPLE cont.
• Total potential energy is minimized with respect to the DOFs
¶P ¶P ¶P ¶P
= 0, = 0, =0 or, =0
¶u1 ¶u 2 ¶ u3 ¶{Q }
ìï u1 üï ì F1 ü
ïï ïï ï ï ï
ï
[ K ]í u2 ý = í F2 ïý
ï [K ]{Q } = {F}
ïï ïï ï ï ï
ï
ïî u3 ïþ ï
î 3ï
F þ Finite element equations
• Global FE equations
é 300 -100 -200 ù ìï 0 üï ìï F1 üï u2 = 6.538mm
ê úï ï ï ï
ê -100 250 -150 ú í u2 ý = í1,000 ï
ï ï ï
ý u3 = 4.231mm
ê úï ï ï ï
êë -200 -150 350 úû îï
ï u3 þï îï 500 þï
ï ï ï F1 = -1,500N

• Forces in the springs P (e ) = k (e ) ( u j - ui )

P (1) = k (1) ( u2 - u1 ) = 654N P (2) = k (2) ( u3 - u1 ) = 846N
P (3) = k (3) ( u3 - u2 ) = -346N
58
 RAYLEIGH-RITZ METHOD
• PMPE is easy to apply to discrete systems (exact solution)
– Unknown DOFs are finite
• Continuous system (DOFs are infinite)
– Use Rayleigh-Ritz method to approximate a continuous system as a
discrete system with finite number of DOFs
– Approximate the displacements by a function containing finite number
of coefficients
• Total potential energy is evaluated in terms of the unknown
coefficients
• Apply PMPE to determine the coefficients that minimizes the
total potential energy
• Solution thus obtained may not be exact
– It is the best solution from among the family of solutions that can be
obtained from the assumed displacement functions

59
 RAYLEIGH-RITZ METHOD cont.
• Assumed displacement (must satisfy the essential BC)
u ( x ) = c1f1( x ) +  + cn fn ( x )
• Determine the strain energy: U
• Find the potential of external forces: V
• Total potential energy in terms of unknown coefficients
P(c1, c2 ,...cn ) = U + V
• PMPE to determine coefficients
¶P
= 0, i = 1, n
¶c i

• After finding coefficients, determine displacement and stress

u ( x ) = c1f1( x ) +  + cn fn ( x )

du
P ( x ) = AE (for 1D bar)
dx
60
 EXAMPLE
bx
F

• L = 1m, A = 100mm2, E = 100 GPa, F = 10kN, bx = 10kN/m

• Approximate solution u( x ) = c1x + c2 x 2
2
L L1 L1 æ du ö
• Strain energy U = ò UL ( x )dx = ò AE ex2dx = ò AE çç ÷÷ dx
0 0 2 0 2 çè dx ÷ø

1 L 2 1 æ 4 ö
U ( c1, c2 ) = AE ò ( c1 + 2c2 x ) dx = AE çç Lc12 + 2L2c1c2 + L3c22 ÷÷÷
2 0 2 è 3 ø
• Potential energy of forces
L L
V ( c1, c2 ) = -ò bx ( x )u( x )dx - (-F )u(L ) = -ò bx ( c1x + c2 x 2 ) dx + F ( c1L + c2L2 )
0 0
æ L2ö æ 2 L3ö
= c1 çç FL - bx ÷÷ + c2 çç FL - bx ÷÷÷
÷
çè 2ø èç 3ø
61
 EXAMPLE cont.
• PMPE P ( c1, c2 ) = U + V
¶P 2 L2
= AELc1 + AEL c2 + FL - bx =0
¶c1 2
¶P 2 4 3 2 L3
= AEL c1 + AEL c2 + FL - bx =0
¶c 2 3 3

107 c1 + 107 c2 = -5,000 c1 = 0

7
4 ´ 10 c2 = -0.5 ´10-3
107 c1 + c2 = -6,667
3

• Approximate solution u( x ) = -0.5 ´10-3 x 2

• Axial force P ( x ) = AEdu / dx = -10,000 x
• Reaction force R = -P(0) = 0

62