Unusual Closed-Form Solutions For Beam Vibrations

EOIS: 2892_chap04 2004/9/27 22:56 page 135 #1
4
Unusual Closed-Form Solutions for
Beam Vibrations
This chapter discusses the vibration of inhomogeneous beams under various
boundary conditions. Associated aspects of stochastic analysis are also elucid-
ated. Specically, an innite number of closed-form solutions is reported for
both the free vibrations and reliabilities of vibrating beams. It is demonstrated
that the stochastic beams may share deterministic vibration frequencies.
4.1 Apparently First Closed-Form Solutions for Frequencies of
Deterministically and/or Stochastically Inhomogeneous
Beams (PinnedPinned Boundary Conditions)
4.1.1 Introductory Remarks
The aim of this section is to nd some closed-form solutions to the dynamic
equation of a beam in which both the Youngs modulus and the density are
polynomial functions, with both the deterministic and stochastic inhomo-
geneities included. We look for the exact mode shape also as a polynomial
function, with an attendant closed-form expression for the natural frequen-
cies. The case considered is that of the beam pinned at both ends. For the
bibliography of investigations on vibration and buckling of inhomogeneous
beams, one may consult the papers by Eisenberger (1997a,b), and Rollot and
Elishakoff (1999).
The importance of the solutions found lies in the possibility of their use
as benchmark solutions against which the efcacy of various approximate
methods could be ascertained. Additionally, presently there is a consid-
erable literature on the so-called stochastic nite element method (SFEM),
which deals with inhomogeneous structures involving random elds. The
latter random functions can be represented as mean functions superimposed
with deviation functions. The solution of the problem with properly chosen
mean functions often constitutes an important part of the analysis (see e.g.
tions, bothindeterministic andstochastic settings possess attractiveanalytical
135
2005 by Issac Elishakoff
Kylog lu et al., 1994; Elishakoff et al., 1995). Thus, the closed-form solu-
EOIS: 2892_chap04 2004/9/27 22:56 page 136 #2
136 Eigenvalues of Inhomogeneous Structures
advantages over approximate solutions, where inherent approximations of
various natures are needed. For alternative formulations of the randomeigen-
value problem, the reader may consult papers by Shinozuka andAstill (1972)
and Zhu and Wu (1991).
4.1.2 Formulation of the Problem
The dynamic behavior of a beam, is described by the following equation
d
2
dx
2
_
D(x)
d
2
w(x)
dx
2
_
R(x)
2
w(x) = 0 (4.1)
where D(x) = EI is the exural rigidity, E the Youngs modulus, the dens-
ity, I the moment of inertia of the cross section, R(x) = A is the inertial
coefcient, A the area of the cross section, w(x) the displacement and the
natural frequency.
In this section, it is assumed that the cross-sectional area is constant, but
both and P are specied as polynomial functions, given by
R() =
m
i=0
a
i
i
(4.2)
D() =
n
i=0
b
i
i
(4.3)
where = x/L is a non-dimensional axial coordinate.
We assume that w() is also polynomial:
w() =
p
i=0
w
i
i
(4.4)
where w
i
are the sought for coefcients. In these expressions, m, n and p are,
respectively, the degree of the polynomials for R(), D() and w().
Equation (4.1) can be re-written as
d
2
d
2
_
D()
d
2
w()
d
2
_
kL
4
R()w() = 0 (4.5)
where
k =
2
(4.6)
EOIS: 2892_chap04 2004/9/27 22:56 page 137 #3
Unusual Closed-Form Solutions for Beam Vibrations 137
As the involved functions are assumed to be polynomial ones, the degrees of
each polynomial function must be linked, namely
n +(p 2) 2 = m+p (4.7)
or, simply
n m = 4 (4.8)
We observe that Eq. (4.8) does not dependon the degree pof the displacement
w(). We arrive at the seemingly unexpected conclusion that any polynomial
function for the displacement may be used in Eq. (4.5) if it also satises the
boundary conditions. This fact will be used at a later stage. In viewof Eq. (4.8)
the expression for D() can be written as follows:
D() =
m+4
i=0
b
i
i
(4.9)
4.1.3 Boundary Conditions
The case of the pinned beam is associated with the following boundary
conditions:
w(0) = 0 (4.10)
D(0)w
(0) = 0 (4.11)
w(1) = 0 (4.12)
D(1)w
(1) = 0 (4.13)
The solution to Eq. (4.11) can be found with either D(0) = 0 or w
(0) = 0.
However, the Youngs modulus being zero at one point has no physical sense;
thus, Eq. (4.11) is equivalent to w
(0) = 0. Hence, we postulate that b

0
> 0.
The same reasoning can be applied to Eq. (4.13). So, the displacement has to
satisfy the following conditions:
w(0) = 0 (4.14)
w
(0) = 0 (4.15)
w(1) = 0 (4.16)
w
(1) = 0 (4.17)
Satisfaction of the boundary conditions (4.144.17) requires that the degree
of the displacement polynomial must be at least 4. Assuming that w() is a
EOIS: 2892_chap04 2004/9/27 22:56 page 138 #4
fourth-order polynomial
w() = w
0
+w
1
+w
2
2
+w
3
3
+w
4
4
(4.18)
The satisfaction of the boundary conditions yields
w() = w
1
( 2
3
+
4
) (4.19)
4.1.4 Expansion of the Differential Equation
By substituting the different expressions for D(), R() and w() in Eq. (4.5),
we obtain
w
1
_
m+4
i=2
i(i 1)b
i
i2
(12 +12
2
) +
m+4
i=0
24b
i
i
+2
m+4
i=1
ib
i
i1
(12 +24)
kL
4
m
i=0
a
i
i
( 2
3
+
4
)
_
= 0 (4.20)
The latter expression can be re-written as follows:
12
m+3
i=1
i(i +1)b
i+1
i
+12
m+4
i=2
i(i 1)b
i
i
+24
m+4
i=0
b
i
i
24
m+3
i=0
(i +1)b
i+1
i
+48
m+4
i=1
ib
i
i
kL
4
m+1
i=1
a
i1
i
+2kL
4
m+3
i=3
a
i3
i
kL
4
m+4
i=4
a
i4
i
= 0
(4.21)
Equation (4.21) has to be satised for any . This requirement yields the
following relations:
24(b
1
b
0
) = 0 for i = 0 (4.22)
kL
4
a
0
+72(b
1
b
2
) = 0 for i = 1 (4.23)
kL
4
a
1
+144(b
2
b
3
) = 0 for i = 2 (4.24)
L
4
(2ka
0
ka
2
) +240(b
3
b
4
) = 0 for i = 3 (4.25)
. . .
EOIS: 2892_chap04 2004/9/27 22:56 page 139 #5
L
4
(2ka
i3
ka
i4
ka
i1
) +12(i +1)(i +2)(b
i
b
i+1
) = 0 for 4 i m+1
(4.26)
. . .
L
4
(2ka
m1
ka
m2
) +12(m+3)(m+4)(b
m+2
b
m+3
) = 0 for i = m+2
(4.27)
L
4
(2ka
m
ka
m1
) +12(m+4)(m+5)(b
m+3
b
m+4
) = 0 for i = m+3
(4.28)
kL
4
a
m
+12(m
2
+11m+30)b
m+4
= 0 for i = m+4
(4.29)
Note that Eqs. (4.22)(4.29) ar e valid only if m3. For cases that satisfy
the inequality m < 3, the reader is referred to Appendix A. Note also that
Eqs. (4.22)(4.29) have a recursive form.
The sole unknown in Eqs. (4.22)(4.29) is the natural frequency coefcient k;
yet, we observe that we have m+5 equations. We conclude that the parameters
b
i
and a
i
have to satisfy some auxiliary conditions so that Eqs. (4.22)(4.29)
are compatible.
4.1.5 Compatibility Conditions
A rst compatibility condition is given by Eq. (4.22), leading to b
0
= b
1
.
From the other equations, several expressions for k can be found. Its values
determined from Eqs. (4.22)(4.29) respectively, are listed below:
k = 72(b
1
b
2
)/L
4
a
0
(4.30)
k = 144(b
2
b
3
)/L
4
a
1
(4.31)
k = 240(a
2
2a
0
)
1
(b
3
b
4
)L
4
(4.32)
. . .
k = 12(i +1)(i +2)(a
i1
+a
i4
2a
i3
)
1
[b
i
b
i+1
]/L
4
for 4 i m+1 (4.33)
. . .
k = 12(m+3)(m+4)(a
m2
2a
m1
)
1
[b
m+2
b
m+3
]/L
4
(4.34)
k = 12(m+4)(m+5)(a
m1
2a
m
)
1
[b
m+3
b
m+4
]/L
4
(4.35)
k = 12(m
2
+11m+30)b
m+4
/L
4
a
m
(4.36)
To check the compatibility of these expressions, all expressions for k have to
be equal to each other. We consider two separate problems: (i) the material
density coefcients a
i
are specied; nd coefcients b
i
so that the closed-form
EOIS: 2892_chap04 2004/9/27 22:56 page 140 #6
solution holds; (ii) the elastic modulus coefcients b
i
are specied; nd
coefcients a
i
so that the closed-form solution is obtainable.
4.1.6 Specied Inertial Coefcient Function
Let us assume that the function R(), of the inertial coefcient, and hence all
a
i
(i = 0, 2, . . . , m) are given. Let us observe that if b
m+4
is specied then the
expression given in Eq. (4.36) is the nal formula for the natural frequency
coefcient k. Then, Eqs. (4.30)(4.36) allow an evaluation of the remaining
parameters b
i
. Note that b
m+4
and a
m
have to have the same sign due to the
positivity of k.
From Eq. (4.35) we get
b
m+3
=
__
m
2
+11m+30
(m+4)(m+5)
_
_
a
m1
a
m
1
_
+1
_
b
m+4
(4.37)
Equation (4.34) yields
b
m+2
=
__
m+5
m+3
_
a
m2
2a
m1
a
m1
2a
m
+1
_
b
m+3

_
m+5
m+3
_
a
m2
2a
m1
a
m1
2a
m
b
m+4
(4.38)
Equation (4.33) results in
b
i
=
__
i +3
i +1
_
a
i1
a
i4
2a
i3
a
i
a
i3
2a
i2
+1
_
b
i+1

_
i +3
i +1
_
a
i1
a
i4
2a
i3
a
i
a
i3
2a
i2
b
i+2
(4.39)
where i belongs to the set {4, 5, . . . , m+1}.
From Eq. (4.32) we obtain
b
3
=
__
3
2
_
a
2
2a
0
a
3
a
0
2a
1
+1
_
b
4

_
3
2
_
a
2
2a
0
a
3
a
0
2a
1
b
5
(4.40)
Equation (4.31) leads to
b
2
=
__
5
3
_
a
1
a
2
2a
0
+1
_
b
3

_
5
3
_
a
1
a
2
2a
0
b
4
(4.41)
From Eq. (4.29) we get
b
1
=
_
2
a
0
a
1
+1
_
b
2
2
a
0
a
1
b
3
(4.42)
EOIS: 2892_chap04 2004/9/27 22:56 page 141 #7
And, nally, Eq. (4.22) yields
b
0
= b
1
(4.43)
Thus, for specied coefcients a
0
, a
1
, . . . , a
m
and b
m+4
, Eqs. (4.37)(4.43) lead
to the set of coefcients of the elastic modulus such that the beam possesses
the mode shape given in Eq. (4.19). Note that if a
i
= a, then the coefcients b
i
do not depend on the parameter a.
To sum up, if
R() =
m
i=0
a
i
i
D() =
m+4
i=0
b
i
i
(4.44)
where b
i
are computed via Eqs. (4.37)(4.43), the fundamental mode shape of
a beam is
w() = w
1
( 2
3
+
4
) (4.45)
and the fundamental natural frequency squared reads
2
= 12(m
2
+11m+30)b
m+4
/a
m
L
4
(4.46)
As we have seen, in order to obtain the closed-from solution it is sufcient
that (1) all coefcients a
i
and (2) the coefcient b
m+4
be specied. Yet, the
requirements are not necessary ones. Indeed, one can assume that the coef-
cients a
i
are given and, instead of b
m+4
, any coefcient b
j
(j = m + 4) is
specied. If this is the case, then from Eq. (4.33) one expresses b
i+1
via b
i
and
k; substitution into subsequent equations allows us to express b
m+2
, b
m+3
,
b
m+4
via b
i
; analogously, substitution of b
i
into Eqs. (4.30)(4.32) yields the
sought for exact solutions.
ant fundamental natural frequency coefcients are given. The polynomial
functions R
j
() were specied as
R
1
() =
m
i=0
i
R
2
() =
m
i=0
(i +1)
i
(4.47)
respectively, in Tables 4.1 and 4.2.
4.1.7 Specied Flexural Rigidity Function
Consider, now, the case whenthe exural rigidityfunctionis specied, imply-
ing that all b
i
(i = 0, . . . , m+4) are given. The following question arises: is it
possible to determine the material density coefcients a
i
(i = 0, . . . , m), such
that the equations corresponding to Eqs. (4.22)(4.29) are compatible? One
In Tables 4.1 and 4.2, some sample specied function D(x) and the attend-
EOIS: 2892_chap04 2004/9/27 22:56 page 142 #8
TABLE 4.1
Flexural Rigidities and Eigenvalues Corresponding to Different Values of m (R
1
()
Given in Eq. (4.47.1))
m D
j
() KL
4
0 b(3 +3 2
2
2
3
+
4
) 360b
1 (b/10)(59 +59 11
2
46
3
4
4
+10
5
) 504b
2 (b/15)(135 +135 5
2
75
3
33
4
5
5
+15
6
) 672b
3 (b/35)(434 +434 +14
2
196
3
70
4
70
5
10
6
+35
7
) 864b
4 (b/28)(452 +452 +32
2
178
3
52
4
52
5
52
6
7
7
+28
8
) 1,080b
5 (b/36)(648 +648 +69
2
224
3
56
4
56
5
56
6
56
7
8
8
+36
9
) 1,320b
6 (b/15)(371 +371 +41
2
124
3
25
4
25
5
25
6
25
7
25
8
3
9
+15
10
) 1,584b
7 (b/55)(1,628 +1,628 +198
2
517
3
88
4
88
5
88
6
88
7
88
8
88
9
10
10
+55
11
)
1,872b
8 (b/330)(5,715 +57,151 +1,492
2
3,513
3
510
4
510
5
510
6
510
7
510
8
510
9
510
10
55
11
+330
12
)
2,184b
9 (b/26)(1,053 +1,053 +143
2
312
3
39
4
39
5
39
6
39
7
39
8
39
9
39
10
39
11
4
12
+26
13
)
2,520b
TABLE 4.2
Flexural Rigidities and Eigenvalues Corresponding to Different Values of m (R
1
()
Given in Eq. (4.47.2))
m D
j
() KL
4
0 b(3 +3 2
2
2
3
+
4
) 360b
1 (b/10)(38 +38 +3
2
32
3
11
4
+10
5
) 252b
2 (b/45)(203 +203 +63
2
77
3
119
4
35
5
+45
6
) 224b
3 (b/140)(725 +725 +305
2
115
3
241
4
325
5
85
6
+140
7
) 216b
4 (b/140)(815 +815 +395
2
25
3
151
4
235
5
295
6
70
7
+140
8
)
216b
5 (b/1,512)(9,751 +9,751 +5,131
2
+511
3
875
4
1,799
5
2,549
6
2,954
7
644
8
+1,512
9
)
220b
6 (b/1,470)(10,388 +10,388 +5,768
2
+1,148
3
238
4
1,162
5
1,822
6
1,617
7
2,702
8
546
9
+1,470
10
)
1584
7
b
immediately observes that there are (m + 5) equations (4.22)(4.29), while
one has only m + 1 unknowns, a
0
, a
1
, . . . , a
m
. In actuality, however, one has
only munknowns. In order for the process of determining the coefcient a
i
to proceed, one of the coefcients a
j
should be specied. The most conveni-
ent assumption is to x either a
0
or a
1
or a
m
, since in these cases only one
equation, respectively, Eq. (4.30) or Eq. (4.31) or Eq. (4.36) will be sufcient to
determine the sought for expression for the natural frequency coefcient. Let
us assume that the coefcient a
0
is given; thus, to check the compatibility of
Eqs. (4.22)(4.29), four b
i
coefcients cannot be chosen arbitrarily.
Note that the natural frequency coefcient k has to be positive; thus, the
difference b
1
b
2
and the coefcient a
0
have to have the same sign. Moreover,
EOIS: 2892_chap04 2004/9/27 22:56 page 143 #9
as the coefcient a
0
is positive, the difference b
1
b
2
should be positive. So,
for b
1
> b
2
, one substitutes the value of k determined from Eq. (4.30) into
Eq. (4.31); this allows as to determine the coefcient a
1
so that the frequency
coefcient k in Eq. (4.31) is positive, and so on.
First, Eq. (4.22) leads to
b
0
= b
1
(4.48)
From Eq. (4.31), we get
a
1
= 2a
0
b
3
b
2
b
2
b
1
(4.49)
Then, Eq. (4.32) yields
a
2
=
5a
1
(b
4
b
3
) +6a
0
(b
3
b
2
)
3(b
3
b
2
)
(4.50)
Equation (4.33), where i = 4, results in
a
3
=
a
0
(4b
4
6b
5
+2b
3
) +a
1
(4b
4
4b
3
) +a
2
(3b
5
3b
4
)
2(b
4
b
3
)
(4.51)
From Eq. (4.33), where 4 i m, we obtain
a
i
=
1
(i +1)(b
i+1
b
i
)
{a
i1
(i +3)(b
i+2
b
i+1
) +2a
i2
(i +1)(b
i+1
b
i
)
+a
i3
[b
i+1
(i +5) 2b
i+2
(i +3) +b
i
(i +1)] +a
i4
(i +3)(b
i+2
b
i+1
)}
(4.52)
Then, from Eqs. (4.36) and (4.30), one can nd an expression for b
m
+ 4, so
that, the compatibility of Eqs. (4.22)(4.29) can be checked:
b
m+4
=
6a
m
(b
1
b
2
)
a
0
(m
2
+11m+30)
(4.53)
From Eqs. (4.35) and (4.36), a relation for b
m+3
can be found
b
m+3
=
a
0
b
m+4
(m
2
+9m+20) +6a
m1
(b
1
b
2
) +12a
m
(b
2
b
1
)
a
0
(m+4)(m+5)
(4.54)
Finally, Eqs. (4.34) and (4.35) yield an evaluation of b
m+2
b
m+2
=
a
0
b
m+3
(m+3)(m+4) +6a
m2
(b
1
b
2
) +12a
m1
(b
2
b
1
)
a
0
(m+3)(m+4)
(4.55)
EOIS: 2892_chap04 2004/9/27 22:56 page 144 #10
To summarize, while specifying the elastic modulus function, only m + 1
coefcients b
i
can be chosen arbitrarily; the remaining four coefcients are
connected with the arbitrary ones via Eq. (4.48) and Eqs. (4.53)(4.55).
Thus, if
R() =
m
i=0
a
i
i
D() =
m+4
i=0
b
i
i
(4.56)
.
the fundamental mode shape of a beam is
w() = w
1
( 2
3
+
4
) (4.57)
The fundamental natural frequency squared reads as
2
= 72(b
1
b
2
)/a
0
L
4
(4.58)
The closed-formsolutions couldbe utilizedfor comparisonwithapproximate
techniques. For example, utilization of the single term BoobnovGalerkin
method for the case
R() = 1 +2 +3
2
+4
3
+5
4
(4.59)
D() = b
8
L
8
_
163
28
+
163
28
+
79
28
5
28
151
140
47
28
59
28
1
2
7
+
8
_
(4.60)
yields, with sin() taken as a comparison function, the following expression
k =
b
8
L
4
22,050
_
6,945,750 409,185
4
+391,612
8
2,716,875
2
+17,356
6
10
4
19
2
+15
_
(4.61)

k =
4.1.8 Stochastic Analysis
The preceding formulation allows one to perform a stochastic analysis to
account for the possible randomness in the material density and elastic
modulus.
where a
i
and four of the coefcients b
i
are computed via Eqs. (4.48)(4.55),
or numerically k = 216.29697b
8
L
4
, which differs from the exact solution k =
216b
8
L
4
by 0.13%.
EOIS: 2892_chap04 2004/9/27 22:56 page 145 #11
4.1.8.1 Probabilistically specied inertial coefcient function
Assume that the coefcients a
i
form a random vector with a joint probability
density f
A
(a
1
, a
2
, . . . , a
m
), where A
T
= (A
1
, A
2
, . . . , Am) and capital letters
denote a randomvariable A
i
whose possible values are denoted by the lower-
case letters a
i
. As Eq. (4.36) suggests, the natural frequency squared,
2
, is
also a random variable denoted by
2
=
2
B
m+4
/A
m
2
= 12(m
2
+11m+30)
(4.62)
where the coefcient B
m+4
constitutes either a deterministic or a randomvari-
able. Several cases allow a closed-form evaluation of the reliability r, dened
in the present circumstances as the probability that the natural frequency
squared
2
does not exceed a pre-selected deterministic value
2
0
:
r = Prob(
2

2
0
) = Prob(
2
B
m+4

2
0
A
m
) (4.63)
Let B
m+4
B be an exponentially distributed random variable with density
f
B
(b) =
1
E(B)
exp
_
b
E(B)
_
b 0 (4.64)
and zero otherwise, E(B) being the mean value of B. Likewise, the coefcient
A
m
A has an exponential density
f
A
(a) =
1
E(A)
exp
_
a
E(A)
_
a 0 (4.65)
and vanishes if a < 0, with E(A) indicating the mean value of A. Since A is
exponentially distributed, the randomvariable
2
0
Ais also exponentially dis-
tributed with mean
2
0
E(A). Likewise
2
B is an exponential random variable
with mean
2
E(B). The reliability is obtained as
r =

2
0
E(A)
2
0
E(A) +
2
E(B)
(4.66)
It is remarkable that although all coefcients A
j
(j = 1, . . . , m) are random,
the probabilistic characterization of only a single coefcient A
m
turns out to
be needed, in addition to that of B
m+4
for the reliability evaluation.
4.1.8.2 Specied exural rigidity function
Assume now that the m+1 coefcients b
i
form a random vector with a joint
probability density f
B
(b
1
, . . . , b
m+1
). The remaining four coefcients b
0
, b
m+2
,
EOIS: 2892_chap04 2004/9/27 22:56 page 146 #12
b
m+3
, b
m+4
are relatedtothe above coefcients via Eqs. (4.48) and(4.53)(4.55).
Due to the randomness of coefcients B
1
, . . . , B
m+1
, we conclude that the
natural frequency squared is itself a random variable
2
=
2
(B
1
B
2
) (4.67)
where
2
is a coefcient
2
= 72/L
4
a
0
(4.68)
The coefcient a
0
in Eq. (4.68) can be treated either

as a deterministic or a ran-
domvariable. For the sake of illustration, a particular case will be considered
now, namely, the case when a
0
is a deterministic variable. The mean natural
frequency squared equals
E[
2
] =
2
[E(B
1
) E(B
2
)] (4.69)
whereas its variance reads as
Var[
2
] =
4
[Var(B
1
) +Var(B
2
)] (4.70)
As in Eq. (4.63) reliability is denedas a probability that the natural frequency
squared,
2
, does not exceed a pre-selected value
2
0
. Once the joint probabil-
ity density of the coefcients b
1
, b
2
anda
0
are specied, the reliability function
r can be derived directly. The reliability is cast as
r = Prob[
2
(B
1
B
2
) <
2
0
] (4.71)
Anew random variable is introduced:
Z =
2
(B
1
B
2
)
2
0
(4.72)
The reliability is re-written as
r = Prob(Z < 0) =
_

db
2
_
_
b
2
+
2
0
/
2
f
B
2
B
1
(b
2
, b
1
) db
1
_
b
2
f
B
2
B
1
(b
2
, b
1
) db
1
_
(4.73)
EOIS: 2892_chap04 2004/9/27 22:56 page 147 #13
Let B
1
and B
2
be independent random variables; then, Eq (4.73) becomes
r =
_

f
B
2
(b
2
)
_
F
B
1
_
b
2
+

2
0
2
_
F
B
1
(b
2
)
_
db
2
(4.74)
F
B
1
_
b
2
+

2
0
2
_
=
_
b
2
+
2
0
/
2
f
B
1
(b
1
) db
1
(4.75)
F
B
1
(b
2
) =
_
b
2
f
B
1
(b
1
) db
1
Let B
1
be a uniformly distributed random variable with density
f
B
1
(b
1
) = ( )
1
if b
1
[, ] (4.76)
and zero otherwise; likewise, the coefcient B
2
has a uniform density
f
B
2
(b
2
) = ( )
1
if b
2
[, ] (4.77)
Let us assume that > . Thus, the positivity of
2
is always assured since
B
1
> B
2
. We rst calculate the expressions in Eq. (4.75):
F
B
1
_
b
2
+

2
0
2
_
=
_
_
0 if b
2
+

2
0
2

b
2

_

2
0
2
_

if < b
2
+

2
0
2
<
1 if b
2
+

2
0
2

(4.78)
F
B
1
(b
2
) = 0
From Eq. (4.74) it follows that for the reliability evaluation we need to nd a
region in which both f
B
2
and F
B
1
are non-zero. As Eq. (4.77) suggests f
B
2
is
non-zero if b
2
. The function F
B
1
differs from both zero and unity if

2
0
/
2
< b
2
<
2
0
/
2
; F
B
1
equals unity if b
2
>
2
0
/
2
. Thus, in
order for the product f
B
2
F
B
1
to be non-zero it is necessary and sufcient that
b
2
belongs to the two following intervals:
I
1
= [, ] I
2
= [
2
0
/
2
, ] (4.79)
It is natural to inquire when these two intervals have no intersection. This
takes place when the lower end of interval I
2
exceeds theupper- endof interval
I
1
, i.e., when
2
0
/
2
, or in terms of
2
0
, when
2
0

2
( ) (see
the reliability r both vanish identically.
Figure 4.1). Under these circumstances, the integrand in Eq. (4.69), and hence
EOIS: 2892_chap04 2004/9/27 22:56 page 148 #14
0
1/( )
1/( )
f
B
2
(b
2
)
f
B
1
(b
2
)

0
2
/
2
F
B
1
(b
2
+
0
2
/
2
)

0
2
/
2
b
2
FIGURE 4.1
Probability density functions of variables B
1
and B
2
, and the probability distribution function
F
B
1
for
2
0

2
( ), leading to zero reliability
0
1/( )
1/( )
f
B
2
(b
2
)
f
B
1
(b
2
)

0
2
/
2
F
B
1
(b
2
+
0
2
/
2
)

0
2
/
2
b
2
FIGURE 4.2
1
and B
2
F
B
1
for
2
0
/
2
; the reliability is given by Eq. (4.80)
0
1/( )
1/( )
f
B
2
(b
2
) f
B
1
(b
2
)

0
2
/
2
F
B
1
(b
2
+
0
2
/
2
)

0
2
/
2
b
2
FIGURE 4.3
1
and B
2
F
B
1
for
2
0
/
2
and
2
0
/
2
; the reliability is given by Eq. (4.81)
Assume now that the lower end of the interval I
2
belongs to the interval I
1
,
i.e.
2
0
/
2
, and, moreover,
2
0
/
2
(Figure 4.2).
This implies that
2
( )
<
2
0

2
( ). The sought for region for b
2
is
the interval [
2
0
/
2
, ]. The reliability is obtained as
r =
2
0
/
2
b
2
(
2
0
/
2
( )( )
db
2
=
[
2
( ) +
2
0
]
2
4
( )( )
(4.80)
Now, consider the following case:
2
0
/
2
and
2
0
/
2

(Figure 4.3) implying
2
( ) <
2
0

2
( ).
The b
2
region is the interval [, ]. But to evaluate the reliability, this region
has to be split into the union of two regions [,
2
0
/
2
], [
2
0
/
2
, ] since
EOIS: 2892_chap04 2004/9/27 22:56 page 149 #15
0
1/( )
1/( )
f
B
2
(b
2
)
f
B
1
(b
2
)
0
2
/
2
F
B
1
(b
2
+
0
2
/
2
)

0
2
/
2
b
2
FIGURE 4.4
1
and B
2
F
B
1
for
2
0
/
2
, leading to unit reliability
the function F
B
1
(b
2
+
2
0
/
2
) takes the value unity at b
2
=
2
0
/
2
. Hence,
the reliability reads as
r =
_

2
0
/
2
r
b
2
(
2
0
/
2
)
( )( )
db
2
+
_

2
0
/
2
1

db
2
=
(
2

2
0
) +2
4
( ) +
4
(
2
2
2
+2
2
0
)
2
4
( )( )
(4.81)
We now consider the case 2
0
/
2
(Figure 4.4), meaning
2
() <
2
0
.
The integration domain is the interval [, ]:
r =
_

1
1

db
2
= 1 (4.82)
of the interval b
1
is

smaller

than

the

length

of

the

interval

b
2
,

then
the two quantities,
2
0
/
2
and
2
0
/
2
belong to the interval b
2
as shown
r =
_

2
0
/
2
2
0
/
2
b
2

_

2
0
/
2
_
( )( )
db
2
+
_

2
0
/
2
1

db
2
=

2
(2 ) +2
2
0
2
2
( )
(4.83)
If, however, the length of the interval b
1
is larger than the length
2
reliability is expressed as follows:
r =
_

r
b
2

_

2
0
/
2
_
( )( )
db
2
=

2
2
2( )( )
+

2
0

2
2
( )
(4.84)
There are two intermediate situations between the ones depicted in Figures 4.2
and 4.3, depending on the lengths of the intervals and . If the length
in Figure 4.5. The reliability is given by
of the interval b , the sought for region is as illustrated in Figure 4.6, and the
EOIS: 2892_chap04 2004/9/27 22:56 page 150 #16
0
1/( )
1/( )
f
B
2
(b
2
)
f
B
1
(b
2
)

0
2
/
2
F
B
1
(b
2
+
0
2
/
2
)

0
2
/
2
b
2
FIGURE 4.5
Case when the length of the interval B
1
is smaller than the length of the interval B2; the
reliability is given in Eq. (4.83)
0
1/( )
1/( )
f
B
2
(b
2
) f
B
1
(b
2
)

0
2
/
2
F
B
1
(b
2
+
0
2
/
2
)

0
2
/
2
b
2
FIGURE 4.6
Case when the length of the interval B
1
is larger than the length of the interval B
2
; the
reliability is given in Eq. (4.84)
The coefcients of variations c
1
and c
2
of the random variables B
1
and B
2
,
respectively:
c
1
=
Var(B
1
)
E(B
1
)
c
2
=
Var(B
2
)
E(B
2
)
(4.85)
are chosen to be equal c
1
= c
2
=c. This implies that for a specied
coefcient of variation c, the upper bounds of the interval are related to the
lower bounds, as follows
=
1 +
3c
1
3c
=
1 +
3c
1
3c
(4.86)
is xed at 14, is xed at 1. For the coefcient of variation 0.3, the reliability
of unity is manifested for y 43.286. Values associated with transition to
unit reliability for the coefcients of variation of 0.4 and 0.5 are, respectively,
y 76.151 and y 193.994.
It should be remarked that the transitional values of y from non-unit to
unit reliability can be predicted for the uniformly distributed B
1
and B
2
,
without resort to the reliability calculations. As the natural frequency squared
is proportional to the difference between B
1
and B
2
in Eq. (4.67), the largest
value of the natural frequency is obtained when B
1
takes on the value of the
EOIS: 2892_chap04 2004/9/27 22:56 page 151 #17
upper bound of the interval, , whereas B
2
takes on the value of the lower
bound of the interval, . Thus, if the maximum natural frequency squared
2
max
=
2
( ) is smaller than y =
2
0
, then the inequality in Eq. (4.71) is
satisedautomatically; hence, the reliability is identically unity. Likewise, the
minimum natural frequency squared
2
min
=
2
( ). If this value exceeds
y =
2
0
, then obviously for any pairs of values in the intervals B
1
and B
2
, the
inequality (4.71) will be violated, with resulting vanishing reliability.
On the other hand, for exponentially distributed variables, unit reliability
is never achieved, as Eq. (4.66) suggests.
4.1.9 Nature of Imposed Restrictions
In this chapter, in order to obtain the closed-form solutions for natur al
frequencies deterministically and/or stochastically for an inhomogeneous
pinned beam, the exural rigidity and the inertial coefcient were assumed to
be polynomial functions whose powers differ by four. One should stress that
the a andb coefcients inEqs. (4.2) and(4.3) cannot be speciedindependently
in order for a closed-form solution to exist.
One should stress that the importance of the derived closed-form solution
is not diminished by the fact that certain conditions must be met. The appear-
ance of conditions is natural too. Indeed, it can be expected that the solution
of the inverse problem would depend up on either part of or the entire given
data. Thus, if the inhomogeneous beam has apolynomial inertial coefcient
with given coefcients, it must be no surprise that the sought for exural
rigidity of the beam possesses the pre-selected mode shape, which is directly
related to the specied inertial coefcients, in order to derive the closed-form
solutions.
The following question arises: is there any resemblance in earlier literature
to the type of thinking adopted in this chapter? The connection with previous
work was found via Saint-Venants semi-inverse method.
4.1.10 Concluding Remarks
The class of deterministic andstochastic solutions describedcontains aninn-
ite number of closed-fom solutions. Indeed, the degree mof the polynomial
in the expression for the inertial coefcient in Eq. (4.44) can be chosen arbitrar-
ily. Likewise, the coefcients a
i
can be prescribed at will subject to a condition
of positivity of both P() and ().
It should be noted that there is a connection between the present work
and the subject of inverse problems of vibration (Gladwell, 1986ac, 1996).
Indeed, whereas mathematical direct problems consist of nding solutions
to equations with known input parameters, mathematical inverse problems
deal with the reconstruction of the parameters of the governing equations
whenthe output quantities are known. According to Gladwell (1986), inverse
EOIS: 2892_chap04 2004/9/27 22:56 page 152 #18
problems are concerned with the construction of a model of a given type;
e.g. a mass-spring system, a string, etc; which has given eigenvalues and/or
eigenvectors or eigenfunctions, i.e. given spectral data. In general, if some
such spectral data are given, there can be no system, a unique system, or
many systems, having these properties. It is remarkable as the present section
demonstrates, that there exist innite beams, corresponding to m = 0, 1, 2, . . .,
that possess the fundamental mode given in Eq. (4.19).
The natural question arises: Is it possible to formulate the problem so as
to obtain a unique solution? The reply to this question is in the afrmative.
Indeed, one can pre-select not only the fundamental mode shape, but also
the fundamental natural frequency denoted by
1
. Then, Eq. (4.46) yields the
coefcient b
m+4
that accomplishes this goal:
b
m+4
=
2
1
a
m
L
4
/12(m
2
+11m+30) (4.87)
The polynomial expressions have been used prior to this study in determ-
inistic analyses; yet, to the best of the authors knowledge, this is the rst
collection of closed-form results in either the deterministic or the prob-
abilistic setting for the natural frequencies and associated reliabilities of
inhomogeneous beams.
It is also notable that whereas in the usual so called stochastic nite ele-
ment method, only small coefcients of variation can be allowed, the present
formulation is not bound to small coefcients of variation. Therefore, the
deterministic and probabilistic closed-form solutions that were uncovered in
this chapter, can be utilized as benchmark solutions.
4.2 Apparently First Closed-Form Solutions for Inhomogeneous
Beams (Other Boundary Conditions)
There are several articles that deal with the vibrations of beams that are
non-uniformalong their axes. Usually, the variation is attributed to the cross-
sectional area. Then, for specic analytic expressions of such deterministic
variations, exact solutions are given in terms of special functions. The rst
solution for the natural frequency of a tapered beam, that of the wedge, was
pioneeredby Kirchhoff (1882). The solution was given in terms of Bessel func-
tions. Several other solutions of Kirchhoffs type followed. The appropriate
bibliography of problems solved in terms of Bessel functions was given by
Naguleswaran (1994a,b). The case where hypergeometric functions arise was
discussed by other authors. In some special cases, for beams that are clamped
at both ends, a transformation of the dependent variable is possible so that
the tapered beam shares a natural frequency of the uniform one (Abrate,
EOIS: 2892_chap04 2004/9/27 22:56 page 153 #19
1995). There are fewer studies that deal, in various approximate settings, with
vibrations of beams with random inhomogeneities. For example, Shinozuka
and Astill (1972), and Manohar and Keane (1993) considered several ran-
dom eigenvalue problems via analytical methods. The nite element method
in the stochastic setting was applied, amongst others, by Hart and Collins
(1970), Nakagari et al. (1987), and Zhu and Wu (1991). For extensive bib-
liography devoted to vibrations of discrete and continuous structures with
random parameters one may consult the extensive review by Ibrahim (1987)
and its recent up-date by Manohar and Ibrahim (1997).
The present section deals with beams which exhibit inhomogeneity both in
the material density and in the elastic modulus. These inhomogeneities are
described in terms of polynomial functions. Simple closed-form expressions
for both mode shapes and fundamental natural frequencies are uncovered
for a special class of problems. Then, the problem is considered in the prob-
abilistic setting with an attendant, seemingly paradoxical conclusion: beams
with random properties may possess the deterministic fundamental natural
frequencies. An extensive numerical study is conducted to substantiate this
nding.
4.2.2 Formulation of the Problem
Consider a non-unifom beam of length L, with cr oss-sectional ar ea A,
moments of inertia, I, that are constant, and variable material density (x)
and modulus of elasticity E(x). The beams vibrations are governed by the
BernouilliEuler equation:
2
x
2
E(x)I
2
w(x, t)
x
2
+(x)A
2
w(x, t)
x
2
= 0 (4.88)
where w(x,t) is the displacement, x the axial coodinate and t the time.
We introduce the non-dimensional coordinate = x/L, as well as consider
harmonic vibration, so that the displacement w(, t) is represented as follows:
w(, t) = W()e
it
(4.89)
where W()is the mode shape and is the sought for natural frequency. Thus,
Eq. (4.82) becomes
d
2
d
2
D()
d
2
W()
d
2
kL
4
R()W() = 0
k =
2
D = EI R = A
(4.90)
EOIS: 2892_chap04 2004/9/27 22:56 page 154 #20
The material density and elastic modulus are represented as polynomial
functions
R() =
m
i=0
a
i
i
D() =
n
i=0
b
i
i
(4.91)
where mand nare positive integers. We restrict our consideration to the case
n=m+4, since the rst term in Eq. (4.88) involves four spatial derivatives.
We are looking for that special class of problems in which the mode shape
W() is represented by the simplest polynomial functions that satisfy a given
set of boundary conditions.
4.2.3 Cantilever Beam
The beam is clamped at = 0 and free at = 1. The boundary conditions
read as follows:
W() = 0
dW()
d
= 0 at = 0
E()I
d
2
W()
d
2
= 0
d
d
_
E()I
d
2
W()
d
2
_
= 0 at = 1
(4.92)
A polynomial function that satises the boundary conditions in Eq. (4.92) is
given by
W() = w
1
(6
2
4
3
+
4
) (4.93)
It coincides with the expression of the rst comparison function in the set of
polynomial functions introduced by Duncan (1937) for studying beam vibra-
tion in the context of the BoobnovGalerkin method; w
1
is an indeterminate
coefcient. We also note that the expression in parentheses is propor-
tional to the static displacement of the uniform cantilever under constant
loading.
We pose the following question: what should the coefcients a
i
and b
i
be
so that the beams vibration mode coincides with Eq. (4.93)?
By substituting the expressions for D(), R(), W()into Eq. (4.90), we obtain
w
1
_
m+4
i=2
i(i 1)b
i
i2
(12
2
24 +12) +
m+4
i=0
24b
i
i
+2
m+4
i=1
ib
i
i1
(24 24)
kL
4
m
i=0
a
i
i
(
4
4
3
+6
2
)
_
= 0 (4.94)
EOIS: 2892_chap04 2004/9/27 22:56 page 155 #21
The latter expression can be re-written as follows, in a more convenient form:
24
m+3
i=1
i(i +1)b
i+1
i
+12
m+4
i=2
i(i 1)b
i
i
+12
m+2
i=0
(i +2)(i +1)b
i+2
i
+24
m+4
i=0
b
i
i
48
m+3
i=0
(i +1)b
i+1
i
+48
m+4
i=1
ib
i
i
6kL
4
m+2
i=2
a
i2
i
+4kL
4
m+3
i=3
a
i3
i
kL
4
m+4
i=4
a
i4
i
= 0 (4.95)
Equation (4.95) has to be satised for any . It will be shown later that one
has to distinguish between two special sub-cases: (a) m 3 and (b) m > 3. It
appears instructive to rst treat the particular cases m=0, m=1, m=2 and
m = 3.
4.2.3.1 Cantilever with uniform mass density (m=0)
In this case, the expressions for D() and R() read as
R() = a
0
D() =
4
i=0
b
i
i
(4.96)
By substituting the latter expressions in Eq. (4.90), we obtain
24
3
i=1
i(i +1)b
i+1
i
+12
4
i=2
i(i 1)b
i
i
+12
2
i=0
(i +1)(i +2)b
i+2
i
+24
4
i=0
b
i
i
48
3
i=0
(i +1)b
i+1
i
+48
4
i=1
ib
i
i
kL
4
a
0
(6
2
5
3
+2
4
) = 0 (4.97)
Equation (4.97) has to be satised for any . This requirement yields
24(b
0
+b
2
) 48b
1
= 0 (4.98)
72(b
1
+b
3
) 144b
2
= 0 (4.99)
144(b
2
+b
4
) 288b
3
6kL
4
a
0
= 0 (4.100)
240b
3
480b
4
+4L
4
ka
0
= 0 (4.101)
kL
4
a
0
+360b
4
= 0 (4.102)
EOIS: 2892_chap04 2004/9/27 22:56 page 156 #22
The sole unknown in Eqs. (4.13)(4.15) is the natural frequency coef-
cient k, yet we have ve equations. We conclude that the parameters b
i
and a
i
have to satisfy some auxiliary conditions so that Eqs. (4.13)(4.15) are
compatible.
Two compatibility conditions are given by Eq. (4.98) and Eq. (4.99),
leading to
b
0
= 2b
1
b
2
(4.103)
b
1
= 2b
2
b
3
(4.104)
From Eqs. (4.100)(4.102), three expressions for k can be found. These are
listed below:
k = [144(b
2
+b
4
) 288b
3
]/6L
4
a
0
(4.105)
k = [240b
3
+480b
4
]/L
4
4a
0
(4.106)
k = 360b
4
/L
4
a
0
(4.107)
To satisfy the compatibility requirement, all expressions for k have to be equal
to each other. We consider the case when the material density coefcients a
i
are specied. Then, the problem is reduced to determining the coefcients b
i
so that Eqs. (4.105)(4.107) are compatible.
Since the function R() of the inertial coefcient is given, so is the coefcient
a
0
. Let us observe that if b
4
is specied, then, the expressiongiveninEq. (4.107)
is the nal formula for the natural frequency coefcient k. Then Eqs. (4.105)
(4.107) allow an evaluation of the remaining parameters b
i
. Note that b
4
and
a
0
have to have the same sign since the natural frequency parameter k must
be positive. From Eq. (4.106), we obtain
b
3
= 4b
4
(4.108)
b
2
=
1
2
b
2
+4b
4
(4.109)
The b
i
where i = {0, 1, 2, 3}, can be re-written as follows:
b
3
= 4b
4
(4.110)
b
2
= 6b
4
(4.111)
b
1
= 16b
4
(4.112)
b
0
= 26b
4
(4.113)
To summarize if conditions (4.96) are satised, where b
i
are given
by Eqs. (4.110)(4.117), then, the fundamental mode shape is expressed
EOIS: 2892_chap04 2004/9/27 22:56 page 157 #23
by Eq. (4.93). The fundamental natural frequency reads as
2
= 360b
4
/a
0
L
4
(4.114)
4.2.3.2 Cantilever with linearly varying inertial coefcient (m=1)
In this case, the expressions for R() and D() read as
R() = a
0
+a
1
D() =
5
i=0
b
i
i
(4.115)
By substituting the latter expressions into Eq. (4.100), we obtain
24
4
i=1
i(i +1)b
i+1
i
+12
5
i=2
i(i 1)b
i
i
+12
3
i=0
(i +1)(i +2)b
i+2
i
+24
5
i=0
b
i
i
48
4
i=0
(i +1)b
i+1
i
+48
5
i=1
ib
i
i
kL
4
(a
0
+a
1
)(6
2
5
3
+2
4
) = 0 (4.116)
24(b
0
+b
2
) 48b
1
= 0 (4.117)
72(b
1
+b
3
) 144b
2
= 0 (4.118)
144(b
2
+b
4
) 288b
3
6kL
4
a
0
= 0 (4.119)
240(b
3
+b
5
) 480b
4
+L
4
(4ka
0
6ka
1
) = 0 (4.120)
360b
4
720b
5
+L
4
(4ka
1
ka
0
) = 0 (4.121)
kL
4
a
1
+504b
5
= 0 (4.122)
The coefcients b
i
, where i = {0, 1, 2, 3, 4}, can be evaluated so that the
compatibility of Eqs. (4.117)(4.121) is checked:
b
4
= (7a
0
18a
1
)b
5
/5a
1
(4.123)
b
3
= 2(11a
1
14a
0
)b
5
/5a
1
(4.124)
b
2
= 2(31a
1
+21a
0
)b
5
/5a
1
(4.125)
b
1
= 2(51a
1
+56a
0
)b
5
/5a
1
(4.126)
b
0
= 2(71a
1
+91a
0
)b
5
/5a
1
(4.127)
EOIS: 2892_chap04 2004/9/27 22:56 page 158 #24
We arrive at the following conclusion: if conditions (4.115) are satised,
where b
i
are given by Eqs. (4.123)(4.127), then the fundamental mode
shape is expressed by Eq. (4.93), where the fundamental natural frequency
reads as
2
= 504b
5
/a
1
L
4
(4.128)
4.2.3.3 Cantilever with parabolically varying inertial coefcient (m=2)
In this case, the expressions for R() and D() read as
R() = a
0
+a
1
+a
2
2
D() =
6
i=0
b
i
i
(4.129)
By substituting the latter expressions into Eq. (4.90), we obtain
24
5
i=1
i(i +1)b
i+1
i
+12
6
i=2
i(i 1)b
i
i
+12
4
i=0
(i +1)(i +2)b
i+2
i
+24
6
i=0
b
i
i
48
5
i=0
(i +1)b
i+1
i
+48
6
i=1
ib
i
i
kL
4
(a
0
+a
1
+a
2
2
)(6
2
5
3
+2
4
) = 0
(4.130)
Equation (4.130) has to be satised for any . This requirement is equivalent to
24(b
0
+b
2
) 48b
1
= 0 (4.131)
72(b
1
+b
3
) 144b
2
= 0 (4.132)
144(b
2
+b
4
) 288b
3
6kL
4
a
0
= 0 (4.133)
240(b
3
+b
5
) 480b
4
+L
4
(4ka
0
6ka
1
) = 0 (4.134)
360(b
4
+b
6
) 720b
5
+L
4
(4ka
1
ka
0
6ka
2
) = 0 (4.135)
504b
5
1008b
6
+L
4
(4ka
2
ka
1
) = 0 (4.136)
kL
4
a
2
+504b
6
= 0 (4.137)
To satisfy the compatibility equations, b
i
, where i = {0, 1, 2, 3, 4, 5}, have to be
b
5
= 2(2a
1
5a
2
)b
6
/3a (4.138)
b
4
= (53a
2
72a
1
+28a
0
)b
6
/15a
2
(4.139)
b
3
= 4(39a
2
+22a
1
28a
0
)b
6
/15a
2
(4.140)
b
2
= (259a
2
+248a
1
+168a
0
)b
6
/15a
2
(4.141)
EOIS: 2892_chap04 2004/9/27 22:56 page 159 #25
b
1
= 2(181a
2
+204a
1
+224a
0
)b
6
/15a
2
(4.142)
b
0
= (465a
2
+568a
1
+728a
0
)b
6
/15a
2
(4.143)
Thus, if conditions (4.129) are satised, where b
i
are given by
Eqs. (4.138)(4.143), the fundamental mode shape is given by Eq. (4.93), where
the fundamental natural frequency is
2
= 672b
6
/a
2
L
4
(4.144)
4.2.3.4 Cantilever with material inertial coeficent represented as
a cubic polynomial (m=3)
In this case, the expressions for R() and D() read
R() = a
0
+a
1
+a
2
2
+a
3
3
D() =
7
i=0
b
i
i
(4.145)
By substituting the latter expressions in Eq. (4.90), we obtain
24
6
i=1
i(i +1)b
i+1
i
+12
7
i=2
i(i 1)b
i
i
+12
5
i=0
(i +1)(i +2)b
i+2
i
+24
7
i=0
b
i
i
48
6
i=0
(i +1)b
i+1
i
+48
7
i=1
ib
i
i
kL
4
(a
0
+a
1
+a
2
2
+a
3
3
)(6
2
5
3
+2
4
) = 0
(4.146)
24(b
0
+b
2
) 48b
1
= 0 (4.147)
72(b
1
+b
3
) 144b
2
= 0 (4.148)
144(b
2
+b
4
) 288b
3
6kL
4
a
0
= 0 (4.149)
240(b
3
+b
5
) 480b
4
+L
4
(4ka
0
6ka
1
) = 0 (4.150)
360(b
4
+b
6
) 720b
5
+L
4
(4ka
1
ka
0
6ka
2
) = 0 (4.151)
504(b
5
+b
7
) 1008b
6
+L
4
(4ka
2
ka
1
6ka
3
) = 0 (4.152)
672b
6
1344b
7
+L
4
(4ka
3
ka
2
) = 0 (4.153)
kL
4
a
3
+864b
7
= 0 (4.154)
EOIS: 2892_chap04 2004/9/27 22:56 page 160 #26
To satisfy the compatibility equations, b
i
, where i = {0, 1, 2, 3, 4, 5, 6}, have
to be
b
6
= (9a
2
22a
3
)b
7
/7a
3
(4.155)
b
5
= 3(7a
3
10a
2
+4a
1
)b
7
/7a
3
(4.156)
b
4
= (320a
3
+195a
2
216a
1
+84a
0
)b
7
/35a
3
(4.157)
b
3
= (535a
3
+468a
2
+264a
1
336a
0
)b
7
/35a
3
(4.158)
b
2
= 3(250a
3
+259a
2
+248a
1
+168a
0
)b
7
/35a
3
(4.159)
b
1
= (965a
3
+1086a
2
+1225a
1
+1334a
0
)b
7
/35a
3
(4.160)
b
0
= (1180a
3
+1395a
2
+1704a
1
+2184a
0
)b
7
/35a
3
(4.161)
In summary, if conditions (4.145) are satised, where b
i
are given by
Eqs. (4.155)(4.161), then the fundamental mode shape is expressed by
Eq. (4.97), where the fundamental natural frequency reads as
2
= 864b
7
/a
3
L
4
(4.162)
4.2.3.5 Cantilever with material inertial coefcient represented as
a higher order polynomial (m>3)
Since Eq. (4.99) is valid for any , we conclude that
24(b
0
+b
2
) 48b
1
= 0 for i = 0 (4.163)
72(b
1
+b
3
) 144b
2
= 0 for i = 1 (4.164)
144(b
2
+b
4
) 288b
3
6kL
4
a
0
= 0 for i = 2 (4.165)
240(b
3
+b
5
) 480b
4
+L
4
(4ka
0
6ka
1
) = 0 for i = 3 (4.166)
. . .
12(i +1)(i +2)(b
i
+b
i+2
) 24(i +1)(i +2)b
i+1
+L
4
(4ka
i3
6ka
i2
ka
i4
) = 0 for 4 i m+2 (4.167)
. . .
12(m
2
+9m+42)b
m+3
24(m+4)(m+5)b
m+4
+L
4
(4ka
m
ka
m1
) = 0
for i = m+3 (4.168)
kL
4
a
m
+12(m
2
+11m+30)b
m+4
= 0 for i = m+4 (4.169)
EOIS: 2892_chap04 2004/9/27 22:56 page 161 #27
The only unknown in Eqs. (4.163)(4.169) is the natural frequency coefcient
k; yet, we have m + 4 equations. We conclude that the parameters b
i
and
a
i
have to satisfy some auxiliary conditions so that Eqs. (4.163)(4.169) are
compatible.
Two compatibility conditions are given by Eqs. (4.163) and (4.164), leading
to
b
0
= 2b
1
b
2
(4.170)
b
1
= 2b
2
b
3
(4.171)
From the other equations, several expressions for k can be found. These are
determined from Eqs. (4.163)(4.169), respectively, and are listed below
k = [144(b
2
+b
4
) 288b
3
]/6L
4
a
0
(4.172)
k =
240(b
3
+b
5
) 480b
4
L
4
(6a
1
4a
0
)
(4.173)
. . . ..
k =
12(i +1)(i +2)(b
i
+b
i+2
) 24(i +1)(i +2)b
i+1
L
4
(a
i4
+6a
i2
4a
i3
)
(4.174)
. . . ..
k =
12(m
2
+9m+42)b
m+3
24(m+4)(m+5)b
m+4
L
4
(a
m1
4a
m
)
(4.175)
k = 12(m
2
+11m+30)b
m+4
/L
4
a
m
(4.176)
To meet the compatibility requirement, all expressions for k have to be equal
to each other. We consider the case when the material density coefcients a
i
are specied. Then, the problem is reduced to determining the coefcients b
i
so that Eqs. (4.172)(4.176) are compatible.
Let us assume that the function R() of the inertial coeffcient, and hence all
a
i
(i = 0, 1, . . . , m) are given. Let us observe that if b
m+4
is specied, then, the
expression given in Eq. (4.176) is the nal formula for the natural frequency
coefcient k. Then Eqs. (4.172)(4.176) allow an evaluation of the remaining
parameters b
i
. Note that b
m+4
and a
m
have to have the same sign since the
natural frequency parameter k must be positive.
From Eq. (4.175), we obtain
b
m+3
=
b
m+4
a
m
(m
2
+9m+42)
[(m
2
+11m+30)a
m1
2(m
2
+13m+40)a
m
]
(4.177)
EOIS: 2892_chap04 2004/9/27 22:56 page 162 #28
b
i
=
1
(i +1)(4a
i2
6a
11
a
i3
)
{[(i +3)a
i4
+2(i +5)a
i3
+(2i 10)a
i2
12(i +1)a
i1
]b
i+1
+[2(i +3)a
i4
(7i 23)a
i3
+8(i +4)a
i2
+6(i +1)a
i1
]b
i+2
+[(i +3)a
i4
+4(i +3)a
i3
+(2i 10)a
i2
]b
i+3
} (4.178)
where i takes values {4, 5, . . . , m+2}.
b
3
=
(4a
o
+12a
2
+a
1
)b
4
+(11a
0
6a
2
14a
1
)b
5
+(9a
1
6a
0
)b
6
a
0
+6a
2
4a
1
(4.179)
From Eq. (4.177) we deduce that
b
2
= [(6a
1
+a
0
)b
3
(3a
1
+8a
0
)b
4
5a
0
b
5
]/(3a
1
2a
0
) (4.180)
We conclude that for specied coefcients a
0
, . . . , a
m
and b
m+4
, Eqs. (4.170),
(4.171), (4.177) and (4.180) result in the set of coefcients for the elastic mod-
ulus such that the beam has a mode shape given in Eq. (4.6). It is remarkable
that if a
i
i
To summarize, if R() and D() vary as in Eq. (4.95) with b
i
computed
via Eqs. (4.170), (4.171) and (4.177)(4.180), the fundamental mode shape
of the beam is given by Eq. (4.97) and the fundamental natural frequency
squared reads
2
= 12(m
2
+11m+30)b
m+4
/a
m
L
4
(4.181)
As we have established, in order for the closed-form solution to be obtainable,
it is sufcient that (1) all coefcients a
i
and(2) the coefcient b
m+4
be specied.
These requirements are not necessary: one can assume that all coefcients a
i
are givenandinsteadof the coefcient b
m+4
anyother coefcient b
j
(j = m+4)
is specied. If this is the case, then from Eq. (4.78) one expresses b
j+1
via b
j
and k; substitution into subsequent equations allows us to express b
m+2
, b
m+3
and b
m+4
via b
i
; analogously, substitution of b
i
into Eqs. (4.77)(4.79) yields
sought for exact solutions.
Although the natural frequency expressions for uniform density in
Eq. (4.100), for linearlyvaryingdensityinEq. (4.119), for parabolicallyvarying
density in Eq. (4.133) and cubically varying density in Eq. (4.149), are derived
separately from the case m>3, all these equations follow from Eq.(4.185) by
substituting appropriate values for m. Hence, Eq. (4.185) is the nal formula
for any integer value m.
EOIS: 2892_chap04 2004/9/27 22:56 page 163 #29
4.2.4 Beam that is Clamped at Both Ends
The beam is clamped at = 0 and = 1. The boundary conditions are
W() = 0
dW()
d
= 0 at = 0 and = 1 (4.182)
The simplest polynomial function that satises the boundary conditions in
Eq. (4.182) is given by
W() = w
1
(
2
2
3
+
4
) (4.183)
By substituting the expressions for D(), R()and W()in Eq. (4.90), we obtain
w
1
_
m+4
i=2
i(i 1)b
i
i2
(12
2
12 +2) +
m+4
i=0
24b
i
i
+2
m+4
i=1
ib
i
i1
(24 12) kL
4
m
i=0
a
i
i
(
4
2
3
+
2
)
_
= 0 (4.184)
The latter expression can be cast in the following form
12
m+3
i=1
i(i +1)b
i+1
i
+12
m+4
i=2
i(i 1)b
i
i
+2
m+2
i=0
(i +2)(i +1)b
i+2
i
+24
m+4
i=0
b
i
i
24
m+3
i=0
(i +1)b
i+1
i
+48
m+4
i=1
ib
i
i
kL
4
m+2
i=2
a
i2
i
+2kL
4
m+3
i=3
a
i3
i
kL
4
m+4
i=4
a
i4
i
= 0 (4.185)
Since Eq. (4.185) has to be satised for any , we arrive at the following
relations
24b
0
+4b
2
24b
1
= 0 for i = 0 (4.186)
72b
1
+12b
3
72b
2
= 0 for i = 1 (4.187)
144b
2
+24b
4
144b
3
kL
4
a
0
= 0 for i = 2 (4.188)
240b
3
+40b
5
240b
4
+L
4
(2ka
0
ka
1
) = 0 for i = 3 (4.189)
. . .
EOIS: 2892_chap04 2004/9/27 22:56 page 164 #30
12(i +1)(i +2)b
i
+2(i +1)(i +2)b
i+2
12(i +1)(i +2)b
i+1
+L
4
(2ka
i3
ka
i2
ka
i4
) = 0 for 4 i m+2 (4.190)
. . .
12(m+4)(m+5)b
m+3
12(m+4)(m+5)b
m+4
+L
4
(2ka
m
ka
m1
) = 0
for i = m+3 (4.191)
kL
4
a
m
+12(m
2
+11m+30)b
m+4
= 0 for i = m+4 (4.192)
It shouldbe borne in mindthat Eqs. (4.186)(4.192) are validonly if m > 3. For
cases that satisfy the inequality m 4, the reader is referred to Appendix A.
Note also that Eqs. (4.186)(4.192) have a recursive form, as do Eqs. (4.163)
(4.169) for the cantilever.
Two compatibility conditions are immediately detected for Eq. (4.186) and
Eq. (4.187), resulting in
b
0
= b
1

1
6
b
2
(4.193)
b
1
= b
2

1
6
b
3
(4.194)
determined from Eqs. (4.188)(4.192), respectively. The alternative analytical
formulas for k are:
k = [144(b
2
b
3
) 24b
4
]/L
4
a
0
(4.195)
k = [240(b
3
b
4
) 40b
5
]/L
4
(a
1
2a
0
) (4.196)
. . .
k =
12(i +1)(i +2)(b
i
b
i+1
) 2(i +1)(i +2)b
i+2
L
4
(a
i4
+a
i2
2a
i3
)
(4.197)
. . .
k =
12(m+4)(m+5)(b
m+3
b
m+4
)
L
4
(a
m1
2a
m
)
(4.198)
k = 12(m
2
+11m+30)b
m+4
/L
4
a
m
(4.199)
Equations (4.195)(4.199) then allowan evaluation of the remaining paramet-
ers b
i
. It is worth noting that b
m+4
and a
m
have to have the same sign due to
the positivity of k. From Eq. (4.198), we get
b
m+3
=
b
m+4
a
m
(m+4)(m+5)
[(m
2
+11m+30)a
m1
(m
2
+13m+40)a
m
]
(4.200)
EOIS: 2892_chap04 2004/9/27 22:56 page 165 #31
b
i
=
1
6(i +1)(2a
i2
a
i1
a
i3
)
{[6(i +3)a
i4
+6(i +5)a
i3
+6(i 1)a
i2
6(i +1)a
i1
]b
i+1
+[6(i +3)a
i4
(37i +13)a
i3
+4(2i +5)a
i2
(i +1)a
i1
]b
i+2
+[(i +3)a
i4
2(i +3)a
i3
+(i +3)a
i2
]b
i+3
} (4.201)
where i belongs to the set {4, 5, . . . , m+2}. Equation (4.196) leads to
b
3
=
6(4a
0
+a
1
2a
2
)b
4
(38a
0
2a
2
22a
1
)b
5
3(2a
0
a
1
)b
6
12(2a
1
a
0
a
2
)
(4.202)
b
2
=
(18a
1
6a
0
)b
3
+(3a
1
36a
0
)b
4
5a
0
b
5
18(a
1
2a
0
)
(4.203)
Therefore, for speciedcoefcients a
0
, a
1
, . . . , a
m
andb
m+4
, Eqs. (4.193), (4.194)
and (4.200)(4.203) lead to the set of coefcients for the elastic modulus such
that the beam possesses the mode shape given in Eq. (4.183). In perfect ana-
logy with the cantilever, if a
i
i
do not depend on
the parameter a. We conclude that, if R() and D() vary as in Eq. (4.91),
with attendant b
i
computed via Eqs. (4.193), (4.194) and (4.200)(4.203), the
fundamental mode shape of the beam is governed by Eq. (4.183) and the
fundamental natural frequency squared reads as
2
= 12(m
2
+11m+30)b
m+4
/L
4
a
m
(4.204)
4.2.5 Beam Clamped at One End and Pinned at the Other
Consider now the beam that is clamped at =0 and pinned at =1. The
boundary conditions are
W() = 0
dW()
d
= 0 at = 0
W() = 0 EI
d
2
W()
d
2
= 0 at = 1
(4.205)
The simplest polynomial function that satises the boundary conditions in
Eq. (4.205), and does not have a node point in the interval (0, 1) is given by
W() = w
1
_
3
2
5
3
+2
4
_
(4.206)
EOIS: 2892_chap04 2004/9/27 22:56 page 166 #32
By substituting the expressions for D(), R()and W()in Eq. (4.90), we obtain
w
1
_
m+4
i=0
i(i 1)b
i
i2
(24
2
30 +6) +
m+4
i=0
48b
i
i
+2
m+4
i=1
ib
i
i1
(48 30)
kL
4
m
i=0
a
i
i
(2
4
5
3
+3
2
)
_
= 0 (4.207)
Equation (4.207) can be re-written as follows:
30
m+3
i=1
i(i +1)b
i+1
i
+24
m+4
i=2
i(i 1)b
i
i
+6
m+2
i=0
(i +2)(i +1)b
i+2
i
+48
m+4
i=0
b
i
i
60
m+3
i=0
(i +1)b
i+1
i
+96
m+4
i=1
ib
i
i
3kL
4
m+2
i=2
a
i2
i
+5kL
4
m+3
i=3
a
i3
i
2kL
4
m+4
i=4
a
i4
i
= 0 (4.208)
Since Eq. (4.208) has to be satised for any , we conclude that
48b
0
60b
1
+12b
2
= 0 for i = 0 (4.209)
144b
1
180b
2
+36b
3
= 0 for i = 1 (4.210)
288b
2
360b
3
+72b
4
3kL
4
a
0
= 0 for i = 2 (4.211)
480b
3
600b
4
+120b
5
+L
4
(5ka
0
3ka
1
) = 0 for i = 3 (4.212)
. . .
24(i +1)(i +2)b
i
30(i +1)(i +2)b
i+1
+6(i +1)(i +2)b
i+2
+L
4
(5ka
i3
3ka
i2
2ka
i4
) = 0 for 4 i m+2 (4.213)
24(m+4)(m+5)b
m+3
30(m+4)(m+5)b
m+4
+L
4
(5ka
m
2ka
m1
) = 0
for i = m+3 (4.214)
kL
4
a
m
+12(m
2
+11m+30)b
m+4
= 0 for i = m+4 (4.215)
It should be borne in mind that Eqs. (4.209)(4.215) are valid only if m > 3.
For cases that satisfy the inequality m3, one can derive a solution in perfect
analogy with Appendix A, which deals with a pinned beam. Note also that
Eqs. (4.209)(4.215) have a recursive form.
EOIS: 2892_chap04 2004/9/27 22:56 page 167 #33
Two compatibility conditions are given by Eqs. (4.209) and (4.210), with
b
0
=
1
4
(5b
1
b
2
) (4.216)
b
1
=
1
4
(5b
2
b
3
) (4.217)
determined from Eqs. (4.211)(4.215), respectively, and are listed below:
k = (96b
2
120b
3
+24b
4
)/L
4
a
0
(4.218)
k = (480b
3
600b
4
+120b
5
)/L
4
(3a
1
5a
0
) (4.219)
. . .
k =
6(i +1)(i +2)(4b
i
5b
i+1
+b
i+2
)
L
4
(2a
i4
+3a
i2
5a
i3
)
(4.220)
. . .
k =
6(m+4)(m+5)(4b
m+3
5b
m+4
)
L
4
(2a
m1
5a
m
)
(4.221)
k = 12(m
2
+11m+30)b
m+4
/L
4
a
m
(4.222)
Then, Eqs. (4.218)(4.222) allow an evaluation of the remaining parameters
b
i
. Note that b
m+4
and a
m
have to have the same sign due to the positivity of
k. From Eq. (4.221), we obtain
b
m+3
=
b
m+4
4a
m
(m+4)
[4(m+6)a
m1
5(m+8)a
m
] (4.223)
b
i
=
1
4(i +1)(5a
i2
3a
i1
2a
i3
)
{[8(i +3)a
i4
+10(i +5)a
i3
+(13i 11)a
i2
15(i +1)a
i1
]b
i+1
+[10(i +3)a
i4
(23i +73)a
i3
+10(i +4)a
i2
+3(i +1)a
i1
]b
i+2
+[2(i +3)a
i4
+5(i +3)a
i3
3(i +3)a
i2
]b
i+3
} (4.224)
where i belongs to the set {4, 5, . . . , m+2}. Equation (4.219) results in
b
3
=
(40a
o
30a
2
+14a
1
)b
4
+(71a
0
+6a
2
+35a
1
)b
5
+(9a
1
+15a
0
)b
6
8(5a
1
3a
2
2a
0
)
(4.225)
EOIS: 2892_chap04 2004/9/27 22:56 page 168 #34
b
2
=
5(5a
1
a
0
)b
3
(3a
1
+20a
0
)b
4
+5a
0
b
5
4(3a
1
5a
0
)
(4.226)
Thus, for specied coefcients a
0
, a
1
, . . . , a
m
andb
m+4
, Eqs. (4.216), (4.217) and
(4.221)(4.226) lead to the set of coefcients for the elastic modulus such that
the beam possesses the mode shape given in Eq. (4.206). Note that if a
i
= a,
then the coefcients b
i
The results can be summarized as follows: if R() and D() vary as in
Eq. (4.90) with b
i
, computed via Eqs. (4.216), (4.217) and (4.224)(4.226),
the fundamental mode shape of the beam is given by Eq. (4.206) and the
fundamental natural frequency squared is
2
= 12(m
2
+11m+30)b
m+4
/L
4
a
m
(4.227)
4.2.6 Random Beams with Deterministic Frequencies
As seen in Eqs. (4.181), (4.204) and (4.227) the fundamental natural frequency
depends only on the terminal coefcients a
m
and b
m+4
. If either of these coef-
cients is random, so is the natural frequency. In the latter case one can ask
whether the evaluation is reliable, the reliability being dened as the probab-
ility that the natural frequency does not exceed any pre-selected value. Such
an analysis, for the beams that were pinned at their ends was conducted by
Candan and Elishakoff (2000).
Yet, Eqs. (4.181), (4.204) and(4.227) for beams with three different boundary
conditions leadto a remarkable conclusion: If the material density coefcients
a
0
, a
1
, . . . , a
m1
and the elastic modulus coefcients b
0
, b
1
, . . . ., b
m+3
are ran-
dom, but the quantities a
m
andb
m+4
are deterministic, the natural frequency is a
deterministic variable too. Thus, although the beam is random, its fundamental
frequency is deterministic. The present writers do not knowof any other study
that reports an analogous occurrence. The closest one is the study by Fraser
and Budiansky (1969), which dealt with buckling of elastic beams on a non-
linear elastic foundation; the beam possessed random initial imperfection
deviation from the straight line with a given autocorrelation function. The
beams were treated as having innite length and the initial imperfections
constituted an ergodic random eld. Fraser and Budiansky (1969) used an
approximate technique which resulted in deterministic buckling loads. The
buckling load was dened as the maximum axial load the beam could support.
utilized different approximate analyses. Amazigo (1976) ascribed this seem-
ingly paradoxical behavior to the property of ergodicity that was postulated
for the random initial imperfection.
Elishakoff (1979) re-examined the FraserBudiansky problem for nite
beams on non-linear elastic foundations. The Monte Car lo method was
Several other studies followed (see, for bibliography; Amazigo, 1976) which
EOIS: 2892_chap04 2004/9/27 22:56 page 169 #35
utilized in conjunction with developing a special procedure for solving a
nonlinear boundary-value problem for each realization of the random beam.
In nite beams, the buckling loads did not turn out to be deterministic. Yet,
the coefcient of variation of the buckling load was shown to be a decreasing
function with increase in length.
How can one detect that a complex structure may possess deterministic
eigenvalues? Such structures obviously are analyzed by approximate ana-
lytical and/or numerical techniques. In order to answer this question we
simulate the realistic situation and apply the Monte Carlo method to check
the validity of the main conclusion of this study, namely, that random beams
may have deterministic frequencies.
The particular case considered hereinafter is m = 2. The coefcients a
0
and a
1
were taken to be exponentially distributed independent random vari-
ables and 1089 realizations of beams were simulated. For each realization
of coefcients a
0
and a
1
the appropriate coefcients b
0
, . . . , b
5
were evalu-
ated. For simplicity the coefcients a
m
= a
2
and b
m+4
= b
6
were xed
at unity. For each realization of the beam the nite element method was
applied. For one of the simulated beams, with material density coefcients
a
0
= 0.557602, a
1
= 0.387297 and a
2
= 1, the associated elastic modulus
ary condition. The convergence of the nite element method is illustrated in
Table 4.3 with the exact solution for the natural frequency coefcient being
k = 672. The percentagewise errors from the exact solution are listed in
to be equal to four, with maximum error, which for the clampedclamped
beam, was only 0.41%. This is in agreement with the observation by Gupta
and Rao (1978) who concluded that the nite element procedure developed
for the eigenvalue analysis of doubly tapered and twisted BresseTimoshenko
beams has beenfoundto give reasonably accurate results evenwithfour nite
TABLE 4.3
Convergence of the Natural Frequency Squared for Different Boundary
Conditions (a
0
= 0.557602, a
1
= 0.387297, a
2
= 1)
Number of
elements Pinned Clampedpinned Clampedclamped Clampedfree
1 791.789673 1146.29273 696.96113
2 678.667018 691.102962 717.006005 673.551345
3 673.321172 675.750735 680.733099 672.307085
4 672.418974 673.188072 674.761629 672.097261
5 672.171821 672.487131 673.131839 672.039859
6 672.082919 672.235081 672.546128 672.019228
7 672.044777 672.126948 672.294903 672.01038
8 672.026255 672.074438 672.172916 672.006086
9 672.016394 672.046481 672.107973 672.0038
10 672.010758 672.030501 672.070852 672.002493
coefcients are given in Tables 4.5, 4.6, 4.7 and 4.8, depending on the bound-
Table 4.4. For the subsequent calculations, the number of elements was taken
EOIS: 2892_chap04 2004/9/27 22:56 page 170 #36
TABLE 4.4
Percentagewise Error
Number of
Elements Pinned Clampedpinned Clampedclamped Clampedfree
1 17.8258442 70.5792753 3.714453869
2 0.992115774 2.842702679 6.697322173 0.230854911
3 0.196602976 0.558145089 1.299568304 0.045697173
4 0.062347321 0.176796429 0.410956696 0.014473363
5 0.025568601 0.072489732 0.168428423 0.005931399
6 0.012339137 0.034982292 0.081269048 0.00286131
7 0.006663244 0.018891071 0.043884375 0.001544643
8 0.003906994 0.011077083 0.025731548 0.000905655
9 0.002439583 0.006916815 0.016067411 0.000565476
10 0.001600893 0.004538839 0.010543452 0.000370982
elements. Sample calculations for ten realizations of the random beam are
beams pinned at both ends, Table 4.5 also lists the results obtainable from the
single-term BoobnovGalerkin method with a sinusoidal comparison func-
tion, sin . It can be observed from Table 4.5 that the sample frequencies are
concentrated around the exact solution k=672 none exceeding the value 673.
clamped-clamped beam all frequencies in Table 4.8 are below 675.
Results of the Monte Carlo simulation for 1089 sample beams are given
level of signicance of 0.01, the maximum absolute difference between exact
and empirical reliabilities is smaller than 1.63/
1089 = 0.049. It lists mean

values, standard deviation and the coefcient of variation for the natural fre-
quencies. It is seen that for all four sets of boundary conditions, the standard
deviation is much smaller than the mean natural frequency. The resulting
coefcients of variation all are less than 10
5
. Extreme smallness of the coef-
cient of variation supports our theoretical nding that the natural frequency
constitutes a deterministic quantity.
Likewise, it appears that if the results of the Monte Carlo simulation of a
complexstructureexhibit small coefcients of variationfor the eigenvalues irre-
spective of the moderate or large coefcients of variationof the input stochastic
quantities, one is facing the phenomenon discovered in this study, namely
the random structures possessing the deterministic eigenvalues. Soize (1999,
2000a,b) also found that some stochastic systems may possess a deterministic
fundamental frequency.
The difference from the paper by Fraser and Budiansky (1969) lies in the
fact that the deterministic property of the buckling loads was possibly due
to the ergodicity of the input random elds (Amazigo, 1976) expanding from
minus innity to plus innity, combined with approximate analysis. Here,
listed in Tables 4.54.8 for various boundary conditions. For the case of the
The same occurs for the CF beams (Table 4.6). For clamped pinned beams
none of the random frequencies in Table 4.7 exceeds 674, whereas for the
in Table 4.9. At this size of the sample, according to Massey (1951), at the
E
O
I
S
:
2
8
9
2
_
c
h
a
p
0
4
2
0
0
4
/
9
/
2
7
2
2
:
5
6
p
a
g
e
1
7
1
#
3
7
U
n
u
s
u
a
l
C
l
o
s
e
d
-
F
o
r
m
S
o
l
u
t
i
o
n
s
f
o
r
B
e
a
m
V
i
b
r
a
t
i
o
n
s
1
7
1
TABLE 4.5
Sample Calculations for Fundamental and Second Natural Frequencies of Pinned Beams
Approximation:
6
i=1
w
i
sin(i)
First natural
frequency coefcient
k
Second natural
frequency coefcient
k
Coefcients a[j] Coefcients b[j]
Boobnov
Galerkin
method FEM
Boobnov
Galerkin
method FEM
a[0] = 0.557602
a[1] = 0.387297
a[2] = 1.000000
b[0] = 5.133779
b[4] = 1.555322
b[1] = 5.133779
b[5] = 1.150270
b[2] = 0.070508
b[6] = 1.000000
b[3] = 1.877895
672.006907 672.000673 10,565.5 10,565.4
a[0] = 0.421035
a[1] = 0.642941
a[2] = 1.000000
b[0] = 4.948464
b[4] = 2.423793
b[1] = 4.948464
b[5] = 0.809412
b[2] = 1.018801
b[6] = 1.000000
b[3] = 1.981591
672.006899 672.000673 10,801.0 10,800.8
a[0] = 0.817484
a[1] = 1.373393
a[2] = 1.000000
b[0] = 8.824267
b[4] = 3.436841
b[1] = 8.824267
b[5] = 0.164524
b[2] = 1.194419
b[6] = 1.000000
b[3] = 5.214750
672.006882 672.000672 10,703.8 10,703.7

2
0
0
5

b
y

I
s
s
a
c

E
l
i
s
h
a
k
o
f
f
E
O
I
S
:
2
8
9
2
_
c
h
a
p
0
4
2
0
0
4
/
9
/
2
7
2
2
:
5
6
p
a
g
e
1
7
2
#
3
8
1
7
2
E
i
g
e
n
v
a
l
u
e
s
o
f
I
n
h
o
m
o
g
e
n
e
o
u
s
S
t
r
u
c
t
u
r
e
s
TABLE 4.5
Continued
Approximation:
6
i=1
w
i
sin(i)
First natural
frequency coefcient
k
Second natural
frequency coefcient
k
Boobnov
Galerkin
Method FEM
Boobnov
Galerkin
Method FEM
a[0] = 0.032587
a[1] = 1.283644
a[2] = 1.000000
b[0] = 4.225417
b[4] = 4.686583
b[1] = 4.225417
b[5] = 0.044859
b[2] = 3.921267
b[6] = 1.000000
b[3] = 2.069073
672.006868 672.000673 11,829.6 11,829.4
a[0] = 0.927113
a[1] = 1.368769
a[2] = 1.000000
b[0] = 9.427710
b[4] = 3.221101
b[1] = 9.427710
b[5] = 0.158359
b[2] = 0.774654
b[6] = 1.000000
b[3] = 5.612935
672.006882 672.000671 10,653.4 10,653.3
a[0] = 0.300851
a[1] = 0.371173
a[2] = 1.000000
b[0] = 3.659425
b[4] = 1.995894
b[1] = 3.659425
b[5] = 1.171769
b[2] = 0.851482
b[6] = 1.000000
b[3] = 0.880659
672.006913 672.000673 10,872.9 10,872.7

2
0
0
5

b
y

I
s
s
a
c

E
l
i
s
h
a
k
o
f
f
E
O
I
S
:
2
8
9
2
_
c
h
a
p
0
4
2
0
0
4
/
9
/
2
7
2
2
:
5
6
p
a
g
e
1
7
3
#
3
9
U
n
u
s
u
a
l
C
l
o
s
e
d
-
F
o
r
m
S
o
l
u
t
i
o
n
s
f
o
r
B
e
a
m
V
i
b
r
a
t
i
o
n
s
1
7
3
a[0] = 1.595434
a[1] = 0.039957
a[2] = 1.000000
b[0] = 10.158333
b[4] = 1.215579
b[1] = 10.158333
b[5] = 1.613390
b[2] = 4.732383
b[6] = 1.000000
b[3] = 4.918851
672.006896 672.000671 10,274.3 10,274.2
a[0] = 0.092588
a[1] = 3.231425
a[2] = 1.000000
b[0] = 8.976387
b[4] = 9.249257
b[1] = 8.976387
b[5] = 2.641900
b[2] = 8.112237
b[6] = 1.000000
b[3] = 6.967747
672.006847 672.000671 11,639.2 11,639.0
a[0] = 1.676221
a[1] = 0.604673
a[2] = 1.000000
b[0] = 11.890765
b[4] = 0.011065
b[1] = 11.890765
b[5] = 0.860436
b[2] = 3.753968
b[6] = 1.000000
b[3] = 6.575775
672.006888 672.000671 10,363.8 10,363.7
a[0] = 1.032913
a[1] = 2.533571
a[2] = 1.000000
b[0] = 12.660406
b[4] = 5.819134
b[1] = 12.660406
b[5] = 1.711428
b[2] = 3.019887
b[6] = 1.000000
b[3] = 8.803445
672.006870 672.000671 10,802.4 10,802.3

2
0
0
5

b
y

I
s
s
a
c

E
l
i
s
h
a
k
o
f
f
EOIS: 2892_chap04 2004/9/27 22:56 page 174 #40
TABLE 4.6
Sample Calculations for Natural Frequencies of ClampedFree Beams
Natural frequency coefcient k
by the FEM
a[0] = 0.557602
a[1] = 0.387297
a[2] = 1.000000
b[0] = 72.727949
b[4] = 2.715164
b[1] = 51.321537
b[5] = 2.816937
b[2] = 29.915125
b[6] = 1.000000
b[3] = 8.508713
672.097261
a[0] = 0.421035
a[1] = 0.642941
a[2] = 1.000000
b[0] = 75.780285
b[4] = 1.233148
b[1] = 54.196254
b[5] = 2.476078
b[2] = 32.612222
b[6] = 1.000000
b[3] = 11.028191
672.097963
a[0] = 0.817484
a[1] = 1.373393
a[2] = 1.000000
b[0] = 122.681037
b[4] = 1.532985
b[1] = 85.905145
b[5] = 1.502142
b[2] = 49.129253
b[6] = 1.000000
b[3] = 12.353362
672.096366
a[0] = 0.032587
a[1] = 1.283644
a[2] = 1.000000
b[0] = 81.188912
b[4] = 2.567330
b[1] = 60.021739
b[5] = 1.621808
b[2] = 38.854566
b[6] = 1.000000
b[3] = 17.687394
672.099769
a[0] = 0.927113
a[1] = 1.368769
a[2] = 1.000000
b[0] = 127.826609
b[4] = 1.306147
b[1] = 89.053628
b[5] = 1.508308
b[2] = 50.280647
b[6] = 1.000000
b[3] = 11.507667
672.096056
a[0] = 0.300851
a[1] = 0.371173
a[2] = 1.000000
b[0] = 59.656390
b[4] = 2.313291
b[1] = 43.214658
b[5] = 2.838436
b[2] = 26.772926
b[6] = 1.000000
b[3] = 10.331195
672.098978
EOIS: 2892_chap04 2004/9/27 22:56 page 175 #41
TABLE 4.6
Continued
by the FEM
a[0] = 1.595434
a[1] = 0.039957
a[2] = 1.000000
b[0] = 109.944777
b[4] = 6.319681
b[1] = 72.870466
b[5] = 3.280057
b[2] = 35.796155
b[6] = 1.000000
b[3] = 1.278156
672.093444
a[0] = 0.092588
a[1] = 3.231425
a[2] = 1.000000
b[0] = 157.856878
b[4] = 11.804677
b[1] = 114.793376
b[5] = 0.975233
b[2] = 71.729875
b[6] = 1.000000
b[3] = 28.666374
672.098338
a[0] = 1.676221
a[1] = 0.604673
a[2] = 1.000000
b[0] = 135.249566
b[4] = 3.759849
b[1] = 90.643587
b[5] = 2.527103
b[2] = 46.037608
b[6] = 1.000000
b[3] = 1.431629
672.093899
a[0] = 1.032913
a[1] = 2.533571
a[2] = 1.000000
b[0] = 177.068593
b[4] = 6.699704
b[1] = 123.896129
b[5] = 0.044762
b[2] = 70.723666
b[6] = 1.000000
b[3] = 17.551202
672.096116
we do not use the ergodicity assumption, the beams have nite length and
the results are obtained in closed form.
4.3 Inhomogeneous Beams that may Possess a Prescribed
Polynomial Second Mode
The exact solutions for the vibration frequencies of non-uniform rods and
beams were considered, respectively, by Ward (1913) and Kirchhoff (1882),
apparently for the rst time, in terms of Bessel functions. General-purpose
EOIS: 2892_chap04 2004/9/27 22:56 page 176 #42
TABLE 4.7
Sample Calculations for Natural Frequencies of ClampedPinned Beams
by the FEM
a[0] = 0.557602
a[1] = 0.387297
a[2] = 1.000000
b[0] = 2.386304
b[4] = 0.175201
b[1] = 2.172488
b[5] = 1.566937
b[2] = 1.317223
b[6] = 1.000000
b[3] = 2.103834
673.188072
a[0] = 0.421035
a[1] = 0.642941
a[2] = 1.000000
b[0] = 2.352934
b[4] = 1.197058
b[1] = 2.172475
b[5] = 1.226078
b[2] = 1.450641
b[6] = 1.000000
b[3] = 1.436697
673.190547
a[0] = 0.817484
a[1] = 1.373393
a[2] = 1.000000
b[0] = 4.127546
b[4] = 2.648377
b[1] = 3.778540
b[5] = 0.252142
b[2] = 2.382518
b[6] = 1.000000
b[3] = 3.201570
673.18739
a[0] = 0.032587
a[1] = 1.283644
a[2] = 1.000000
b[0] = 2.159529
b[4] = 3.844270
b[1] = 2.078161
b[5] = 0.371808
b[2] = 1.752691
b[6] = 1.000000
b[3] = 0.450809
673.197488
a[0] = 0.927113
a[1] = 1.368769
a[2] = 1.000000
b[0] = 4.382875
b[4] = 2.429862
b[1] = 3.997323
b[5] = 0.258308
b[2] = 2.455116
b[6] = 1.000000
b[3] = 3.713713
673.186588[6pt]
a[0] = 0.300851
a[1] = 0.371173
a[2] = 1.000000
b[0] = 1.757614
b[4] = 0.606097
b[1] = 1.630702
b[5] = 1.588436
b[2] = 1.123053
b[6] = 1.000000
b[3] = 0.907543
673.192617
EOIS: 2892_chap04 2004/9/27 22:56 page 177 #43
TABLE 4.7
Continued
by the FEM
a[0] = 1.595434
a[1] = 0.039957
a[2] = 1.000000
b[0] = 4.457584
b[4] = 2.804104
b[1] = 3.912576
b[5] = 2.030057
b[2] = 1.732547
b[6] = 1.000000
b[3] = 6.987571
673.179294
a[0] = 0.092588
a[1] = 3.231425
a[2] = 1.000000
b[0] = 4.521305
b[4] = 9.575612
b[1] = 4.324959
b[5] = 2.225233
b[2] = 3.539577
b[6] = 1.000000
b[3] = 0.398050
673.194604
a[0] = 1.676221
a[1] = 0.604673
a[2] = 1.000000
b[0] = 5.293016
b[4] = 1.260761
b[1] = 4.693411
b[5] = 1.277103
b[2] = 2.294989
b[6] = 1.000000
b[3] = 7.298696
673.180924
a[0] = 1.032913
a[1] = 2.533571
a[2] = 1.000000
b[0] = 5.961455
b[4] = 5.726776
b[1] = 5.483693
b[5] = 1.294762
b[2] = 3.572644
b[6] = 1.000000
b[3] = 4.071551
673.187544
nite element codes can handle this problem successfully. Since then numer-
ous studies have been performed to obtain both exact and approximate solu-
tions. Presently, the attraction to exact solutions appears to be disappearing
due to the nite element methods versality and its fast convergence.
The natural question arises: Does it makes sense to still look for closed form
solutions? One argument for obtaining exact solutions is that they can serve
as benchmark solutions. The second argument is less articulated: if the exact
solution is extremely simple and reveals some interesting underpinnings of
the problem, it appears to be worth pursuing. Closed-form solutions will be
obtained in the present section for the natural frequency and mode shape of
inhomogeneous beams that possess uniform cross-section along the beams
EOIS: 2892_chap04 2004/9/27 22:56 page 178 #44
TABLE 4.8
Sample Calculations for Natural Frequencies of ClampedClamped Beams
by the FEM
a[0] = 0.557602
a[1] = 0.387297
a[2] = 1.000000
b[0] = 1.013858
b[4] = 0.144678
b[1] = 1.159905
b[5] = 1.150270
b[2] = 0.876280
b[6] = 1.000000
b[3] = 1.701751
674.761629
a[0] = 0.421035
a[1] = 0.642941
a[2] = 1.000000
b[0] = 0.973584
b[4] = 0.723793
b[1] = 1.130086
b[5] = 0.809412
b[2] = 0.939010
b[6] = 1.000000
b[3] = 1.146453
674.76113
a[0] = 0.817484
a[1] = 1.373393
a[2] = 1.000000
b[0] = 1.755885
b[4] = 1.736841
b[1] = 2.023840
b[5] = 0.164524
b[2] = 1.607728
b[6] = 1.000000
b[3] = 2.496669
674.764592
a[0] = 0.032587
a[1] = 1.283644
a[2] = 1.000000
b[0] = 0.819969
b[4] = 2.986583
b[1] = 0.997884
b[5] = 0.044859
b[2] = 1.067493
b[6] = 1.000000
b[3] = 0.417655
674.758823
a[0] = 0.927113
a[1] = 1.368769
a[2] = 1.000000
b[0] = 1.878854
b[4] = 1.521101
b[1] = 2.157733
b[5] = 0.158359
b[2] = 1.673270
b[6] = 1.000000
b[3] = 2.906774
674.764933
a[0] = 0.300851
a[1] = 0.371173
a[2] = 1.000000
b[0] = 0.713686
b[4] = 0.295894
b[1] = 0.831554
b[5] = 1.171769
b[2] = 0.707207
b[6] = 1.000000
b[3] = 0.746080
674.758412
EOIS: 2892_chap04 2004/9/27 22:56 page 179 #45
TABLE 4.8
Continued
by the FEM
a[0] = 1.595434
a[1] = 0.039957
a[2] = 1.000000
b[0] = 2.040809
b[4] = 2.915579
b[1] = 2.261035
b[5] = 1.613390
b[2] = 1.321356
b[6] = 1.000000
b[3] = 5.638072
674.765655
a[0] = 0.092588
a[1] = 3.231425
a[2] = 1.000000
b[0] = 1.768522
b[4] = 7.549257
b[1] = 2.140223
b[5] = 2.641900
b[2] = 2.230211
b[6] = 1.000000
b[3] = 0.539927
674.763433
a[0] = 1.676221
a[1] = 0.604673
a[2] = 1.000000
b[0] = 2.388134
b[4] = 1.711065
b[1] = 2.671119
b[5] = 0.860436
b[2] = 1.697904
b[6] = 1.000000
b[3] = 5.839285
674.766146
a[0] = 1.032913
a[1] = 2.533571
a[2] = 1.000000
b[0] = 2.525862
b[4] = 4.119134
b[1] = 2.927879
b[5] = 1.711428
b[2] = 2.412098
b[6] = 1.000000
b[3] = 3.094684
674.767408
TABLE 4.9
Statistical Properties of Natural Frequency for Different Boundary Conditions (1089
Samples)
Boundary
conditions Pinnedpinned Clampedfree Clampedclamped Clampedpinned
Mean 672.4186337 672.0966348 674.762564 673.1879431
Standard
deviation
0.000767083 0.002587103 0.0046755 0.006558834
Coefcient of
variation
1.14078 10
6
3.8493 10
6
6.9291 10
6
9.74295 10
6
EOIS: 2892_chap04 2004/9/27 22:56 page 180 #46
axis. Usually, once the properties of the structure are known one determines
the locationof nodes andanti-nodes (Grgoze, 1999). We ask a questionabout
the type of inverse problems that exist: nd the distribution of the elastic
modulus and the material density that yield a pre-selected, polynomial mode
shape. We pose the additional question: Is it possible to determine the mode
shape that has a nodal point at a pre-selected location on the beams axis? A
somewhat related problem, for spring-mass systems, was treated by Cha and
Peirre (1999). They studied a case of a one-dimensional arbitrarily pinned
chain of oscillators and showed that the desired node position can either
case, it is always possible to get a node at any pre-selected location on the
structure whereas in the latter case one can induce the node for certain normal
modes only. Earlier, McLaughlin (1988) and Hald and McLaughlin (1998b)
studied the inverse problemof the special type, namely, of reconstructing the
structure from the set of nodal points or nodal lines. The motivation for this
section stems from the work by Candan and Elishakoff (2000) who focused
on the fundamental natural frequency of inhomogenous beams. Here, we are
concerned with the second frequency. This section closely follows the study
by Neuringer and Elishakoff (2001).
4.3.2 Basic Equation
The dynamical behavior of a uniform, inhomogeneous beam is described by
the following equation
d
2
dx
2
_
E(x)I
d
2
W
dx
2
_
(x)
2
AW(x) = 0 (4.228)
where E is the modulus of elasticity, (x) the density, I the moment of inertia,
A the cross-sectional area, W(x) the modal displacement and the natural
frequency. Letting = x/L represent the non-dimensional axial coordin-
ate, where L is the length of the beam, introducing the natural frequency
coefcient
k =
2
A/I (4.229)
and carrying out the differentiation in Eq. (4.228), we obtain
d
2
E
d
2
d
2
W
d
2
+2
dE
d
d
3
W
d
3
+E
d
4
W
d
4
kL
4
()W = 0 (4.230)
The objective of this section is to determine the polynomial function W() that
satises the boundary conditions of simple support W = W
= 0 at = 0 and
= 1, and has an arbitrarily prescribed node at = (0 < < 1), and serves
coincide with the oscillator chain location or differ from it. In the former
EOIS: 2892_chap04 2004/9/27 22:56 page 181 #47
as the mode shape satisfying Eq. (4.230) with polynomial variations of E()
and (). The mode shape satisfying the conditions
W(0) = W
(0) = W(1) = W
(1) = W() = 0 (4.231)

will be sought for the fth-order polynomial in terms of :
W() =
0
+
1
+
2
2
+
3
3
+
4
4
+
5
5
(4.232)
Satisfying the boundary conditions (4.232) yields the following set of
equations:
0
= 0
2
= 0
0
+
1
+
2
+
3
+
4
+
5
= 0
2
2
+6
3
+12
4
+20
5
= 0
0
+
1
+
2
2
+
3
3
+
4
4
+
5
5
= 0
(4.233)
We have ve equations with six unknowns. One of the unknowns, namely
1
, is set equal to unity, resulting in ve inhomogeneous equations in ve
unknowns. The determinant of the system equals
1 0 0 0 0
0 1 0 0 0
1 1 1 1 1
0 2 6 12 20
1
2
= 6
5
14
4
+8
3
= 2
3
( 1)(3 4) (4.234)
and vanishes at = 0, = 1 and =
4
3
. We exclude values = 0 and
= 1 from consideration, since these cases do not correspond to internal
nodes. Likewise =
4
3
is not included in the study, since it corresponds to
a node outside the beam. Hence, for = 0, = 1 and =
4
3
the above
inhomogeneous set has a unique solution, which leads to the following mode
shape:
W() = +
3
3
+
4
4
+
5
5
(4.235)
EOIS: 2892_chap04 2004/9/27 22:56 page 182 #48
where
3
= 2
_
3
3
2
2
2 2
_
/
4
=
_
3
3
+2
2
7 7
_
/
5
= 3
_
2
1
_
/ (4.236)
=
2
(3 4)
If we are given the material density in the polynomial form
() =
m
i=0
a
i
i
(4.237)
with upper limit m, then it is easily seen that the largest derivative in Eq. (4.230)
is of the fourth order, and the polynomial for E(), to be used in Eq. (4.234)
must have the upper limit m+4:
E() =
m+4
i=0
b
i
i
(4.238)
The required expressions in Eq. (4.230) are
W() = +
3
3
+
4
4
+
5
5
W
() = 6
3
+12
4
2
+20
5
3
W
() = 6
3
+24
4
+60
5
2
W
IV
() = 24
4
+120
5
() =
m+4
i=0
ib
i
i1
E
() =
m+4
i=0
i(i 1)b
i
i2
(4.239)
We will start the analysis with particular cases.
4.3.3 A Beam with Constant Mass Density
Let us rst take the case where the density is uniform, i.e. m = 0 so that
a
0
> 0 a
1
= a
2
= = a
m
= 0 (4.240)
EOIS: 2892_chap04 2004/9/27 22:56 page 183 #49
Substituting Eq. (4.238) and (4.239) into Eq. (4.230) we get the following
expressions for each term of the governing equation
d
2
E
d
2
d
2
W
d
2
=
_
2b
2
+6b
3
+12b
4
2
_ _
6
3
+12
4
2
+20
5
3
_
(4.241)
2
dE
d
d
3
W
d
3
= 2
_
b
1
+2b
2
+3b
3
2
+4b
4
3
_ _
6
3
+24
4
+60
5
2
_
(4.242)
E
d
4
W
d
4
=
_
b
0
+b
1
+b
2
2
+b
3
3
+b
4
4
_ _
24
4
+120
5
2
_
(4.243)
kL
4
() W () = kL
4
a
0
_
+
3
3
+
4
4
+
5
5
_
(4.244)
Setting the sum of Eqs. (4.241)(4.244) to zero and setting the coefcient of
the various power of equal to zero, we get a set of six linear algebraic
homogeneous equations for the six unknowns b
i,i=0...4
and k:
2
4
b
0
+
3
b
1
= 0
120
5
b
0
+72
4
b
1
+36
3
b
2
kL
4
a
0
= 0
240
5
b
1
+144
4
b
2
+72
3
b
3
= 0
400
5
b
2
+240
4
b
3
+120
3
b
4
kL
4
3
a
0
= 0
600
5
b
3
+360
4
b
4
kL
4
4
a
0
= 0
840
5
b
4
kL
4
5
a
0
= 0
(4.245)
In order to have a non-trivial solution the determinant of the matrix of the
set (4.245) must vanish, yielding the following equation:
265,420,800L
4
5
a
0
_
108
4
4
675
3
2
4
5
4,375
3
5
+750
2
3
2
5
_
= 0 (4.246)
Substituting the expressions for
3
,
4
and
5
in terms of into Eq. (4.246), we
obtain
21,499,084,800L
4
a
0
(972
17
8,262
16
+15,498
15
+49,005
14
188,595
13
18,603
12
+688,293
11
370,557
10
1,214,845
9
+890,775
8
+1,261,667
7
839,737
6
846,457
5
+317,555
4
+333,335
3
8,668
2
46,032 7,672)/
_
12
(3 4)
6
_
= 0 (4.247)
The solution of this equation for within the interval [0; 1] is =
1
2
. Hence,
for this value of ,
3
= 10,
4
= 15 and
5
= 6. We conclude that for
EOIS: 2892_chap04 2004/9/27 22:56 page 184 #50
0
0 0.2 0.4 0.6 0.8 1
0.05
0.05
0.1
0.15
0.2
0.1
0.15
0.2
W
(
)
FIGURE 4.7
Second mode shape of the pinned beam with constant mass density
the case m = 0, there exists only a single fth-degree polynomial solution for
w(), namely
W() = 10
3
+15
4
6
5
(4.248)
which is depicted in Figure 4.7.
Hence, with the mode shape (4.248), we obtain the following expression for
k from the last equation in the set (4.245):
k =
840b
4
L
4
a
0
(4.249)
Note that the fundamental frequency coefcient is (Candan and
Elishakoff, 2000):
k =
360b
4
L
4
a
0
(4.250)
Solving for the b
i
(i = 0, 1, . . . , 4) in terms of b
4
, Eq. (4.238) for E() becomes
E() =
_
1
3
+ 2
3
+
4
_
b
4
(4.251)
EOIS: 2892_chap04 2004/9/27 22:56 page 185 #51
0.3
0 0.2 0.4 0.6 0.8 1
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
D
(
FIGURE 4.8
Variation of the stiffness of a pinned beam with constant mass density
The variation of the obtained exural rigidity is portrayed in Figure 4.8.
Thus, for the case m=0, there exists only a single fth-degree polynomial
solution for W() satisfying certain prescribed boundary conditions and hav-
ing a denite nodal position, which is generated by a unique elastic modulus
distribution of the fourth degree. It is remarkable that in the case of the uni-
form material density (m = 0) the variation of the elastic modulus does not
depend on a
0
.
4.3.4 A Beam with Linearly Varying Mass Density
In the case of a linear variation in the material density (), namely
() = a
0
+a
1
(4.252)
in Eq. (4.252) as follows:
() = a
0
(1 +) where = a
1
/a
0
= 0 (4.253)
consider rst the case when a
0
differs from zero. We rewrite the mass density
EOIS: 2892_chap04 2004/9/27 22:56 page 186 #52
The modulus of elasticity, according to Eq. (4.238), takes the form
E() =
5
i=0
b
i
i
(4.254)
Substitution of Eqs. (4.253) and (4.229) into Eq. (4.230) along with Eq. (4.235)
yields the polynomial equation
C
6
6
+C
5
5
+C
4
4
+C
3
3
+C
2
2
+C
1
+C
0
= 0 (4.255)
where
C
0
= 24
4
b
0
+12
3
b
1
C
1
= 120
5
b
0
+72
4
b
2
+36
3
b
3
kL
4
a
0
C
2
= 240
5
b
1
+144
4
b
2
+72
3
b
3
kL
4
a
0
C
3
= 400
5
b
2
+240
4
b
3
+120
3
b
4
kL
4
a
0
3
C
4
= 600
5
b
3
+360
4
b
4
+180
3
b
5
kL
4
a
0
(
4

3
)
C
5
= 840
5
b
4
+504
4
b
5
kL
4
a
0
(
5
+
4
)
C
6
= 1120
5
b
5
kL
4
a
0
5
(4.256)
In order for Eq. (4.256) to be valid for any , we demand all coefcients C
i
(i =
0,1,...,6)to vanish, leading to seven equations for the seven unknowns,
namely b
i
(i = 0, 1, . . . , 5) and k. Since this set is homogeneous, for non-
triviality of the solution, its determinant
=
24
4
12
3
0 0 0 0 0
120
5
72
4
36
3
0 0 0 L
4
a
0
0 240
5
144
4
72L
3
0 0 L
4
a
0
0 0 400
5
240
4
120
3
0 L
4
a
0
3
0 0 0 600
5
360
4
180
3
L
4
a
0
(
4
+
3
)
0 0 0 0 840
5
504
4
L
4
a
0
(
5
+
4
)
0 0 0 0 0 1120
5
L
4
a
0
(4.257)
must vanish. This leads to the following equation
L
4
a
0
5
(G
1
+G
2
) = 0 (4.258)
EOIS: 2892_chap04 2004/9/27 22:56 page 187 #53
where
1
= 1,857,945,600
G
1
=
1
_
700,000
4
4
5
120,000
3
5
2
3
+108,000
2
5
3
4
3
17,280
5
5
4
_
G
2
=
1
_
175,000
4
5
3
210,000
3
5
2
4
17,625
3
5
3
3
+78,200
2
5
2
4
2
3
49,140
5
4
4
3
+7,128
6
4
_
(4.259)
Since
5
= 0 (i.e. the roots of
2
1 lie outside [0; 1] accordingtoEq. (4.236))
and it was assumed that a
0
differs from zero, the expression in parentheses
in (4.258) must vanish, leading to the following expression for :
= G
1
/G
2
(4.260)
The latter expr ession allows us to pose the following question: What is
the value of the ratio of the mass density coefcients corresponding to a
vibration node placed at an arbitrary location ?
The expressions for G
1
and G
2
in terms of read as follows:
G
1
= 20(972
17
8,262
16
15,498
15
+49,005
14
188,595
13
18,603
12
+688,293
11
370,557
10
1,214,845
9
+890,775
8
+1,261,667
7
839,737
6
846,457
5
+31,755
4
+333,335
3
8,668
2
46,032 7,672)
G
2
= 8,016
18
62,451
17
+102,384
16
+310,824
15
987,795
14
+6,129
13
+1,026,444
12
+3,368,349
11
4,463,541
10
8,101,030
9
+9,604,699
8
+9,339,499
7
8,927,306
6
7,068,461
5
+3,583,555
4
+3,150,739
3
225,591
2
511,026 85,171 (4.261)
When G
2
equals zero, the value of tends to innity, implying that either a
1
increases without limit or that a
0
tends to zero. The approximate value of
for this limiting case is
= 0.5571986621 (4.262)
The following question ar ises: Can take on other values besides =
1
for nite values of ? The reply must include the consideration of physical
EOIS: 2892_chap04 2004/9/27 22:56 page 188 #54
realizability of the problem, which demands that () must be positive when
[0; 1], i.e.
1 + 0 or 1 (4.263)
Therefore, the allowable region of variation of is
[
0
;
1
] (4.264)
where
0
denotes the value of , corresponding to = 1, namely
0
= 0.4428013379 (4.265)
Various mode shapes corresponding to in the allowable region are presented
in Figure 4.9.
Physical realizability also imposes the requirement that the elastic modulus
E() be positive. This latter propertycanbe checkedstraightforwardly; infact,
the exural rigidity coefcient b
i
that involve and
i
are functions of , since
and
i
depend on .
0.25
0.2 0 0.3 0.1 0.4 0.5 0.6 0.7 0.8 0.9 1
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
D
(
=
0
=
1
Allowable region of node's position
FIGURE 4.9
Various second mode shapes of the inhomogeneous pinned beam with linearly varying mass
density (() = a
0
(1 +))
EOIS: 2892_chap04 2004/9/27 22:56 page 189 #55
Consider in the allowable region. The system (4.256) has a non-trivial
solution. The second frequency reads as
k = 1120b
5
/L
4
a
1
(4.266)
The coefcients b
i
in the elastic modulus E() are derived fromthe set (4.256):
b
0
=
b
5
37,500
5
5
__
21,600
3
3
4
5
+23,625
2
3
2
5
+3,564
5
4
105,000
4
3
5
_
+
_
46,800
3
2
4
2
5
27,000
2
3
3
5
8,640
4
4
5
+350,000
5
__
b
1
=
b
5
7,500
4
5
__
1,188
4
4
6,210
3
2
4
5
+3,525
2
3
2
5
35,000
5
_
2, 880
3
4
5
+13,200
3
b
2
=
b
5
125
3
5
__
33
3
4
+145
3
5
_
+80
2
4
5
300
3
2
5
_
b
3
=
b
5
150
5
2
__
66
2
4
+235
3
5
_
+160
4
5
_
b
4
=
b
5
15
5
(11
4
+20
5
)
(4.267)
Consider separately the case a
0
= 0, () = a
1
. We again obtain seven
equations for seven unknowns. The requirement of its determinant to vanish
reads as
2
a
1
L
4
G
3
/G
4
= 0 (4.268)
where
2
= 3,611,846,246,40
G
3
= 8,019
20
70,470
19
+15,681
18
+270,891
17
14,011,003
16
+6,831,003
15
+2,008,110
14
+2,335,577
13
8,858,334
12
7,005,838
11
+22,169,270
10
+78,355,830
9
27,871,504
8
7,480,654
7
+19,579,322
6
+6,635,645
5
6,959,885
4
3,436,174
3
+651,446
2
+596,197 +85,171
G
4
=
14
(3 4)
(4.269)
EOIS: 2892_chap04 2004/9/27 22:56 page 190 #56
The solution of this equation yields =
1
= 0.55719, which matches the root
of G
2
in Eq. (4.262).
The frequency coefcient k is given by
k = 1120b
5
/L
4
a
1
(4.270)
We denote
3
,
4
,
5
the values of the mode shape coefcients for =
1
.
The coefcients in the expression for the exural rigidity read
b
0
=

4
b
5
12,500
5
5
_
7,200
3

2
4

5
+7,875
2
3

2
5
+1,188
4
4
35,000
3
5
_
= 0.1169104775b
5
b
1
=
b
5
7,500
4
5
_
1,188
4
4
6,210
3

2
4

5
+3,525
2
3

2
5
35,000
3
5
_
= 0.3434953999b
5
(4.271)
b
2
=

4
b
5
125
3
5
_
33
3
4
+145
3

5
_
= 0.7826410468b
5
b
3
=
b
5
150
2
5
_
66
2
4
+235
3

5
_
= 0.114362033b
5
b
4
=
11
4
b
5
15
5
= 1.852849279b
5
4.3.5 A Beam with Parabolically Varying Mass Density
Let us study now the parabolically varying material density case, namely
(x) = a
0
+a
1
+a
2
2
(4.272)
We study rst the case when a
0
differs from zero. We rewrite the mass density
in Eq. (4.272) as follows:
(x) = a
0
_
1 +
1
+
2
2
_

1
= a
1
/a
0

2
= a
2
/a
0
= 0 (4.273)
The elastic exural rigidity takes the form
E() =
6
i=0
b
i
i
(4.274)
Figure 4.10 depicts the variation of the exural rigidity in this case.
EOIS: 2892_chap04 2004/9/27 22:56 page 191 #57
0.1
0
0.15
0.2
0.25
0.3
0.35
0.4
0.2 0.4 0.6 0.8 1
0.45
D()
b
5
FIGURE 4.10
Variation of the exural rigidity of a pinned beam with linearly varying mass density () =
1
The governing differential equation (4.230) yields the following set of eight
equations for eight unknowns:
24
4
b
0
+12
3
b
1
= 0
120
5
b
0
+72
4
b
1
+36
3
b
2
+kL
4
a
0
= 0
240
5
b
1
+144
4
b
2
+72
3
b
3
kL
4
a
0
1
= 0
400
5
b
2
+240
4
b
3
+120
3
b
2
kL
4
a
0
(
3
+
2
) = 0
600
5
b
3
+360
4
b
4
+180
3
b
5
kL
4
a
0
(
4
+
3
1
) = 0
840
5
b
4
+504
4
b
5
+252
3
b
6
kL
4
a
0
(
5
+
4
1
+
3
2
) = 0
1120
5
b
5
+672
4
b
6
kL
4
a
0
(
5
1
+
4
2
) = 0
1440
5
b
6
kL
4
a
0
2
= 0
(4.275)
In order to nd a non-trivial solution, the determinant of the matrix of the
set (4.275) must vanish. This leads to the following determinantal equation:
L
4
a
0
5
(A
2
+B
1
+C) = 0 (4.276)
EOIS: 2892_chap04 2004/9/27 22:56 page 192 #58
where
A =
3
_
15,552
7
4
+118,260
5
5
4
3
+378,000
3
5
3
4
243,000
2
5
3
4
2
3
630,000
4
5
3
+120,375
5
2
3
_
B =
3
_
35,640
5
6
4
245,700
2
5
4
4
3
1,050,000
4
5
2
4
+391,500
2
5
3
4
2
3
+875,000
5
5
3
88,125
4
5
3
3
_
C =
3
_
86,400
2
5
5
4
+540,000
3
5
3
4
3
+3,500,000
5
5
4
600,000
4
5
2
3
_
3
= 535,088,332,800 (4.277)
Substituting Eq. (4.256) into Eq. (4.277), we get the following expression for
A, B and C in terms of :
A = G
5
G
6
B = G
7
G
8
C = G
8
G
9
(4.278)
where
G
5
= 5,832
18
53,703
17
+163,377
16
102,978
15
382,085
14
+711,387
13
634,768
12
+1,165,347
11
94,623
10
4,158,540
9
+2,937,847
8
+4,518,597
7
3,352,918
6
3,061,283
5
+1,432,465
4
+1,292,917
3
110,598
2
219,228 36,538 (4.279)
G
6
= 3
3
+3
2
7 7 (4.280)
G
7
= 5(8,019
18
62,451
17
+102,384
16
+310,824
15
987,795
14
+6,129
13
+1,026,444 +3,368,349
11
4,463,541
10
8,101,030
9
+9,604,699
8
+9,339,499
7
8,907,306
6
7,068,461
5
+3,583,555
4
+3,150,739
3
225,591
2
511,026 85,171) (4.281)
G
8
=
2
Ga 1 (4.282)
EOIS: 2892_chap04 2004/9/27 22:56 page 193 #59
G
9
= 5(19,440
17
165,240
16
+309,960
15
+980,100
14
+3,771,900
13
372,060
12
+13,765,860
11
7,411,140
10
24,296,900
9
+17,815,500
8
+25,233,340
7
+16,794,740
6
16,929,140
5
+6,351,100
4
+6,666,700
3
173,360
2
920,640 153,440)
(4.283)
Since
5
= 0 and a
0
was assumed to be non-zero, Eq. (4.276) becomes
A
2
+B
1
+C = 0 (4.284)
Solving Eq. (4.284) for
2
we get
2
= (B
1
+C) /A (4.285)
When A = 0 the value of
2
grows without bound, implying that
2
tends to
innity or that a
0
tends to zero. Consider
3
the value of that satises A = 0;
3
equals
3
= 0.6020659819 (4.286)
An approximation for the right limit
4
in the allowed interval (
3
,
4
) is
obtained as follows:
We know that the system (4.275) has a non-trivial solution when the rela-
tionship (4.285) between
1
,
2
2
in terms of
1
and , as given in Eq. (4.285) by the gray surface; the interval
of variation for
1
is taken as [20; 20].
Along with the relationship (4.285) one has also to take into account the
requirement of physical realizability of the problem, which imposes the con-
dition that both mass density and exural rigidity have to be non-negative for
any within the interval [0; 1]. At this stage the non-negativity of the mass
density can be checked, whereas this property for the exural rigidity can be
checked by direct numerical evaluation.
Consider the expression for the mass density given in Eq. (4.273). Since a
0
is the value taken by ()at =0, it must be positive. Thus, in order for the
mass density to be positive the following trinomial
()/a
0
= 1 +
1
+
2
2
(4.287)
has to be positive within the interval [0; 1]. Note that a trinomial can either
have real roots or not, and in each case it does not have the same sign. When
it has no real root (i.e. its discriminant is negative), it keeps the same sign for
all real numbers and its sign is that of the coefcient of the term
2
. When it
has real roots (i.e. its discriminant is positive) it shares the same sign as the
coefcient of the term
2
outside the interval dened by its roots.
and holds. Figure 4.11 shows the variation of
EOIS: 2892_chap04 2004/9/27 22:56 page 194 #60
0
10
20
30
40
50
20
20
10
10
20
0
20
40
60
80
100
0
2
=0
1
2
4
=
FIGURE 4.11
Variation of the parameter
2
given in Eq. (4.285) in terms of [0.61; 1] and
1
We demand the trinomial (4.287) to be positive for all real numbers, leading
to the following requirement
=
2
1
4
2
< 0
2
> 0 (4.288)
where is the discriminant of the trinomial (4.287). In addition to the surface
discriminant vanishes ( = 0).
The two surfaces intersect onthe curve that is showninbold. The observable
part of the surface that satises Eq. (4.285) is above the bold line; in this region
the discriminant is negative, while
2
> 0, implying that the mass density
is non-negative. The maximum value of on the bold curve is
4
0.66.
Thus, we arrive at the interval (
3
,
4
) = (0.602; 0.66) in which the node can
be placed so as to lead to a physically realizable problem.
Consider in an allowable region. The system(4.275) has a non-trivial solu-
tion. The second frequency coefcient is derived from the last equation of the
set (4.275):
k = 1440b
6
/L
4
2
(4.289)
representing Eq. (4.285), Figure 4.12 also depicts the surface on which the
EOIS: 2892_chap04 2004/9/27 22:56 page 195 #61
20
10
10
20
1
0.98
0.94
0.9
0.86
0.82
0.78
0.74
0.7
0.66
0.62
0
200
100
100
0
2
=0
1
2
4
=
Region where the
discriminant of the
trinomial (A) is negative
FIGURE 4.12
Variation of the parameter
2
given in Eq. (4.309) in terms of [0, 0.58] and
1
The coefcient b
i
in the elastic modulus E() reads as
b
0
=
3b
6
875,000
6
5
2
__
315,000
4
5
4
105,300
5
4
4
3
+166,050
2
5
2
4
2
3
37,125
3
5
3
3
378,000
3
5
3
4
+15,552
6
4
_
2
+
_
216,000
2
5
3
4
236,250
3
5
2
3
35,640
5
5
4
+1,050,000
4
5
4
_
1
46,800
3
2
4
2
5
27,000
2
3
3
5
8,640
4
4
5
+350,000
5
_
b
1
=
3b
6
875,000
5
5
2
__
630,000
3
5
4
+2,592
5
4
15,390
5
3
4
3
+16,650
2
5
2
3
_
2
+
_
31,050
2
5
4
4
3
5,940
5
4
4
17,625
3
5
2
3
+175,000
4
5
_
1
+14,400
2
5
3
4
66,000
3
5
3
_
b
2
=
9b
6
17,500
4
5
2
__
7,000
3
5
+288
4
4
1,470
5
2
4
3
+825
2
5
2
3
_

2
+
_
2,900
2
5
3
660
5
3
4
_
1
+1,600
2
5
2
4
6,000
3
5
3
_
EOIS: 2892_chap04 2004/9/27 22:56 page 196 #62
b
3
=
3b
6
1,750
3
5
2
__
144
3
4
615
4
3
_
2
+
_
1,175
2
5
3
330
5
2
4
_
1
+800
2
5
4
_
b
4
=
3b
6
350
2
5
2
__
48
2
4
165
5
3
_
2
110
5
1
200
2
5
_
b
5
=
3b
6
35
5
2
(8
4
2
15
5
1
)
(4.290)
Now, consider the case a
0
= 0, with () = a
1
+ a
2
2
. We rewrite the mass
density as follows:
() = a
1
(1 +) = a
2
/a
1
= 0 (4.291)
The parameter a
1
is assumed to be non-zero. We obtain eight equations for
eight unknowns:
24
4
b
0
+12
3
b
1
= 0
120
5
b
0
+72
4
b
1
+36
3
b
2
= 0
240
5
b
1
+144
4
b
2
+72
3
b
3
kL
4
a
0
= 0
400
5
b
2
+240
4
b
3
+120
3
b
2
kL
4
a
1
= 0
600
5
b
3
+360
4
b
4
+180
3
b
5
kL
4
a
1
3
= 0
840
5
b
4
+504
4
b
5
+252
3
b
6
kL
4
a
1
(
4
+
3
) = 0
1120
5
b
5
+672
4
b
6
kL
4
a
1
(
5
+
4
) = 0
1440
5
b
6
kL
4
a
1
5
= 0
(4.292)
The requirement that the determinant of the system (4.292) should vanish
reads as
A +B = 0 (4.293)
with A and B dened in Eq. (4.277). The solution of this equation for yields
the following expression
= B/A (4.294)
Note that when the denominator in Eq. (4.294) equals zero, the value of tends
to innity, implying that a
1
tends to zero, or that a
2
grows without bound. The
corresponding node location , which is a solution of the equation A = 0, is
3
given in Eq. (4.286).
EOIS: 2892_chap04 2004/9/27 22:56 page 197 #63
Moreover, the requirement of positivity of the mass density, demands
1 + 0 or 1 (4.295)
Hence, the allowable region of variation of is
[
2
,
3
] (4.296)
where
2
corresponding to the value of when = 1, reads as
2
= 0.5 (4.297)
Various mode shapes corresponding to in the allowable region are presented
in Figure 4.13.
The requirement of physical realizability also imposes the condition that the
elastic modulus E() be positive. This latter property can be checked straight-
forwardly; in fact, the exural rigidity coefcients b
i
that involve and
i
are
functions of , since and
i
depend upon .
0.44
0
0.52
0.5
0.48
0.46
100 200 300 400 500
0.54
FIGURE 4.13
Variation of the node position in terms of the parameter
EOIS: 2892_chap04 2004/9/27 22:56 page 198 #64
Consider in the allowable region. The system (4.292) has a non-trivial
solution. The second frequency is derived from the last equation of the
set (4.292):
k = 1440b
6
/L
4
a
1
(4.298)
The coefcients b
i
in the elastic modulus E() are given in Eq. (4.290) with
2
= .
Consider, nally, the case a
0
= 0 and a
1
= 0. The mass density, then, equals
() = a
2
2
(4.299)
Again we obtain eight equations for eight unknowns:
24
4
b
0
+12
3
b
1
= 0
120
5
b
0
+72
4
b
1
+36
3
b
2
= 0
240
5
b
1
+144
4
b
2
+72
3
b
3
= 0
400
5
b
2
+240
4
b
3
+120
3
b
2
kL
4
a
2
= 0
600
5
b
3
+360
4
b
4
+180
3
b
5
= 0
840
5
b
4
+504
4
b
5
+252
3
b
6
kL
4
a
2
3
= 0
1120
5
b
5
+672
4
b
6
kL
4
a
2
4
= 0
1440
5
b
6
kL
4
a
2
5
= 0
(4.300)
The requirement for the determinant of (4.300) to vanish reads
3
L
2
a
2
G
10
/G
11
= 0 (4.301)
where
3
= 3,120,635,156,889,600
G
10
= 17,496
23
161,109
22
+414,315
21
+371,709
20
3,027,699
19
+227,126
18
+5,284,392
17
4,903,488
16
237,482
15
18,436,704
14
+2,640,556
13
+79,775,162
12
28,942,438
11
135,634,072
10
+46,346,616
9
+13,757,347
8
25,073,663
7
88,091,388
6
3,925,062
5
+3,130,985
4
+8,634,725
3
+3,733,764
2
2,046,128 255,766
G
11
=
16
(3 4)
8
(4.302)
EOIS: 2892_chap04 2004/9/27 22:56 page 199 #65
The denominator of the determinant is zero when =0, which is one end
of the beam or =
4
3
, which is a node outside the beam. The solution of the
determinantal equation (4.301) is =
3
given in Eq. (4.286). Denoting by
3
,

4
and
5
the value taken by the coefcients
i
in this case, the mode shape
becomes
W() = +
3
3
+
4
4
+
5
5
(4.303)
This yields the following solutions for the second frequency and the coef-
cients b
i
in the exural rigidity:
k = 1440b
6
/L
4
a
2
(4.304)
b
0
=
27b
6
875,000
6
5
_
11,700
3

4
4

5
18,450
2
3

2
4

2
5
+4,125
3
3

3
5
35,000
3

4
5
1,728
6
4
+42,000
2
4
_
0.4899b
6
b
1
=
27
4
b
6
43,750
5
5
_
144
4
4
855
3

2
4

5
+925
2
3

2
5
3,500
3
5
_
0.1417b
6
b
2
=
9b
6
17,500
4
5
_
288
4
4
1,470
3

2
4

5
+825
2
3

2
5
7,000
3
5
_
0.3172b
6
b
3
=
9
4
b
6
1750
3
5
_
205
3

5
48
2
4
_
0.6495b
6
b
4
=
9b
6
350
2
5
_
55
3

5
16
2
4
_
0.1788b
6
b
5
=
24
4
b
6
35
5
1.7466b
6
(4.305)
general variation of the mass density is considered in Appendix B.
4.4 Concluding Remarks
In this section for three sets of boundary conditions the closed-form solu-
tions have been derived for the mode shapes and natural frequencies of
inhomogeneous beams. The following conclusions have been reached.
Figure 4.14 depicts the exural rigidity obtained in this case. The case of the
EOIS: 2892_chap04 2004/9/27 22:56 page 200 #66
0.2
0.15
0.05
0.05
0.1
0.15
0.2
0
0.2 0.4
Allowable region of nodes position
0.6 0.8 1
0
0.1
=
2
=
3
FIGURE 4.14
Various second mode shapes of the inhomogeneous pinned beam with parabolically varying
mass density (() = a
1
(1 +))
Inhomogeneous beams may possess the natural mode that is coincident
with the static deection of the associated uniform beam under a uniformly
distributed load.
The fundamental frequencies in all three cases coincide with each other,
as a comparison of Eqs. (4.181), (4.204) and (4.227) reveals. The remain-
ing case of the inhomogeneous beam that is pinned at both its ends was
studied by Candan and Elishakoff (2000). There too, the beam turned
out to possess the fundamental frequency given in Eqs. (4.181), (4.204)
and (4.227).
Although the expressions for fundamental frequencies of inhomogeneous
beams with four different boundary conditions coalesce, the beams char-
acteristics in each case are different. Namely, although they share the same
material density variation as in Eq. (4.2), the b
i
coefcients in the elastic
modulus variation differ. This leads to the interesting conclusion that the
beams with different elastic modulus variation may have the same natural
frequency, although the beams are under differing boundary conditions.
This conclusion may at rst glance appear to be counterintuitive. Indeed,
if one anticipates that the fundamental frequency of the clampedclamped
beam must be greater than its counterpart for the beam that is pinned
at both its ends. Yet, it must be borne in mind that in the cases that
we have considered the beams characteristics are different: The clamped
clamped beam and the pinned beam have different expressions for the elastic
modulus.
EOIS: 2892_chap04 2004/9/27 22:56 page 201 #67
Theintricateconnectionof thesubject of this studywiththeinverseproblems
should be mentioned. As Gladwell (1996) stressed,
classical direct problems have involved the analysis and derivation of the
behavior of the system(e.g., forced response, natural frequencies, current
ow, stresses, etc) from its properties such as density or mass, conduct-
ivity, elastic constant, crack lengths, etc. Inverse problem are concerned
with the determination or estimation of such properties from behavior.
It turns out that although each of the beams has different boundary condi-
tions, and, moreover, each of them has different D() these beams have the
same frequency. Two vibrating system which have the same natural frequen-
cies are called isospectral. In our particular case, beams of different boundary
conditions share the rst natural frequency. Gottlieb (1991), Driscoll (1997)
and others have constructed examples of isospectral structures. In particu-
lar, Gottlieb (1991) showed that clamped inhomogeneous circular plates have
the same vibration spectrum as their homogeneous counterparts. In our
cases, the second and other frequencies do not coincide. For example, the
second natural frequency squared of the pinned beam is 10881.18bI/(AL
4
),
while the clamped-clamped beam has a second natural frequency squared
5607.68I/(AL
4
). Clamped-free and clamped-pinned beams second natural
frequencies squared are respectively 42727.97I/(AL
4
) and 8013.24I/(AL
4
).
These values are obtained by the nite element method. The difference
between the present work and those associated with the inverse vibration
problem lies in our desire to obtain closed-form solutions to nd any beam
that has a polynomial mode shape.
The expressions for the squaredfundamental frequency dependsolely upon
two coefcients a
m
and b
m+4
. If, by any procedure, these coefcients could be
xed by the designer, one can have a beam that has a pre-selected fundamental
natural frequency so that the unwanted resonance condition can be avoided.
Whereas we are unaware of a procedure with such a derivable feature at
present, its possible development in the future cannot be a priori ruled out.
If the coefcients a
m
and b
m+4
are deterministic but the remaining coef-
cients are random, the natural fundamental frequency squared is a determ-
inistic quantity. In complex structures that must be analyzed by approximate
methods this remarkable phenomenon could be validated if the coefcient
of variation of the output quantity turns out to be much smaller than its
counterparts for the input parameter.

Unusual Closed-Form Solutions For Beam Vibrations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Unusual Closed-Form Solutions For Beam Vibrations

Uploaded by

Copyright:

Available Formats

EOIS: 2892_chap04 2004/9/27 22:56 page 135 #1

(0) = 0. Hence, we postulate that b

1089 = 0.049. It lists mean

(1) = W() = 0 (4.231)

You might also like