18

Introduction to the
Finite Element Method

PREVIEW

The classical analysis of distributed-mass systems with infinite number of DOFs is not
feasible for practical structures, for reasons mentioned in Chapter 17. In this chapter,
two methods are presented for discretizing one-dimensional distributed-mass systems:
the Rayleigh–Ritz method and the finite element method. As a result, the governing
partial differential equation is replaced by a system of ordinary differential equations, as
many as the DOFs in the discretized system, which can be solved by the methods
presented in Chapters 10 to 16. The consistent mass matrix concept is introduced and the
accuracy and convergence of the approximate natural frequencies of a cantilever beam,
determined by the finite element method using consistent or lumped-mass matrices, is
demonstrated. The chapter ends with a short discussion on application of the finite
element method to the dynamic analysis of structural continua.

PART A: RAYLEIGH–RITZ METHOD

18.1 FORMULATION USING CONSERVATION OF ENERGY

Developed in Chapter 15 for lumped-mass systems, the Rayleigh–Ritz method is also

applicable to systems with distributed mass and elasticity. It was for the latter class of
systems that the method was originally developed by W. Ritz in 1909. The method, ap-
plicable to any one-, two-, or three-dimensional system with distributed mass and elas-
ticity, reduces the system with an infinite number of DOFs to one with a finite number of

729
730 Introduction to the Finite Element Method Chap. 18

DOFs. In this section we present the Rayleigh–Ritz method for the transverse vibration
of a straight beam.
Consider such a beam with flexural rigidity EI (x) and mass m(x) per unit length,
both of which may vary arbitrarily with position x . The deflections u(x) of the system
are expressed as a linear combination of several trial functions ψj (x):
.
N
u(x) = z j ψj (x) = Ψ(x)z (18.1.1)
j =1

where zj are the generalized coordinates, which vary with time in a dynamic problem, z is
the N × 1 vector of generalized coordinates, and the 1 × N matrix of trial functions is
Ψ(x) = [ψ1 (x) ψ2 (x) · · · ψN (x)]
Each trial function ψj (x)—also known as Ritz function or shape function—must be
admis- sible: that is, continuous and have a continuous first derivative, and satisfy the
displacement boundary conditions on the system. All the trial functions must be linearly
independent and are selected appropriate for the system to be analyzed.
The starting point for formulation of the Rayleigh–Ritz method is the Rayleigh’s
quotient [Eq. (8.5.11)] for a function φ˜ (χ ) defined consistent with Eq. (18.1.1):

φ˜ (x) = Ψ(x)χ (18.1.2)

Replacing ψ(x) in Eq. (8.5.11) by φ˜ (x) and substituting Eq. (18.1.2) leads to
. .
ρ(χ) = N N χ χ
i j
i j
. =1 . =1 ˜ (18.1.3)
k ij
i j m˜ i j
N N χ χ
i j
=1 =1
where
∫ ∫
L L
k˜ij = EI (x)ψi ××
(x)ψ j
×× dx
(x)
m˜ i j m(x)ψ i (x)ψ j (x) dx (18.1.4)
0 0
=
Rayleigh’s quotient cannot be determined from Eq. (18.1.3) because the N generalized
coordinates zi are unknown. Our objective is to find the values that provide the “best”
approximate solution for the natural vibration frequencies and modes of the system.
For this purpose we invoke the property that Rayleigh’s quotient is stationary in the
neighborhood of the true modes (or true values of χ), which implies that ∂ρ/∂ χ i =0—
i = 1, 2,. , N . We do not need to go through the details of the derivation because the
ρ( χ ) for distributed-mass systems, Eq. (18.1.3), is of the same form as that for discretized
systems, Eq. (15.3.8). Thus the stationary condition on Eq. (18.1.3) leads to the
eigenvalue problem of Eq. (15.3.11), which is repeated here for convenience:
k˜ χ = ρm˜ χ (18.1.5)
where k˜ and m˜ are square matrices of order N , the number of Ritz functions used to rep-
resent the deflections u(x) in Eq. (18.1.1), with their elements given by Eq. (18.1.4). The
original eigenvalue problem for distributed-mass systems, Eq. (17.3.6), has been reduced
to a matrix eigenvalue problem of order N by using Rayleigh’s stationary condition.
 Sec. 18.1 Formulation Using Conservation of Energy 731

The solution of Eq. (18.1.5), obtained by the methods of Chapter 10, yields N
eigen- values ρ1, ρ2, , ρN and the corresponding eigenvectors
χn = ∗χ 1n χ 2n ··· χ N n )T n = 1, 2, . . . , N (18.1.6)
The eigenvalues provide
√
ω˜ n = ρn (18.1.7)
When arranged in increasing order of magnitude, ω ˜ n are upper bound approximations to
the true natural frequencies ωn of the system, that is,
ω˜ n ≥ ωn n = 1, 2, . . . , N (18.1.8)
Furthermore, an approximate frequency approaches the exact value from above as the
num- ber N of Ritz functions is increased. The eigenvectors χn substituted in Eq. (18.1.2)
pro- vide the functions:
N
.
˜
φ n (x) = χ jn ψ j (x) = Ψ(x)χn n = 1, 2, . . . , N (18.1.9)
j =1
which are approximations to the true natural modes φ n (x) of the system. The quality of
these approximate results is generally better for the lower modes than for higher modes.
Therefore, more Ritz functions should be included than the number of modes desired for
dynamic response analysis of the system. Although the natural modes φ˜ n (x) are
approx-
imate, they satisfy the orthogonality properties of Eq. (17.4.6)—as demonstrated for dis-
cretized systems (Section 15.3.4)—and can therefore be used in modal analysis of the
system as described in Sections 17.5 to 17.7.
Example 18.1
Find approximations for the first two natural frequencies and modes of lateral vibration of a
uniform cantilever beam of Fig. E18.1a by the Rayleigh–Ritz method using the shape func-
tions shown in Fig. E18.1b:
πx
3πx
ψ1 (x) = 1 − cos ψ2 (x) = 1 − 2L (a)
cos 2L

(a)
(b) (c)

ψ1(x) ψ2(x)
~1(x)
L ~2(x)
1(x)
m, EI
x
2(x)

Figure E18.1
732 Introduction to the Finite Element Method Chap. 18

These functions are admissible because they are continuous, have a continuous first
derivative, and satisfy the two displacement boundary conditions at the clamped end.
However, they do not satisfy one of the force boundary conditions at the free end. Note that
the two shape functions satisfy the requirement of linear independence.
Solution
1. Set up k˜ and m˜ . For the selected Ritz functions of Eq. (a), the stiffness
coefficients are computed from Eq. (18.1.4a):
∫L.
1 π 4 EI πx 1 π 4 EI
k˜11 = Σ
cos
2
2L
dx = (b1)
L4 0
32 L3
∫ L
4
9 π EI πx 3 πx
k˜12 = cos cos dx = 0 (b2)
L4 0 2L 2L
k˜21 = k˜12 = 0
16 (b3)
∫ L. Σ2
4 4
81 π EI 3πx 81 π EI
k˜22 = 4
cos
2L dx = (b4)
L 0 32 L 3
16
Similarly, the mass coefficients are determined from Eq. (18.1.4b):
∫ L. Σ2
πx
m˜ 11 = 1 − cos
m
0 2L dx = 0.2268mL (c1)
∫ L . Σ. Σ
πx 3πx
m˜ 12 = 1 − cos 1− dx = 0.5756mL (c2)
0 2L
m 2L cos
m˜ 21 = m˜ 12 = 0.5756m L (c3)

m˜ 22 =m∫ L . 2Σ
3πx
1− dx = 1.9244mL (c4)
0 2L
cos
Substituting Eqs. (b) and (c) in Eq. (18.1.5) gives

Σ Σ. χ Σ Σ Σ. Σ
10 81
0 χ 21 0.5756 0.5756 χ 2χ 1
0.2268 1.9244
=ρ (d)
where
32mL 4
ρ= ρ (e)
π 4 EI
2. Solve the reduced eigenvalue problem.

ρ 1 = 4.0775 ρ2 = 188.87
=Σχ Σ=Σ Σ =Σχ Σ=Σ Σ
1 1
χ1 χ 11 χ2 χ 12
0.0321 −0.3848
3. Determine the approximate frequencies from Eqs. (e) and (18.1.7).
, ,
3.523 EI 23.978 (f)
ω˜ 1 = m ω˜ 2 = 2 m
L2 L
EI
 Sec. 18.2 Formulation Using Virtual Work 733

4. Determine the approximate modes from Eq. (18.1.9).

πx 3π x πx
φ˜ 1 (x) = .1 − cos Σ + 0.0321 .1 − cos Σ = 1.0321 − cos − 0.0321 cos
2L 2L 2L 2L
3π x (g1)

πx 3π x πx 3π x
φ˜ 2 (x) = .1 − cos Σ − 0.3848 .1 − cos Σ = 0.6152 − cos + 0.3848 cos (g2)
2L 2L 2L 2L
These approximate modes are plotted in Fig. E18.1c.
5. Compare with the exact solution. The exact values for natural frequencies and
modes of a cantilever beam were determined in Section 17.3.2. The exact frequencies are
, ,
3.516 EI 22.03 (h)
ω1 = m ω2 =
L2
EI
L2 m
Both approximate frequencies are higher than the corresponding exact values, and the error
is larger in the second frequency. The exact modes are also plotted in Fig. E18.1c. It is clear
that the approximate solution is excellent for the first mode but not as good for the second
mode.
The approximate value for the fundamental frequency in Eq. (h) is lower and hence
better than the result obtained in Example 8.2 by Rayleigh’s method with ψ1(x) as the only
trial function.

18.2 FORMULATION USING VIRTUAL WORK

In this section the equation governing the transverse vibration of a straight beam due to
external forces will be formulated using the principle of virtual displacements. At each
time instant the system is in equilibrium under the action of the external forces p(x, t),
internal resisting bending moments M (x, t), and the fictitious inertia forces, which by
D’Alembert’s principle are
f I (x, t) = −m(x)u¨(x, t) (18.2.1)
If the system in equilibrium is subjected to virtual displacements δu(x), the external virtual
work δW E is equal to the internal virtual work δW I :
δW I = δW E (18.2.2)
Based on the development of Section 8.3.2, these work quantities are
L
∫ ∫ L
δWE =− m(x)u¨(x, t)δu(x) d x p(x, t)δu(x) dx (18.2.3a)
0
0
+
∫
L

δWI = EI (x)u×× (x, t)δ[u××(x)] dx (18.2.3b)

0

The displacements u(x, t) are given by Eq. (18.1.1) and δu(x) is any admissible virtual
displacement:
δu(x) = ψi (x) δz i i = 1, 2, . . . , N (18.2.4)
734 Introduction to the Finite Element Method Chap. 18

Substituting Eqs. (18.1.1) and (18.2.4) in Eq. (18.2.3) leads to

.
N
δW E = −δz i z¨ j m˜ ij + δz i p˜i (t) (18.2.5a)
j =1

.
N

δW I = δz i z j k˜ i j (18.2.5b)
j =1

where

m˜ ij =∫ L
m(x)ψi (x)ψj (x) dx
0
∫L ×× ××
k˜ij = EI (x)ψ i (x)ψj (x) (18.2.6)
dx
0
∫ L
p˜i (t)
p(x, t)ψi (x) dx
= 0

Substituting Eq. (18.2.5) in Eq. (18.2.2) gives

⎛ .j z k˜ ⎞⎠ = δz p˜ (t) (18.2.7)
j ij i i
N =1
.
δzi ⎝ z¨ j m˜ i j N
j
=1
+

and δzi can be dropped from both sides because this equation is valid for any δzi .
Corresponding to the N independent virtual displacements of Eq. (18.2.4), there are
N equations like Eq. (18.2.7). Together they can be expressed in matrix notation:

m˜ z¨ + k˜ z = p˜ (t ) (18.2.8)

where z is the vector of N generalized coordinates, m˜ the generalized mass matrix with
its elements defined by Eq. (18.2.6a), k˜ the generalized stiffness matrix whose
elements are given by Eq. (18.2.6b), and p˜ (t ) the generalized applied force vector with
its elements defined by Eq. (18.2.6c). It is obvious from Eq. (18.2.6) that m˜ and k˜ are
symmetric matri- ces. A damping matrix can also be included in the virtual work
formulation if the damping mechanisms can be defined. The system of coupled
differential equations (18.2.8) can be solved for the unknowns z j (t) using the numerical
procedures presented in Chapter 16. Then at each time instant the displacement u(x) is
determined from Eq. (18.1.1). This is an alternative to the classical modal analysis of the
system mentioned at the end of Sec- tion 18.1.
The mass and stiffness matrices obtained using the principle of virtual
displacements are identical to those derived in Section 18.1 from the principle of energy
conservation, the concept underlying the original development of the Rayleigh–Ritz
method. We will draw upon the virtual work approach when we introduce the finite
element method in Part B of this chapter.
 Sec. 18.4 Finite Element Approximation 735

18.3 DISADVANTAGES OF RAYLEIGH–RITZ METHOD

The Rayleigh–Ritz method leads to natural frequencies ω˜n and natural modes φ˜ n (x) that
approximate the true values ωn and φn (x), respectively, best among the admissible class
of functions described by Eq. (18.1.1). However, the method is not practical for general
appli- cation and automated computer implementation to analyze complex structures for
several reasons: (1) It is difficult to select the Ritz trial functions because they should be
suitable for the particular system and its boundary conditions. (2) It is not clear how to
select addi- tional functions to improve the accuracy of an approximate solution obtained
using fewer functions. (3) It may be difficult to evaluate the integrals of Eq. (18.2.6) over
the entire structure, especially with higher-order trial functions. (4) It is computationally
demanding
to work with the matrices m˜ and k˜ because they are full matrices. (5) It is difficult to in-
terpret the generalized coordinates, as they do not necessarily represent displacements at
physical locations on the structure. These difficulties are overcome by the finite element
method introduced next.

PART B: FINITE ELEMENT METHOD

The finite element method is one of the most important developments in applied
mechanics. Although the method is applicable to a wide range of problems, only an
introduction is included here with reference to systems that can be idealized as an
assemblage of one- dimensional structural finite elements. The presentation, although
self-contained, is based on the presumption that the reader is familiar with the finite
element method for analysis of static structural problems.

18.4 FINITE ELEMENT APPROXIMATION

In the finite element method the trial functions are selected in a special way to overcome
the aforementioned difficulties of the Rayleigh–Ritz method. To illustrate this concept,
consider the cantilever beam shown in Fig. 18.4.1, which is subdivided into a number of
segments, called finite elements. Their size is arbitrary; they may be all of the same size
or all different. The elements are interconnected only at nodes or nodal points. In this
simple case the nodal points are the ends of the element, and each node has two DOFs,
transverse displacement and rotation. In the finite element method nodal displacements
are selected as the generalized coordinates, and the equations of motion are formulated in
terms of these physically meaningful displacements.
The deflection of the beam is expressed in terms of the nodal displacements
through trial functions ψˆ i (x) shown in Fig. 18.4.1. Corresponding to each DOF, a trial
function is selected with the following properties: unit value at the DOF; zero value at all
other DOFs;
continuous function with continuous first derivative. No trial functions are shown for the
736 Introduction to the Finite Element Method Chap. 18

node at the clamped end because the displacement and slope are both zero. These trial
functions satisfy the requirements of admissibility because they are linearly independent,
continuous with continuous first derivative, and consistent with the geometric boundary
conditions. The deflection of the beam is expressed as
.
u(x) = u i ψˆ i (x) (18.4.1)
i

Finite element
u2 u1 u3 u5
4 6 uN-1
u u uN
• • • •
• Nodal point
u1 = 1

ψ^ 1(x)

u2 = 1
ψ^ 2(x)

u3 = 1

ψ^ 3(x)

u4 = 1
ψ^ 4(x)
•
•
•
uN-1 = 1

ψ^ N-
1(x)

uN = 1
ψ^ N(x)

Figure 18.4.1
 Sec. 18.5 Analysis Procedure 737

where u i is the nodal displacement in the i th DOF and ψˆ i (x) is the associated trial func-
tion. Because these functions define the displacements between nodal points (in contrast
to global displacements of the structure in the Rayleigh–Ritz method), they are called
interpolation functions.
The finite element method offers several important advantages over the Rayleigh–
Ritz method. Stated in the same sequence as the disadvantages of the Rayleigh–Ritz
method mentioned at the end of Section 18.3, the advantages of the finite element
method are: (1) Simple interpolation functions can be chosen for each finite element. (2)
Accu- racy of the solution can be improved by increasing the number of finite elements
in the structural idealization. (3) Computation of the integrals of Eq. (18.2.6) is much
easier be- cause the interpolation functions are simple, and the same functions may be
chosen for each finite element. (4) The structural stiffness and mass matrices developed
by the finite element method are narrowly banded, a property that reduces the
computational effort nec- essary to solve the equations of motion. (5) The generalized
displacements are physically meaningful, as they give the nodal displacements directly.

18.5 ANALYSIS PROCEDURE

The formulation of the equations of motion for a structure by the finite element method
may be summarized as a sequence of the following steps:

1. Idealize the structure as an assemblage of finite elements interconnected only at

nodes (Fig. 18.5.1a); define the DOF u at these nodes (Fig. 18.5.1b).
2. For each finite element form the element stiffness matrix ke, the element mass
matrix me, and the element (applied) force vector pe (t) with reference to the DOF
for the element (Fig. 18.5.1c). For each element the force–displacement relation
and the

5
6
4 4
8 2 5
Node b c
5 9 3 6
3 7 1 4 4
Finite element
2 b
a d 3 
5 1

1 2

(a)
(b) (c)

Figure 18.5.1 (a) Finite elements and nodes; (b) assemblage DOF u; (c) element DOF ue.
738 Introduction to the Finite Element Method Chap. 18

inertia force–acceleration relation are

(f S )e = ke ue (f I )e = me u¨ e (18.5.1)

In the finite element formulation these relations are obtained by assuming the dis-
placement field over the element, expressed in terms of nodal displacements.
3. Form the transformation matrix ae that relates the displacements ue and forces pe
for the element to the displacements u and forces p for the finite element
assemblage:

ue = a e u p(t) = aTe pe (t) (18.5.2)

where ae is a Boolean matrix consisting of zeros and ones. It simply locates the
elements of ke, me, and pe at the proper locations in the mass matrix, stiffness matrix,
and (applied) force vector for the finite element assemblage. Therefore, it is not
necessary to carry out the transformations: kˆ e = aeT ke ae , mˆ e = eaT me ae , or pˆ e (t) =
aTe pe(t) to transform the element stiffness and mass matrices and applied force vector
to the nodal displacements for the assemblage.
4. Assemble the element matrices to determine the stiffness and mass matrices and
the applied force vector for the assemblage of finite elements:

k
Ne
= Ae N
=
e=1
k m Ne m
e=1
e p(t) = A e pe (t) (18.5.3)
A e=1
The operator Adenotes the direct assembly procedure for assembling according to
the matrix ae, the element stiffness matrix, element mass matrix, and the element
force vector—for each element e 1 to Ne, where Ne is the number of elements—
=
into the assemblage stiffness matrix, assemblage mass matrix, and assemblage
force vector, respectively.
5. Formulate the equations of motion for the finite element assemblage:

mu¨ + cu˙ + ku = p(t) (18.5.4)

where the damping matrix c is established by the methods of Chapter 11.

The governing equations (18.5.4) for a finite element system are of the same form
as formulated in Chapter 9 for frame structures. It should be clear from the outline above
that the only difference between the displacement method for analysis of frame structures
and the finite element method is in the formulation of the element mass and stiffness
matri- ces. Therefore, Eq. (18.5.4) can be solved for u(t) by the methods developed in
preceding chapters. The classical modal analysis procedure of Chapters 12 and 13 is
applicable if the system has classical damping and the direct methods of Chapter 16
enable analysis of nonclassically damped systems.
In the subsequent sections, which are restricted to assemblages of one-dimensional
finite elements, we define the interpolation functions and develop the element stiffness
matrix ke, the element mass matrix me, and the element (applied) force vector pe(t). As-
sembly of these element matrices to construct the corresponding matrices for the finite
element assemblage is illustrated by an example.