1

MS5019 FEM

3.1. Definition of the Stiffness Matrix

z

We will consider now the derivation of the stiffness matrix for the
linear-elastic, constant-cross-sectional area (prismatic) bar element
shown in Figure 3-1.

y
y

x , u

d2 x , f2 x

1
T
d1x , f1x

Figure 3-1 Bar subjected to tensile forces T; positive nodal displacements and forces
2

MS5019 FEM

The bar element is assumed to have constant cross-sectional area A,

modulus elasticity E, and initial length L. The nodal d.o.f are local axial
displacements (longitudinal displacements directed along the length of
the bar).
From Hooke' s law and the strain/displacement relationship, we have
= E
(a)
du
=
(b)
dx
From force equilibrium, we have
A x = T = constant
(c )
for no distributed load acting on the bar. Using Eq. (b) in (a) and then
(a) in (c) and differentiating with respect to x , we obtain
d
du
(d )
AE = 0
dx
dx
3

MS5019 FEM

The following assumptions are used in deriving the bar elements

stiffness matrix:
1. The bar cannot sustain shear force; that is f1 y = 0 and f2 y = 0.

2. Any effect of transverse displacement is ignored.

3. Hooke' s law applies; that is, axial stress x is related to axial strain
x by x = E x .
The steps previously outlined in Chapter 1 are now used to derive the
stiffness matrix for bar element.

MS5019 FEM

Step 1 Select Element Type

Represent the bar by labeling nodes at each end and in general by

labeling the element number (see Figure 3-1).
z

Step 2 Select a Displacement Functions

Assume a linear displacement variation along the local axis of the bar
because a linear function with specified endpointshas a unique path.

u = a1 + a2 x

(3.1.1)

with the total number of coefficients ai always equal to the total

number of d.o.f associated with the element. Using the same procedure
as in Section 2.2 for the spring element, we express Eq. (3.1.1) as

d d
u = 2 x 1x x + d1x
L

(3.1.2)
5

MS5019 FEM

or in matrix form, Eq. (3.1.2) becomes

d
u = [N1 N 2 ] 1x
d2 x
with shape functions given by
x
x
and N 2 =
N1 = 1
L
L

(3.1.3)

(3.1.4)

The linear displacement function plotted over the length of the bar
element os shown in Figure 3-2. The bar is shown with the same
orientation as in Figure 3-1.

MS5019 FEM

d2 x

y
u

x
2
L

d1x

1
T

x
Figure 3-2 Linear displacement plotted over the length of the element

MS5019 FEM

Step 3 Define the Strain-displacement and Stress-strain

Relationships
The strain-displacement relationship is.

x =

du d2 x d1x
=
dx
L

(3.1.5)

where, Eqs. (3.1.3) and (3.1.4) have been used to obtain Eq. (3.1.5),
and the stress/strain relationship is

x = E x

(3.1.6)

MS5019 FEM

Step 4 Derive the Element Matrix and Equations

From elementary mechanics, we have

T = A x
(3.1.7)
Using Eqs. (3.1.5) and (3.1.6) in Eq. (3.1.7), we obtain
d d
T = AE 2 x 1x
(3.1.8)
L

Also, by the nodal force sign convention of Figure 3 - 1,

f = T
(3.1.9)
1x

or, by using Eq. (3.1.8), Eq. (3.1.9) becomes

AE
f1x =
d1x d2 x
(3.1.10)
L

MS5019 FEM

Similarly,

f2 x = T

(3.1.11)

or, by using Eq. (3.1.8), Eq. (3.1.11) becomes

AE
f2 x =
d 2 x d1x
(3.1.12)
L
Expressing Eqs. (3.1.10) and (3.1.12) together in matrix form, we have
f1x AE 1 1 d1x
(3.1.13)
=

f 2 x L 1 1 d2 x
Now, because f = k d , we have, from Eq. (3.1.3)
AE 1 1
k =
(3.1.14)
L 1 1
Eq. (3.1.14) represents the stiffness matrix for a truss or bar element.
In Eq. (3.1.14), AE L for a bar element is analogous to the spring
constant k for a spring element.

10

MS5019 FEM

Step 5 Assemble the Element Equations to Obtain the Total/

Global Equations
Assemble the global stiffness and forces vectors and global equations
using the direct stiffness method described in Chapter 2 can still be
adopted in this case.The method applies for structures composedd of
more than one element such that
N

K = [K ] = k ( e )
e =1

and

F = {F } = f ( e )

(3.1.15)

e =1

where now all local element stiffness matrices k must be transformed

to global element stiffness matrices k before the ditrect method is
applied as indicated by Eq. (3.1.15). (This concept of coordinate
and stiffness matrix transformations is described in Sections 3.3
and 3.4).
11

MS5019 FEM

Step 6 Solve for the Nodal Displacements

Determine the displacement by imposing boundary conditions and
simultaneously solving a system of equations, F = K d.

Step 7 Solve for the Element Forces

Finaly, determine the strains and stress in each element by backsubstitution of the displacement into equations similar to Eqs. (3.1.5)
and (3.1.6).

Example 3.1

12

MS5019 FEM

3.2. Selecting Approximation Functions for Displacement

Consider the following guidelines, as they relate to the one-dimensional
bar element, when selecting a displacement function. Further discussion
will be provided in Chapter 4 (for the beam element).
k
1. Common approximation functions (AF) are usually polynomials
such as that given by Eq. (3.1.1) or equivalently by Eq. (3.1.3),
where the functionis expressed in terms of the shape functions.
2. The AF should be continuous within the bar element. The simple
linear function of Eq. (3.1.1) certainly is continuous within the
element.
3. The AF should provide interelement continuity for all d.o.f at
each node for discrete line elements, and along common boundary
lines and surfaces for two- and three-dimensional elements.
2

13

MS5019 FEM

For the bar element, we must ensure that nodes common to two or
more elements remain common to the these elements upon
deformation and thus prevent overlaps or voids between elements.
For the two-bar structure (Figure 3-3), the linear function for
displacement within each element will ensure that elements 1 and 2
remain connected; that is, the displacement at node 2 for element 1
will equal the displacement at the same node 2 for element 2. The
linear function is then called a conforming (or compatible) function
for the bar element because it ensure both the satisfaction of
continuity between adjacent elements and of continuity within the
element.
1

1
L

Figure 3-3 Interelement continuity of a two-bar structure

14

MS5019 FEM

4. The approximation function should allow for rigid-body motion

displacement and for a state of constant strain within the element. The
1D displacement function, Eq. (3.1.1), satisfies these criteria because
a1 term allows for rigid-body motion and the a2 x term allows for
constant strain since x = du dx = a2 is a constant. (This state of
constant strain in the element can, if fact, occur if elements are chosen
small enough).
The simple polynomial Eq. (3.1.1) satisfying this fourth guidelines is
said to be complete for the bar element. The completeness of a
function is a necessary condition for convergence to the exact answer,
for instant, for displacement and stresses.
The idea that the interpolation (approximation) function must allow for a
rigid-body displacement means that the function must be capable of
yielding a constant value (say, a1), because such a value can, in fact, occur.
15

MS5019 FEM

Therefore, we must consider the case

u = a1
a = d = d
or
1

1x

(3.2.1)
(3.2.2)

2x

Using Eq. (3.2.2) in Eq. (3.1.3), we have

u = N1d1x + N 2 d2 x = ( N1 + N 2 )a1

(3.2.3)

From Eqs. (3.2.1) and (3.2.3), we then have

u = a1 = ( N1 + N 2 )a1

(3.2.4)

Therefore, by Eq. (3.2.4), we obtain

N1 + N 2 = 1

(3.2.5)

Thus, Eq. (3.2.5) shows that the displacement interpolation functions must add
to unity at every point within the element so that u will yield a constant value
when a rigid - body displacement occurs.
16

MS5019 FEM

3.3 Transformation of Vectors in Two Dimension

z

In many problem it is convenient to introduce both local and global

coordinates. Local coordinates are always chosen to conveniently
represent the individual element. Global coordinates are chosen to be
convenient for the whole structures.
Given the nodal displacement of an element, represented by the
vector d in Figure 3-4, we want to relate the components of this
vector in one coordinate system to components in another.
y
y

d
j

i
i

Figure 3-4 General displacement vector d

17

MS5019 FEM

For general purposes, we will assume that d is not coincident with

neither the local nor the global axes. In this case, we want to relate
global displacement components to local ones.

In doing that, we will develop a transformation matrix that will

subsequently be used to develop the global stiffness matrix for a bar
element.
We define the angle to be positive when measured CCW from x to x.
We can express vector displacement d in both coordinate systems
(3.3.1)
d = d i + d j = d i + d j
x

where i and j are unit vectors in the x and y directions; i and j are unit
vectors in x and y directions. We will now relate i and j to i and j through
use of Figure 3 - 5.
18

MS5019 FEM

y
y

x
b' j

a
j

a'

i
Figure 3-5 Relationship between local and global unit vectors

19

MS5019 FEM

Using Figure 3 - 5 and vector addition, we obtain

(3.3.2)
a+b = i
Also, from the law of cosines,
(3.3.3)
a = i cos
and because i is a unit vector, its magnitude is given by
(3.3.4)
i =1
Therefore, we obtain
Similarly,

a = cos

(3.3.5)

b = sin

(3.3.6)

Now a is in the i direction and b is in the - j direction. Therefore,

20

MS5019 FEM

10

a = a i = (cos )i
b = b (j) = (sin )(j)

and

(3.3.7)
(3.3.8)

Using Eqs. (3.3.7) and (3.3.8) in Eq. (3.3.2) yields

i = cos i sin j
(3.3.9)
Similarly, from Figure 3 - 5, we obtain
a'+ b' = j
(3.3.10)
a' = cos j
(3.3.11)
b' = sin i
(3.3.12)
Using Eqs. (3.3.11) and (3.3.12) in Eq. (3.3.10), we have
j = sin i + cos j
(3.3.13)
21

MS5019 FEM

Now, using Eqs. (3.3.9) qnd (3.3.13) in Eq. (3.3.1), we have

d (cos i sin j) + d (sin i + cos j) = d i + d j
x

(3.3.14)

Combining like coefficient of i and j in Eq. (3.3.14), we obtain

d cos + d sin = d
x

d x sin + d y cos = d y

(3.3.15)

In matrix form, Eqs. (3.1.15) are written

d x C S d x
=

d y S C d y
where C = cos and S = sin .

(3.3.16)

and

Eq. (3.3.16) relates the global displacement d to the local displacement d . The matrix
C S
(3.3.17)
S C

is called the transformation matrix.

22

MS5019 FEM

11

For the case of d y = 0, we have, from Eq. (3.3.1)

d i + d j = d i
x

(3.3.18)

Figure 3 - 6 shows d x expressed in term of global x and y components.

Using trigonometry and Figure 3 - 6, we then obtain the magnitude of d x as
d = Cd + Sd
(3.3.19)
x

Eq. (3.3.19) is equivalent to the first equation of Eq. (3.3.16).

y
y

dy

d x

dx

Figure 3-6 Relationship between local and global displacements

23

MS5019 FEM

3.4. Global Stiffness Matrix

We will now use the transformation relationship Eq. (3.3.16) to obtain
the global stiffness matrix for a bar element. We need the global
stiffness matrix of each element to assemble the global stiffness matrix
of the structure. We have shown in Eq. (3.1.13) that for a bar element in
the local coordinate system,
f1x AE 1 1 d1x
(3.4.1)
=

f 2 x L 1 1 d2 x
or
f = k d
(3.4.2)

We now want to relate the global element nodal forces f to the global
nodal displacement d for a bar element arbitrarily oriented with respect
to the global axes as was shown in Figure 3-1.
24

MS5019 FEM

12

The relationship between f and d will yield the global stiffness matrix k
of the element such that
f1x
d1x
f
d
1y
1y
=
(3.4.3)
k
f2x
d 2 x
f 2 y
d 2 y
or, in simplified matrix form, Eq. (3.4.3) becomes
f = kd
(3.4.4)
We observe from Eq. (3.4.3) that a total of four components of force
and four of displacement arise when global coordinates are used.
However, a total of two components of force and two of displacement
appear for the local-coordinate representation of a spring or bar, as
shown by Eq. (3.4.1).
25

MS5019 FEM

By using relationship between local and global force components and

between local and global displacement components, we will be able to
obtain the stiffness matrix.
We know from transformation relationship Eq. (3.3.15) that
d1x = d1x cos + d1 y sin
(3.4.5)
d = d cos + d sin
2x

2x

2y

or, in matrix form, Eqs. (3.4.5) can be written

d1x
d1x C S 0 0 d1 y
=

d 2 x 0 0 C S d 2 x
d 2 y
or as
d = T*d

(3.4.6)

(3.4.7)
26

MS5019 FEM

13

C S 0 0
(3.4.8)
T* =

0 0 C S
Similarly, for force transformation, we have
f1 x
f1x C S 0 0 f1 y
(3.4.9)
=

f 2 x 0 0 C S f 2 x
f 2 y
or as
(3.4.10)
f = T*f
Now, substituting Eq. (3.4.7) into Eq. (3.4.2), we obtain
(3.4.11)
f = k T*d
where

and using Eq. (3.4.10) in (3.4.11)

T*f = k T*d

(3.4.12)
27

MS5019 FEM

However, to write the final expression relating global nodal forces to

global nodal displacement fo an element, we must invert T* in Eq. (3.4.12).
This is not immediately possible because T* is not a square matrix.
Therefore, we must expand d , f , and k to the order that is consistent with
the use of global coordinates even though f1 y and f2 y are zeros.
Using Eq. (3.3.16) for each nodal displacement we obtain
d1x C S 0 0 d1x


d1 y S C 0 0 d1 y
(3.4.13)

=
d 2 x 0 0 C S d 2 x
d2 y 0 0 S C d 2 y

or as
(3.4.14)
d = Td

where

C
S
T=
0

0
0
S

(3.4.15)

28

MS5019 FEM

14

Similarly, we can write

f = T f

(3.4.16)
because forces are like displacement - both are vectors. Also, k must be expanded
to 4 4 matrix. Therefore, Eq. (3.4.1) in expanded form becomes
f1x
1 0 1 0 d1x

f1 y AE 0 0 0 0 d1 y
(3.4.17)
=

f 2 x L 1 0 1 0 d 2 x

f2 y
0 0 0 0 d2 y

In Eq. (3.4.17), since f1 y and f2 y are zero, rows of zeros corresponding to the row
numbers f and f appear in k . Now using Eqs. (3.4.14) and (3.4.16) in Eq. (3.4.18),
1y

2y

we obtain
T f = k Td
(3.4.18)
Equation (3.4.18) is Eq. (3.4.12) expanded. Now, premultiplying both sides of
Eq. (3.4.18) by T 1 , we have
29

MS5019 FEM

f = T 1k Td

(3.4.19)

where T is the inverse of T. However, it can be shown, that

-1

T 1 = TT

(3.4.20)

where T is the transpose of T. The property of square matrices such as T given

by Eq. (3.4.10) defines T to be an orthogonal matrix. The transformation matrix
T between rectangular coordinate frames is orthogonal. This property of T is
used throughout this text. Substituting Eq. (3.4.20) into Eq. (3.4.19), we obtain.
(3.4.21)
f = T T k Td
Equating Eqs. (3.4.4) and Eq. (3.4.21, we obtain the global stiffness matrix for
an element as
k = T T k T
(3.4.22)
Substituting Eq. (3.4.15) for T and the expanded form of k given in Eq. (3.4.17)
into Eq. (3.4.22), we obtain k given in explicit form by
30

MS5019 FEM

15

 C2
CS C 2 CS

S 2 CS S 2
AE
(3.4.23)
k=
L
C2
CS

S 2
symetry
Now, since the trial displacement functiom Eq. (3.1.1) was assumed piecewise continuous element by element, the stiffness matrix for each element can be summed
using the direct stiffness method to obtain.
N

(e)

=K

(3.4.24)

e =1

where K is the total stiffness matrix and N is the total number of elements.
Similarly, each element global nodal force matrix can be summed such that
N

(e)

=F

(3.4.25)

e =1

K is now related the global nodal force F and the global nodal displacement d for
the whole structure by
F = Kd
(3.4.26)
31

MS5019 FEM

3.5. Computational of Stress for a Bar in x-y Plane

We will now consider the determination of the stress in a bar element.
For a bar, the local forces are related to the local displacement by Eq.
(3.1.13). This equation is repeated here for convenience.
f1x AE 1 1 d1x
(3.5.1)
=

f 2 x L 1 1 d2 x
The usual definition of axial tensile stress is
f
= 2x
(3.5.2)
A
where f2 x is used because it pulls on the bar as shown in Figure 3 - 7.
From Eq. (3.5.1) we have

AE
[ 1 1] d1x
f2 x =
(3.5.3)
L
d 2 x
32

MS5019 FEM

16

f2 x

L
f1x

Figure 3-7 Basic bar element with positive nodal forces

Therefore, combining Eqs. (3.5.2) and (3.5.3) yields

E
= [ 1 1]d
L
Now, using Eq. (3.4.7), we obtain
E
= [ 1 1]T*d
L

(3.5.4)

(3.5.5)
33

MS5019 FEM

Equation (3.5.5) can be expressed in similar form as

= C' d
where, using Eq. (3.4.8),
C S 0 0
E
C' = [ 1 1]

L
0 0 C S
After multiplying the matrix in Eq. (3.5.7), we have
E
C' = [ C S C S ]
L

(3.5.6)

(3.5.7)

(3.5.8)

34

MS5019 FEM

17

3.6. Solution of a Plane Truss

We will now illustrate the use of equations developed in Section 3.4
qnd 3.5, along with the direct stiffness method of assembling the total
matrix and equatons, to solve the following plane truss example
problem.
A plane truss is a structure composed of bar elements all lying in a
common plane that connected together by frictionless pins. The plane
truss also must have loads acting only in common plane.

EXAMPLE 3.5

35

MS5019 FEM

3.7. Transformation Matrix and Stiffness Matrix

for a Bar in Three-Dimensional Space
We will now derive the tranformation matrix for a bar element in 3-D
space as shown in Figure 3-8.

Figure 3-8 Bar in 3-D space

36

MS5019 FEM

18

Let the node 1 and 2 have the coordinates ( x1 , y1 , z1 ) and ( x2 , y2 , z 2 ) respectively.

Also, let x , y , and z be the angles measured from global x, y, and z axes,
respectively, to the local axis x. Here x directed along the element from node 1 to
node 2. We must determine T* such that d = T*d. We begin the derivation of T*
by considering the vector d = d expressed in 3 - D as
d i + d j + d k = d i + d j + d k

(3.7.1)

Taking the dot product of Eq. (3.7.1) with i , we have

d + 0 + 0 =d (i i ) + d (i j) + d (i k )

(3.7.2)

and, by definition of the dot product,

i i = x2 x1 = C
x
L
i j = y2 y1 = C
y
L
i k = z2 z1 =C
z
L

(3.7.3)

37

MS5019 FEM

where L = ( x2 x1 ) 2 + ( y2 y1 ) 2 + ( z 2 z1 ) 2
and
(3.7.4)
C x = cos x
C y = cos y
C z = cos z
Here C , C , and C are the projections of i on i, j, and k , respectively.
x

Therefore, using Eqs. (3.7.3) in Eq. (3.7.2), we have

d = C d + C d + C d
x

(3.7.5)

For a vector in space directed along the x axis, Eq. (3.7.5) gives the
components of that vector in the global x, y, and z directions.
Now using Eq. (3.7.5), d = T*d can be written in explicit form as

d 1x C x
=
d 2 x 0

Cy

Cz

Cx

Cy

d1x
d
1y
0 d1z
=
C z d 2 x
d 2 y

d 2 z

(3.7.6)

38

MS5019 FEM

19

where
0
0
C x C y C z 0
(3.7.7)
T =
0
0 C x C y C z
0
is the transformation matrix, which enables the local displacement matrix d to be
expressed in terms of displacement components in the global coordinate system.
We have shown in Section 3.4 that the global stiffness matrix is given in general by
k = TT k T. This equation will now be used to express the general form of the stiffness
matrix of a bar arbitrary oriented in space. In general, we must expand the transformation
matrix in a manner analogous to that done in expanding T* to T in Section 3.4. However,
the same result will be obtained here by simply using T* , defined by Eq. (3.7.7), in place
of T. Then k is obtained using the equation k = (T* ) T k T* as follows :

39

MS5019 FEM

where
C x
C
y
C
k= z
0
0

0
0
0
0
0 AE 1 1 C x C y C z 0

0
0 C x C y C z
C x L 1 1
Cy

C z
Simplifying Eq. (3.7.8), we obtain the explicit form of k as

(3.7.8)

C x2
C x2
C xC y C xC z
C xC y C xC z

2
C y2
C yCz
Cy
C y C z C xC y

C xC z C yC z
C z2
C z2
AE
(3.7.9)
k=

L
C x2
C xC y
C xC z

C y2
C yCz

C z2
Symetry
Equation (3.7.9) is the basic form of the stiffness matrix for a bar element in 3 - D space.
40

MS5019 FEM

20

3.8. Potential Energy Approach

We now present the principle of minimum potential energy (POMPE)
to derive the bar element equations. Recall from Section 2.6 that the
total PE, p was defined as the sum of the internal strain energy U and
the potential energy of the external forces as

p =U +

(3.8.1)

To evaluate the strain energy for a bar, we consider only the work done
by the internal forces during deformation. Because we are dealing with
a 1-D bar, the internal force doing work is given in Figure 3-9 as
x(y)(z), due only to normal stress x. The displacement of the x face
of element is x(x); the displacement of x + x face is x(x + dx). The
change in displacement is then xdx, where dx is differential change in
strain occuring over element x.
41

MS5019 FEM

Figure 3-9 Internal force in a 1-D bar

The differential internal work (or strain energy) dU is the internal force multiplied by
the displacement through which the force moves, given by
dx4
F 48 6
647
47
8
dU = x ( x )( y )( z )d x

(3.8.2)

Rearranging and letting the volume of the element approach zero, we obtain, from
Eq. (3.8.2).
dU = x d x dV
(3.8.3)
For the whole bar, we the have
x

U = x d x dV

V
0

(3.8.4)
42

MS5019 FEM

21

Now, for a linear-elastic (Hookes law) material as shown in Figure 3-10,

we see that x = Ex. Hence, substituting this relationship into Eq. (3.8.4),
integrating with respect to x, and then substituting x for Ex, we have

U=

1
x x dV
2
V

(3.8.5)

as the expression for the strain energy for 1 - D stress.

Figure 3-10 Linear-elastic (Hookes law) material

43

MS5019 FEM

The PE of the external forces, being opposite in sign from the external
work expression because the PE of external forces is lost when the work
is done by the external forces, is given by
M

= X budV Tx udS fix dix

1 24
i=
V 424
S1
4
3
1
3 1
424
3 1
body forec

(3.8.6)

nodal force

surface loading

where the first, second, and third terms on the right side of Eq. (3.8.6)
represent the PE of (1) body force X (in units of force per unit
b

volume), (2) surface loading Tx (in units of force per unit surface area),
and (3) nodal concentrated forces f .
ix

The forces X b , Tx , and fix are considered to act in the local x direction
of the bar as shown in Figure 3 - 11.
44

MS5019 FEM

22

In Eq. (3.8.5) and (3.8.6), V is the volume of the body

and S1 is the part of the surface S on which surface
loading acts. For a bar element with two nodes and one
d.o.f per node, M = 2.
We are now ready to describe the FE formulation of the
bar element equations using th POMPE.

Figure 3-11
General forces acting
on a 1-D bar

The FE process seeks a minimum in the PE within the

constraint of an assumed displacement pattern within
each element. The greater the number of d.o.f.
associated with the element, the more closely will the
solution approximate the true one and ensure complete
equilibrium. An approximate FE solution using the
stiffness method will always provide an approximation
value of PE greater than or equal to the correct one.
45

MS5019 FEM

The method also results in a structure behavior that is predicted to be

physically stiffer than, or at best to have the same stiffness as, the actual
one. This is explained by the fact that structure model is allowed to
displaced only into shapes defined by the erms of the assumed
displacement field within each element of the structure. The correct shape
is ussually only approximated bya the assumed field, although the correct
shape can be the same as the assumed field. The assumed field effectively
constraints the structure from deforming in its natural manner, This
constraint effect stiffens the predicted behavior of the structure.
Apply the following steps when using the POMPE to derive the FE
equations.
1. Formulate an expression for the total PE.
2. Assume the displacement pattern to vary with a finite set of
undetermined parameters (nodal displacements).
3. Obtain a set of simultaneous equations minimizing the total PE with
respect to these nodal displacements.
46

MS5019 FEM

23

The resulting equations are the approximate (or possibly exact)

equilibrium equations whose solution for the nodal parameters seeks to
minimize the PE when back-substitued into the PE expression. The
proceeding three steps will now followed to derive the bar element
equations and stiffness matrix.
Consider the bar element of length L, with constant cross-sectional area A,
shown in Figure 3-11. Using Eqs. (3.8.5) and (3.8.6), the total PE, Eq.
(3.8.1), becomes.
L

A
p = x x dx f1x d1x f2 x d2 x uTx dS uX b dV
(3.8.7)
20
S
V
since A is contant and variables x and x at most vary with x.

47

MS5019 FEM

From Eqs. (3.1.3) and (3.1.4), we have the axial displacement function
expressed in terms of the shape functions and nodal displacement by
(3.8.8)
u = [N ] d

{}

where

[N ] = 1

x
L

x
L

(3.8.9)

and

{d}= dd

1x
(3.8.10)

2x
Then, using the strain/displacement relationship x = du dx , the axial
strain can be written as
{ x } = 1 1 d
(3.8.11)
L L

{}

48

MS5019 FEM

24

or

{ x } = [B ] {d}

(3.8.12)

where we define

[B] = 1

1
L

(3.8.13)

The axial stress/strain relationship is given by

[ x ] = [D] { x }
where

(3.8.14)

[D] = [E ]

(3.8.15)
for the 1 - D stress/strain relationship and E is the modulus of elasticity.
Now, by Eq. (3.8.12), we can express Eq. (3.8.14) as

[ x ] = [D][B]{d}

(3.8.16)
49

MS5019 FEM

Using Eq. (3.8.7) expressed in matrix notation form, we have the total
PE given by

{}

{ }

{ }

T
A
p = { x }T { x }dx d {P} {u}T T x dS {u}T X b dV (3.8.17)
20
V
S
where {P} now represents the concentrated nodal loads and where in general
both x and x are column matrices. For proper matrix multiplication we must
place the transpose on { }. Similarly,{u} and T in general are column

{}
x

matrices, so for proper matrix multiplication {u}is transposed in Eq. (3.8.17).

Using Eqs. (3.8.11), (3.8.12), and (3.8.16) in Eq. (3.8.17), we obtain

p =

{ } [B] [D] [B]{d}dx {d} {P} {d } [N ] {T }dS

{d } [N ] {X }dV
L

A
d
2 0

(3.8.18)

50

MS5019 FEM

25

In Eq. (3.8.18), p is seen to be a function of {d}; that is, p = p (d1x , d2 x ).

However,[ B] adn [ D], Eqs. (3.8.13) and (3.8.15), and the nodal d.o.f d and d
1x

2x

are not functions of x. Therefore, integrating Eq. (3.8.18) with respect to x yields
T
AL T T T
d [B ] [D] [B ] d d f
(3.8.19)
p =
2
where
T
T
f = {P}+ [N ] T dS + [N ] X dV
(3.8.20)

{}

{}

{} {}{ }

{ }

{ }

From Eq. (3.8.20), we observe three separate types of load contributions from
body forces, surface tractions, and concentrated nodal forces. We define these
surface tractions and body - force vectors as
T
f = [N ] T dS
(3.8.20a)

{ } { }
{f }= [N ] {X }dV
s

(3.8.20b)

51

MS5019 FEM

Minimizing of p with respect to each nodal displaceme nt requires that

p
=0
d

p
=0
d

and

1x

(3.8.21)

2x

Now we explicitly evaluate p given by Eq. (3.8.19) to apply Eq. (3.8.21).

We define the following for convenienc e :

{U } = {d} [B ] [D ] [B ]{d }
T

(3.8.22)
Using Eqs. (3.8.10), (3.8.13), and (3.8.15) in Eq. (3.8.22) yields
d
1
U * = d1 x d2 x 1 L [E ][ L1 L1 ] 1 x
(3.8.23)
d2 x
L
Simplifyin g Eq. (3.8.23), we obtain
E
U * = 2 d12x 2 d1 x d2 x + d22x
(3.8.24)
L

{ } [
{ }

52

MS5019 FEM

26

{ } {f } is

Also, the explicit expression for d

{d} {f }= d

f + d2 x f 2 x
(3.8.25)
Therefore, using Eqs. (3.8.24) and (3.8.25) in Eq. (3.8.19) and the applying
Eqs. (3.8.21), we obtain
p AL E

2d1 x 2d2 x f1x = 0

=
2

2 L
d 1 x

and
(3.8.26)
p AL E

=
2d1 x + 2d2 x f2 x = 0
2

2 L
d

1x 1x

2x

In matrix form, we express Eq. (3.8.26) yields

p AE 1 - 1 d1 x f1 x 0
=
=
L - 1 1 d2 x f2 x 0
d

{}

(3.8.27 )

53

MS5019 FEM

{ } [ ] {d}, we have the stiffness matrix for the bar element as

[k] = AEL -11 -11
(3.8.28)

or since f = k

Finally, instead of the cumbersome process of explicitly evaluating p , we can

use the matrix differenti ation and apply it directly to Eq. (3.8.19) to obtain
p
T
= AL[B ] [D ][B ] d f = 0
(3.8.29)
d 1 x
where [D ] = [D ] has been used in writtin g Eq. (3.8.29). The result of the
T
evaluation of AL[B ] [D ][B ] is then equal to k given by Eq. (3.2.28).

{} {}

[]

Throughout this text, we will use this matrix differenti ation concept, which
greatly simplifies the task of evaluating k .

[]

54

MS5019 FEM

27

3.9. Galerkins Residual Method

We have develop the bar FE equations by the direct method in Section
3.1 and by the PE method (one of number of variation methods) in
Section 3.8. In fields other than structural/solids mechanics, it is quite
probable that a variational principle, analogous to the principle of
minimum PE, for instance, may not be known or even exist. In some
flow problems in fluid mechanics and in mass transport problems, we
often have the differential equations and BC available. However, the FE
method can still be applied.
The weighted residual method (WRM) applied directly to the
differential equation can be used to develop the FE equations. In this
section, we describe Galerkins residual method (GRM) in general and
the apply it to the bar element.
55

MS5019 FEM

This development provides the basis for later applications of GRM to

the beam element in Chapter 4 and to the non-structural problems.
There are a number of other WRM. Among these are collocation, subdomain method, least square, and least square collocation. (For more on
these methods, see Reference [4].). However, since GRM is more well
known than the other WRM, it is the only one described in this text.
In WRM, a trial or approximate function is chosen to approximate the
independent variable, such as a displacement or a temperature, in a
problem defined by a differential equation. This trail function will not,
in general, satisfy the governing differential equation. Thus, the
substitution of the trail function into the differential equation result in a
residual over the whole region of the problem as follows
R dV = minimum

(3.9.1)
56

MS5019 FEM

28

In the WRM, we require that a weighted value of the residual be a

minimum over the whole region. The weighting functions allow the
weighted integral of residuals to go to zero. Denoting the weighting
function by W, the general form of the weighted residual integral is
RW dV = minimum

(3.9.2)

Using GRM, we chose the interpolation function, such as Eq. (3.1.3), in

terms of Ni shape functions for the independent variable in the
differential equation. In general, this substitution yields the residual R
0. By the Galerkin criterion, the shape functions are chosen to play the
role of the weighting functions W. Thus, for each i we have.
RN

dV = 0 (i = 1,2,3, L , n)

(3.9.3)

Eq. (3.9.3) results in total of n equations.

57

MS5019 FEM

Equation (3.9.3) applies to points within the region of a body without

reference to BC such as specified applied loads or displacement. To
obtain BC, we apply integration by parts to Eq. (3.9.3), which yields
integrals applicable for the region and its boundary.

Bar Element Formulation

We now use GRM to formulate the bar element stiffness equations. We
begin with the basic element differential equation, without distributed
load, derived in Section 3.1 as

d
du
( AE ) = 0
dx
dx

(3.9.4)

where constant A and E are now assumed.

58

MS5019 FEM

29

The residual R is now defined to be Eq. (3.9.4). Applying Galerkins

criterion, Eq. (3.9.3), to Eq. (3.9.4), we have
L

du

dx ( AE dx ) N dx = 0
i

(i = 1,2)

(3.9.5)

We now apply integration by parts to Eq. (3.9.5). Integration by parts

is given in general by

u dv = uv v du

(3.9.6)

where u and v are simply variables in the general equation. Letting

dN i
u = Ni
du =
dx
dx
(3.9.7)
d
du
du
dv = ( AE )dx
v = AE
dx
dx
dx
in Eq. (3.9.5) and integrating by parts according to Eq. (3.9.6), Eq. (3.9.5)
becomes
59

MS5019 FEM

du
du dN i
N i AE
dx = 0
AE
(3.9.8)
dx 0 0
dx dx
where the integration by parts introduces the boundary conditions.
Recall that, because u = [N ] d , we have

{}

du dN1
dN 2
d 1x +
d2x
(3.9.9)
=
dx
dx
dx
or using Eqs. (3.1.4) for N1 and N 2 , we obtain
du 1 1 d 1x
(3.9.10)
=

dx L L d2 x
Using Eq. (3.9.10) in Eq. (3.9.8), we then express Eq. (3.9.8) as
L
L
dN i 1 1 d 1x
du
AE
(3.9.11)
L L dx d = N iAE dx

d
x
0
0
2x
60

MS5019 FEM

30

Equation (3.9.11) is really 2 equations (one for N i = N1 and one for N i = N 2 ).

First, using the weighting function N i = N1 , we have
L

dN 1
AE 1
dx L
0
Substituting for

L
1 d 1x
du
dx = N1 AE
L d2 x
dx 0

(3.9.12)

dN1
, we obtain
dx

1 1
AE
L L
0

1 d 1x
dx = f1x
L d2 x

(3.9.13)

where f1x = AE du dx because N1 = 1 at x = 0 and N1 = 0 at x = L.

Evaluating Eq. (3.9.13) yields

AE
d 1x d2 x = f1x
L

(3.9.14)
61

MS5019 FEM

Similarly, using N i = N 2 , we obtain

L

1 1
AE
L
L
0

du
1 d 1x

d
x
N
AE
=

2
L d2 x
dx 0

(3.9.15)

Simplifying Eq. (3.9.15) yields

AE
d 2 x d1x = f2 x
L

(3.9.16)

where f 2 x = AE (du dx ) because N 2 = 1 at x = L and N 2 = 0 at x = 0.

Equations (3.9.14) and (3.9.16) can be written in matrix form as
f1x AE 1 1 d1x
(3.9.17)
=

f 2 x L 1 1 d 2 x
Eq. (3.9.17) then seen to be the same as Eqs. (3.1.13) and (3.8.27) derived,
respectively, by the direct and variational methods.
62

MS5019 FEM

31

Reference:
1. Logan, D.L., 1992, A First Course in the Finite Element Method,
PWS-KENT Publishing Co., Boston.
2. Imbert, J.F.,1984, Analyse des Structures par Elements Finis,
2nd Ed., Cepadues.
3. Zienkiewics, O.C., 1977, The Finite Eelement Method, 3rd ed.,
McGraw-Hill, London.
4. Finlayson, B.A., 1972, The Method of Weighted Residuals and
Variational Principles, Academic Press, New York.

63

MS5019 FEM

32