Last-Modified: 3 Oct 2012

Chapter 2

FINITE ELEMENT FORMULATION

The basic aim of structural mechanics problem is to determine the distribution of
displacements and stresses under the loading and boundary conditions. A mathematical
model of the structural problem is necessary to find the desired distributions by using
FEM. An understanding of all the basic equations of structural mechanics is essential to
devise an appropriate or adequate mathematical model. Hence, the basic equations of solid
mechanics are summarized in the following sections for ready reference in the formulation
of FE equations.

2.1 BASIC EQUATIONS OF AN ELASTIC CONTINUA

In this book we assume that the continuum to be analyzed consist of an elastic material that
undergoes small strains. For a reasonable solution of a problem, following basic equations
must be satisfied.
Strain displacement relations
Constitutive relations
Equilibrium equations
Compatibility equations
Boundary conditions

2.1.1 Strain-Displacement Relations

The deformed shape of an elastic body under the external loads and temperature
distribution can be completely described by the three components of displacement u, v, and
w in the x, y, and z directions, respectively. In general, each of these components u, v, and
w is a function of the coordinates x, y, and z. The strains induced in the body can be
expressed in terms of the displacement componets.

In this section, we shall assume the deformations to be small so that the strain-
displacement relations remain linear. If the deformations are small enough, then the
equilibrium equations can be written using original geometry rather than deformed
geometry.

For the 3-D case, the strain-displacement relations can be written as follows,
 18

0 0
x
0 0
x y
y
0 0 u
z z
v (2.1)
xy
0 w
y x
yz

zx 0
z y

0
z x

where x, y, and z are the normal strains, and xy, yz, and xz are the shear strains. The
relations can be written in compact format as

d (2.2)

2.1.2 Constitutive Relations

In the case of linearly elastic isotropic three-dimensional solid, the stress-strain relations
are given by Hooke’s law as:

x 1 0 0 0 x
y 1 0 0 0 y

z E 1 0 0 0 z
1 (2.3)
xy 1 1 2 0 0 0 2
1 2 0 0 xy
0 0 0 0 1 1 2 0
yz 2 yz
0 0 0 0 0 1 1 2
zx 2 zx

where E is the Young modulus, is the Poisson ratio Eq. (2.3) can be written in a compact
format as

E (2.4)

where [ ] is a matrix of elastic coefficients.

Sometimes the expressions for strains in terms of stresses will be needed and Eqs. (2.4) can
be inverted to be obtain
 19

x 1 0 0 0 x
y 1 0 0 0 y

z 1 1 0 0 0 z
(2.5)
xy E 0 0 0 21 0 0 xy

yz
0 0 0 0 21 0 yz
0 0 0 0 0 21
zx zx

or, briefly

C (2.6)

2.1.3 Equations of Equilibrium

The equilibrium equations result from the conservation of the momentum which is an
axiom of the continuum mechanics. Two types of equilibrium equations can be written for
an elastic body: (i) external equilibrium equations, and (ii) internal equilibrium equations

The external equilibrium equations are used to find the unknown support reactions. If a
body is in equilibrium under specified static loads, the reactive forces and moments
developed at the support points must balance the externally applied forces and moments. In
other words, the force and moment equilibrium equations for the overall body have to be
satisfied.

Φ
(a)
(b)

P1
S y
y

P2
yx

yz
xy
x
zy xz

zx x
y S z

z
z x

Figure 2.1 (a) A solid body under the loads, (b) stresses on an elemental volume inside the body.
 20

The internal equilibrium equations are used to find the governing differential equations
which the stress field should satisfy inside the body. Due to the application of loads,
stresses will be developed inside the body, Fig 2.1a. If we consider an element of material
inside the body, it must be in equilibrium due to the internal stresses developed by the
loads, Fig. 2.1b. This leads to equations known as internal equilibrium equations.

x xy xz
bx 0
x y z
xy y yz
by 0 (2.7)
x y z
xz yz z bz 0
x y z

where bx, by and bz are the body forces per unit volume acting along the directions x, y and
z, respectively.

Eq. (2.2) can be writen in matrix format as

x
0 0 0
x y z y
bx 0
z
0 0 0 by 0 (2.8)
y x z xy
bz 0
0 0 0 yz
z y x
xz

This equation it can be written in a compact format as

T
d b 0 (2.9)

where the operator matrix [d] is given for the strain-displacement relations, (section 2.1.1).

The numbers of fundamental equations and unknowns are given for a 3-D problem of solid
mechanics in Table 2.1 and 2.2, respectively. Thus we have to solve as much number of
equations as there are number of unknowns to find the solution of a stress analysis
problem.

2.1.4 Compatibility Equations

When a body is continuous before deformation, it should remain continuous after

deformation. In other words, no cracks or gaps should appear in the body and no part
should overlap another due to deformation. Thus the displacement field must be
continuous as well as single-valued. This is known as the “condition of compatibility”. The
condition of compatibility can also be seen from another point of view. We can see from
 21

Table 2.1 The number of basic equations.

Equilibrium Strain-Displacement Constitutive Relations Total Number of

Equations Rel. Eqs.
3 6 6 15

Table 2.2 The number of unknowns.

Stresses Total number of

Displacements Strains
unknowns
u, v, w x, y, z, xy, yz, zx x, y, z, xy, yz, zx 15

Eq. (2.1) that the six strains x, y, z, xy, yz, and xz can be derived from only two
displacements u, v, and w. This implies that a definite relation must exist between the
strains if they correspond to compatible deformation. These definite relations are called
“compatibility equation”. Thus, in three-dimensional elasticity problems, there are totally
six compatibility equations. Two-dimensional elasticity problems are required only one.
The compatibility equations are not needed for the FE derivations and are not included in
the book. The equations can be found in the books on the theory of elasticity.

2.1.5. Boundary Conditions

Boundary conditions can be either on displacements or on stresses. The boundary

conditions on displacements require certain displacements to prevail at certain points on
the boundary of the body. The boundary conditions on stresses require that the stresses
induced must be in equilibrium with the external force applied at certain points on the
boundary of body.

x x x yx y zx z

y xy x y y zy z on S
z xz x yz y z z

where Φx, Φy, and Φz are the components of the surface traction vector Φ( )in the x, y and
z coordinate axis, respectively. x, y, and z, are the components of the unit outer normal
vector to the surface S . These equations are called Cauchy’s equations or the equilibrium
boundary conditions and their derivations can be found at the elasticity books.

We shall use the displacement based formulation method for FE derivations. Therefore we
need the displacement boundary conditions defined on the surface of the body. On the
boundary of the body it is required that the displacement fields satisfy
 22

u xS , y S , z S uS
v xS , y S , z S vS on S
w xS , y S , z S wS

where the subscript S on x, y, and z indicates that these quantities are defined on the
boundary and uS, vS, and wS, are the displacements on the boundary S at the coordinate
location (xS, yS, zS), Fig.2.1a.

2.1.6 Potential Energy

The potential energy of an elastic body is defined as

=U–W (2.10)

In this equation U is the strain energy and W is the work done on the body by the external
forces. The strain energy of a linear elastic body is defined as

1 T 1 T
U dV E dV (2.11)
2V 2V

The work done by the external forces can be stated as

I
T T T
W b dV dS1 i Pi (2.12)
V S1 i 1

where {b} is the vector of body forces,{T} is the vector of distributed surface forces, and
{P} is the vector of concentrated loads. S1 is the surface part which the distributed load
acts on, as shown at Fig. 2.1a.

2.1.7 Kinetic Energy

The kinetic energy of an elastic body is defined as

u
1 2 2 2 1 1  T
 dV
T u v w dV u v w v dV
2V 2V 2V
w

where denotes the derivative with respect to time and is the density of material.

2.2 FORMULATIONS OF SOLID AND STRUCTURAL MECHANICS

Most continuum problems, including solid and structural mechanics problems can be
formulated according to one of the two methods: differential equation method and
variational method. As seen in Chapter 1, the equations necessary in the FEA can be
derived by using either a differential equation formulation method or variational
 23

formulation method. In the case of solid and structural mechanics problems, each of the
differential equation and variational formulation methods can be classified into three
categories as shown in Table 2.3.

The displacement, force and displacement-force methods of differential equation

formulation are closely related to the principles minimum potential energy, minimum
complementary energy and stationary Reissner energy formulations, respectively. We shall
use the displacement method or the principle of minimum potential energy for presenting
the various concepts of the FEM as they have been extensively used in the literature.

Table 2.3 Methods of formulating solid and structural mechanics problems.

Formulation Methods

Differential equation
formulation methods Variational
formulation methods
Displacement method
Principle of minimum
potential energy
Force method
Principle of minimum
Displacement-force complementary energy
method
Principle of stationary
Reissner energy

2.3 PRINCIPLE OF MINIMUM POTENTIAL ENERGY

Most of the equations in this book will be derived from the potential energy theory. We
assume that strains and displacements are small and no energy dissipated in the static
loading process. That is, the external work of gradually applied loads is equal to the energy
stored in the structure, and the system is said to be conservative.

The principle says that among all the displacement states of a conservative system that
satisfy compatibility and boundary restraints, those that also satisfy equilibrium make the
potential energy, given by Eq. (2.10) stationary.

= U- W=0 (2.13)

where is the variation symbol. The variations in above equation are taken with respect to
the displacements since the potential energy depends on the displacements.

(u,v,w) = U(u,v,w) - W(u,v,w) (2.14)

 24

Example 2.1: A spring system is shown in Fig. 2.4. Determine the unknown displacements
using the principle of minimum potential energy

k3
Q
4
k1
Q1 Q2
F2
Q3
F3

k2

Figure E2.1

Solution: The total potential energy of the spring sytem cen be written as

1k 2 1k 2 1k 2 F2Q2 F3Q3
2 1 1 2 2 2 2 3 3

where 1 , 2 and 3 are the extensions of springs. They can be stated in terms of the
displacements, Q.

1 Q2 Q1 2 Q3 Q2 3 Q4 Q2

Let us substitute these relations into the total potential energy to get

1 2 1 2 1 2
2 k1 Q2 Q1 2 k 2 Q3 Q2 2 k 3 Q4 Q2 F2 Q 2 F3 Q 3

Since the total potential energy is minimum for the equilibrium condition, the first
derivative of the will be zero
0 i = 1,2,3,4
Qi

The minimization of the potential energy yields

k1 Q 2 Q1 0
Q1

k1 Q 2 Q1 k2 Q3 Q2 k 3 Q4 Q2 F2 0
Q2

k2 Q3 Q2 F3 0
Q3

k 3 Q4 Q2 0
Q4
 25

We obtain the four equilibrium equations for the four unknown displacements and may
write them in a matrix format as follows

k1 k1 0 0 Q1 0
k1 k1 k2 k3 k2 k3 Q2 F2
0 k2 k2 0 Q3 F3
0 k3 0 k 3 Q4 0

However, we know some displacements from the boundary conditions,

Q1 = 0, Q4 = 0

Hence we can reduce the equation system ignoring the first and fourth equations and
eliminating Q1 and Q4 from the retaining equations. With the other words, we eliminate the
first and the last equation completely and delete the first and last column of the coefficient
matrix.

k1 k2 k3 k2 Q2 F2
k2 k2 Q3 F3

Example 2.2: Obtain the strain energy of a prismatic rod under the simple tension. Write
the strain energy in the matrix format.

F
x

F z

Figure E2.2

Solution:

If the extension of the rod under the tension load F is , the strains are
 26

x xy 0
L

y xz 0 (a)
L

z yz 0
L

where is the Poisson ratio.

The strain energy of the rod can be determined using Eq. 2.11. Since the shear strains, and
consequently the shear stresses, are zero, the strain energy can be written as follows,

1 x
E
U x y z 1 y dV
2 1 1 2 V
1 z
(b)
1 EA 2 1 2
k
2 L 2

where k = EA/L is the extensional stiffness of the rod and it is counterpart of the stiffness
of a spring.

Now, let us write the strain energy in terms of the end displacements of the rod.

Q1 L Q2

Q2 Q1 (c)

Substituting Equation (c) into Equation (b), we get

1 EA 2
U Q 2 Q1
2 L
1 EA
Q2 2 2Q2Q1 Q12
2 L (d)
1 EA
Q2 2 Q2Q1 Q12 Q 2Q1
2 L
1 EA
Q2 Q 2 Q1 Q1 Q1 Q 2
2 L

The strain energy can be stated in a matrix multiplication format as

 27

1 EA Q1 Q2
U Q1 Q2
2 L Q2 Q1
1 EA 1 1 Q1
Q1 Q2 (e)
2 L 1 1 Q2
1 EA 1 1 Q1
Q1 Q2
2 L 1 1 Q2

If we denote

Q1
Q
Q2
(f)
EA 1 1
K
L 1 1

Equation (e) can be written as

1 T
U Q K Q (g)
2

Example 2.3: Find the displacements of rod at the points 2 and 3 using the minimum
potential energy principle.

E, 2A F E, A F
1 2 3

L L

Figure E2.3

Solution: Let us write the strain energy of the stepped rod in terms of the extensions

1 E(2A) 2 1 EA 2
U 1 2 (a)
2 L 2 L

F F
1 2 3

Q1 = 0
Q2 Q3

and in terms of the displacements

 28

EA 2 1 EA 2
U Q2 Q1 Q3 Q2
L 2 L
(b)
EA 2 1 EA 2
Q2 Q3 Q2
L 2 L

The work of external loads is

W FQ2 FQ3 (c)

The potential energy of the stepped rod can be written as

U W
EA 2 1 EA 2 (d)
Q2 Q3 Q 2 F Q 2 Q3
L 2 L

Minimization of the potential energy gives

EA EA
2Q2 Q3 Q2 F 3Q2 Q3 F 0 (e)
Q2 L L
EA
Q3 Q 2 F 0 (f)
Q3 L
Equations (e) and (f) can be written in matrix format as follows

EA 3 1 Q2 F
(g)
L 1 1 Q3 F

The solution of linear algebraic equation system given in Equation (g) gives the unknown
displacements

2FL 3FL
Q2 Q3 (h)
EA EA

Example 2.4: Solve the nodal displacements of the truss system using minimum potential
energy principle.
 29

E, 2A E, 2A

L L

L F

E, A

Figure E2.4

Solution:

Q6

Q5

Q2 Q4

Q1 Q3

The strain energy of the truss system can be written as,

1 EA 2 1 2EA 2 1 2EA 2
U 1 2 3 (a)
2 L 2 L 2 L

The elongations of the rods can be expressed as

1 Q3 Q1
2 Q5 cos60 Q6 cos30 Q3 cos60 Q 4 cos30 (b)

3 Q5 cos60 Q6 cos30 Q1 cos60 Q 2 cos30

Since the boundary conditions are

 30

Q1 Q2 Q4 0 (c)

the elongations can be written as

1 Q3
1 3 1
2 Q5 Q6 Q3 (d)
2 2 2
1 3
3 Q5 Q6
2 2

The work of external forces are

W FQ3 FQ6 (e)

The potential energy of the truss can be obtained as

2 2
EA 1 3 1 1 3
Q 32 2 Q5 Q6 Q3 2 Q5 Q6
2L 2 2 2 2 2 (f)
FQ3 FQ6

Let us apply the minimization procedure to find

EA 1 3 1
Q3 Q5 Q6 Q3 F 0
Q3 L 2 2 2

EA 1 3 1 1 3
Q5 Q6 Q3 Q5 Q6 0 (g)
Q5 L 2 2 2 2 2

EA 1 3 1 1 3
3 Q5 Q6 Q3 3 Q5 Q6 F 0
Q6 L 2 2 2 2 2

and with simplifications

EA 3 1 3
Q3 Q5 Q6 F
L 2 2 2
EA 1
Q 3 Q5 0 (h)
L 2
EA 3 3
Q3 Q6 F
L 2 2

In the matrix format equations can be stated as

 31

3 1 3
2 2 2 Q3 F
EA 1
1 0 Q5 0 (i)
L 2
Q6 F
3 3
0
2 2

The solution of linear algebraic equations gives

FL FL FL
Q3 1.2887 Q5 0.64434 Q3 0.70535 (j)
EA EA EA

2.4 FINITE ELEMENT FORMULATION FOR STATIC ANALYSIS

In this section, the finite element formulation for a three dimensional solid will be derived.
As a first step the displacement field of an element is approximated by using interpolation
functions and nodal displacements.

(a) q6 (b)
q11
q5 4
3
q10
q4 q12
y, v
y, v q3
2
e q2 e q8
q1
q2 q3 1 q7
q5 q9 3
1 q1 q6 q4
x, u 2 x, u
z, w

Figure 2.2 Nodal displacements (a) a 2-D element (b) a 3-D element

u ( x, y , z )
v ( x, y , z ) N q (2.15)
w( x, y, z )

where [N] is the shape function matrix, and {q} is the nodal displacement vector.
Substituting Eq. (2.15) into Eq. (2.2), Strain vector can be expresses in terms of nodal
 32

displacements {q}. For example, Figure 2.2 shows the nodal displacements some two-
dimensional and three-dimensional elements.

d d N q B q (2.16)

where [B] is the strain shape function matrix.

Stress vector can be expresses in terms of nodal displacements, {q}, by using the
constitutive relations given by Eq. (2.4).

E E B q (2.17)

The potential energy of an element, given by Eqs. (2.10)-(2.12) can be written in terms of
nodal displacements. Now, let us write the strain energy as
e 1 T
2
E dV
Ve
1 T T
2
q B E B q dV
Ve
1 T T (2.18)
2
q B E B dV q
Ve
1 T
2
q k q

T
where k B E B dV , element stiffness matrix.
e
V
The formulae for the element stiffness and load vector remain the same irrespective of the
type of the element. However, the order of the stiffness matrix and the load vector will
change for different types of elements.

The work done by the external forces for an element can also be written in terms of nodal
displacements as
T T
We b dV dS1
Ve S1e
T T T T
q N b dV q N dS1
e e
V S
1

T T T T (2.19)
q N b dV q N dS1
Ve S1e
T T
q fb q fs
where
 33

T
fb N b dV , element body force vector, (2.20)
Ve
and
T
fs N dS , element surface load vector. (2.21)
e
S 1

If an element has not a boundary at the surface under distributed load, the contribution of
surface forces will be zero for this element.

The potential energy of an element can be written as

e T T T
Ue We 1
2
q k q q fb q fs (2.22)

2.4.1 Assembly Procedure

The potential energy of the entire structure can be obtained from the summation of the
element energies.

E
e T
Q Fc (2.23)
e 1

{Q} is the vector of nodal displacements of the entire structure

Q1
Q2 E
Q q (2.24)
 e 1
QN

Accordingly, {q} for each element may be replaced by {Q} if the remaining element
matrices and vectors (like [B], [N], {b} and {T} in the expression e are enlarged by
adding the required number zero terms. In other words, the summation of the equation
implies the expansion of element matrices to “structure size” followed by summation of
overlapping elements.

Hence, the potential energy of entire structure can be obtained as

E E E
1 T T T T
2
q k q q fb q fs Q Fc
e 1 e 1 e 1
1 T T
2
Q K Q Q Fb Fs Fc

1 T T
2
Q K Q Q F (2.25)

Assembly procedure gives us the global stiffness matrix and load vector, respectively.
 34

E
K k
e 1

E
F fb fs Fc
e 1

The element stiffness matrix and the global stiffness matrix are always symmetric. Some of
the contributions to the load vector {F} may be zero in a particular problem. In particular,
the contribution of surface forces will be nonzero only for those element boundaries that
are also part of the boundary of the structure which is subjected to externally applied
distributed loading. Some of the components of the load vectors may be moments or even
higher order quantities if the corresponding nodal displacements represent rotations, strains
and curvatures.

F2 Q19
F1
Q21
Q20
7

5
x, u 4
z, w

Q7
y, v Q9
6 Q8
3
Q3 Q1 Q4
Q2 Q6
Q5
1
2
Figurer 2.3 Finite element model and assembly.

2.4.2 Minimization of the Total Potential Energy

The static equilibrium equations of the structure can be obtained from the principle of
minimum potential energy.

0 i 1, 2,3, N (2.26)
Qi
 35

If Eq. (2.25) is substituted into Eq. (2.26), the set of linear algebraic equations of the
overall structure can now be obtained as

K Q F (2.27)

The required solution for the nodal displacements and element stresses can be obtained
after solving above equations.

The equilibrium equation, [K]{Q}={F}, cannot be solved since the stiffness matrices [k]
and [K] are singular, and hence their inverses do not exist. The physical significance of this
is that a loaded structure is free to undergo unlimited rigid body motion (translation and/or
rotation) unless some support or boundary constraints are imposed on the structure to
suppress the rigid body motion. These constraints are called boundary conditions. We
solve the equations after incorporating the prescribed boundary conditions to obtain the
displacements.

2.4.3 Initial Strains

If some initial strains, such as temperature changes, exist in the structure, the elastic
portion of the total strain vector can be stated as

e 0 (2.28)

and stress vector is written as

E 0 (2.29)

The strain energy of an element can be found

1 T
Ue 0 E 0 dV
2V
(2.30)
1 T T 1 T
E dV E 0 dV 0 E 0 dV
2Ve V e 2Ve

T T

Figure 2.4 Temperature effects on the beams.

 36

The first term yield the element stiffness matrix. Since the last term is constant, its
contribution is zero during the minimization of the total potential energy. The second term
yields the element nodal force vector due to initial strains.

T
Wte ε E ε 0 dV
e
V
T T
q B E ε 0 dV
Ve (2.31)
T T
q B E ε 0 dV
e
V
T
q ft

where the element nodal force vector due to initial strains

T
ft B E ε0 dV (2.32)
e
V

2.5 FINITE ELEMENT FORMULATION FOR DYNAMIC ANALYSIS

In the previous section, we have discussed the static analysis of structures. Static analysis
holds when the loads are slowly applied. When the loads are suddenly applied, or when the
loads are of a variable nature, the mass and acceleration effects come into the picture. If a
solid body, such as an engineering structure, is deformed elastically and suddenly released,
it tends to vibrate about its equilibrium position. This periodic motion due to the restoring
strain energy is called free vibration. The number of cycles per unit time is called the
frequency. The maximum displacement from the equilibrium position is the amplitude. In
the real word, the vibrations subside with time due to damping action. In the simplest
vibration model, the damping effects are neglected. The undamped free vibration model of
a structure gives significant information about its dynamic behavior. We present here the
considerations needed to apply finite elements to the analysis of undamped free vibrations
of structures.

2.5.1 Lagrange Equations

We define the Lagrangean L by (2.33)

L T

where T is the kinetic energy and is the potential energy.

Hamilton’s principle: For an arbitrary time interval from t1 to t2, the state of motion of a
body extremizes the functional

t2
I Ldt (2.34)
t1
 37

If L can be expressed in terms of the generalized variables q1, q2 ,..., qn , q1, q2 ,..., qn
where qi dqi dt , then the equations of motion are given by

d L L
0 i 1, 2,..., n (2.35)
dt qi qi

2.5.2 Dynamic Equations of Motion

Now, if we consider a solid body with distributed mass (Fig. 10.1), the generalized
variables will be nodal displacements and velocities. For a solid body the potential energy
expression has already been given in the previous section. The kinetic energy is given by

v, v

u, u

w, w
y
dV

z x

Figure 10.1 Body with distributed mass.

T
T 1
2
δ δ dV (2.36)
V

where is the density (mass per unit volume) of the material and δ is the velocity vector
of point as x, given by

δ
T
u v w (2.37)

In the finite element method, we divide the body into elements, and each element we
express { } in terms of the nodal displacements {q}, using shape functions [N]. Thus,
 38

δ N q (2.38)

In dynamic analysis, the elements of {q} are dependent on time, while [N] represents
(spatial) shape functions defined on a master element. The velocity vector is then given by

δ N q (2.39)

When Eq. is substituted into Eq. , the kinetic energy Te in element e is

T T
Te 1
2
q N N dV q (2.40)
e
V

where the bracket expression is the element mass matrix

T
m N N dV (2.41)
e
V

This mass matrix is consistent with the shape functions chosen and is called the consistent
mass matrix. Mass matrices for various elements are given in the next section. On taking
summation over all the elements, we get

E E T
T Te 1 q
T
m q 1 
Q 
M Q (2.42)
2 2
e 1 e 1

The potential energy is given by

1 T T
2
Q K Q Q F (2.43)

Using the Lagrangean L = T – , we obtain the equations of motion:


M Q K Q F (2.44)

For free vibrations the force {F} is zero. Thus,


M Q K Q 0 (2.45)

For the steady-state condition, starting from the equilibrium state, we set

Q Q sin t (2.46)

where { Q }is the vector of nodal amplitudes of vibration and (rad/s) is the circular
frequency (=2 f, f= cycles/s or Hz). Introducing Eq. into Eq. , we have
 39

2
K Q M Q (2.47)

This is the generalized eigenvalue problem

K Q M Q (2.48)

where { Q }is the eigenvector, representing the vibrating mode, corresponding to the
eigenvalue . The eigenvalue is the square of the circular frequency . The frequency f
in hertz (cycles per second) is obtained from f = /(2 ).

The above equations can also be obtained by using D’Alambert’s principle and the
principle of virtual work. Galerkin’s approach applied to equations of motion of an elastic
body also yields the above set of equations.

2.5.3 Eigenvalue Problem for Vibrations

The generalized problem in free vibration is that of evaluating an eigenvalue , which is a

measure of the frequency of vibration together with the corresponding eigenvector A
indicating the mode shape, as in Eq. (2.48).

We observe here that [K] and [M] are symmetric matrices. Further, [K] is positive definite
for properly constrained problems.

Properties of Eigenvectors

For a positive definite symmetric stiffness matrix of size n, there are n real eigenvalues and
corresponding eigenvectors satisfying Eq. (2.48). The eigenvalues may be arranged in
ascending order:

0 1 2 ... n (2.49)

If Q 1 , Q 2 ,..., Q n
are the corresponding eigenvectors, we have

K Q i M Q (2.50)
i i

The eigenvectors posses the property of being orthogonal with respect to both the stiffness
and mass matrices:

T
Q
i
M Q
j
0 if i j (2.51)

T
Q
i
K Q
j
0 if i j (2.52)

The lengths of eigenvectors are generally normalized so that

 40

T
Q M Q 1 (2.53)
i i

The above normalization of the eigenvectors leads to the relation

T
Q K Q i (2.54)
i i

In many codes, other normalization schemes are also used. The length of an eigenvector
may be fixed by setting its largest component to a preset value, say unity.

Eigenvalue-Eigenvector Evaluation

The eigenvalue-eigenvector evaluation procedures fall into the following basic categories:

1. Characteristic polynomial technique

2. Vector iteration methods
3. Transformation methods

Characteristic polynomial. From Eq. (2.48) , we have

K M Q 0 (2.55)

If the eigenvector is to be nontrivial, the required condition is

det K M 0 (2.56)

This represents the characteristic polynomial in .

2.5.4 Reduction Methods

The basic idea of matrix condensation is simply Gaussian elimination of chosen

displacements to reduce the size of a problem. In static analysis no loss of accuracy results
from such a reduction because the dependent displacements are recovered (exactly) in the
back-substitution phase. In dynamic (or vibrational) analysis is smilar type of condensation
can be used to reduce the number of degrees of freedom (11), but a new type of
approximation is involved.

We begin the discussion with stiffness (or static) reduction and rewrite Eq. (2.55) in
expanded form, as follows,


Q
M AA M AB A K AA K AB QA 0
(2.57)
M BA M BB 
Q K BA K BB QB 0
B
 41

In this equation the subscript A denotes the displacements that are to be eliminated, while

the subscript B refers to those that will be retained. Let the accelerations Q 
A and Q B
be null, and write the remaining static equations as two sets:

K AA Q A K AB Q B 0 (2.58)

K BA Q A K BB Q B 0 (2.59)

Solve for the vector Q A in Eq. (2.58), as follows

1
QA K AA K AB Q B (2.60)

Substitute Eq. into Eq. (2.59) to obtain,

1
K BA K AA K AB Q B K BB Q B 0 (2.61)

and

K*BB QB 0 (2.62)

in which

1
K*BB K BB K BA K AA K AB (2.63)

From Eq. (2.62) we see that Eqs. (2.58) and (2.59) have been reduced to a smaller set,
having the same order as [KBB]. Moreover, Eq. may now be viewed as the back sustitution
process that is necessary to find vector {QA} from {QB}.

Turning next to the vibrational problem in Eq. (2.57), we shall now consider mass (or

dynamic) reduction. As a new approximation, assume that the acceleration vector Q A is

dependent upon Q B in the same manner that the vector Q A is related to Q B in Eq.
(2.60). Thus,


Q K AA
1 
K AB Q (2.64)
A B

Then equate the virtual work done by the inertial actions in the reduced system to that of
the inertial actions in the original system. Thus,
 42

T T
QB M BB QB Q M Q

Q (2.65)
T T M AA M AB A
QA QB
M BA M BB 
Q B

But from Eq. (2.60) we have,

1
QA K AA K AB QB (2.66)

Substitute Eqs. (2.66) and (2.64) into Eq. (2.65) to find

T
M*BB TB M TB (2.67)

In this equation the transformation matrix [TB] is

1
K AA K AB
TB (2.68)
IB

where [IB] is an identity matrix of the same order as [MBB]. Due to the virtual work
equality in Eq. , the mass terms in matrix M*BB are energy-equivalent to those in the
original mass matrix M . However, the reduction in size caused by Eq. (2.67) represents
an additional approximation inherent to the method.