Structural Engineering and Mechanics, Vol. 20, No.

3 (2005) 000-000 1

Free vibration analysis of rotating beams

with random properties
S. A. A. Hosseini† and S. E. Khadem‡
Department of Mechanical Engineering, Tarbiat Modarres University, P.O. Box 14115-177, Tehran, Iran

(Received June 22, 2004, Accepted March 31, 2005)

Abstract. In this paper, free vibration of rotating beam with random properties is studied. The cross-
sectional area, elasticity modulus, moment of inertia, shear modulus and density are modeled as random
fields and the rotational speed as a random variable. To study uncertainty, stochastic finite element
method based on second order perturbation method is applied. To discretize random fields, the three
methods of midpoint, interpolation and local average are applied and compared. The effects of rotational
speed, setting angle, random property variances, discretization scheme, number of elements, correlation of
random fields, correlation function form and correlation length on Coefficient of variation (C.O.V.) of first
mode eigenvalue are investigated completely. To determine the significant random properties on the
variation of first mode eigenvalue the sensitivity analysis is performed. The results are studied for both
Timoshenko and Bernoulli-Euler rotating beam. It is shown that the C.O.V. of first mode eigenvalue of
Timoshenko and Bernoulli-Euler rotating beams are approximately identical. Also, compared to
uncorrelated random fields, the correlated case has larger C.O.V. value. Another important result is, where
correlation length is small, the convergence rate is lower and more number of elements are necessary for
convergence of final response.
Key words: rotating beam; random properties; random fields; stochastic finite element method; second
order perturbation.

1. Introduction
The vibration analysis of rotating beam is important in understanding behavior of rotating
structures such as propellers, turbine blades, helicopter rotors and satellite booms. To study vibration
of rotating beams, Abbas (1986) and Yokoyama (1988) used finite element methods. Lee and Lin
(1994) studied the vibration of non-uniform rotating beam by neglecting the Coriolis effect. Yang
and Tsao (1997) studied the vibration and stability of a pretwisted blade under nonconstant rotating
speed. Pohit et al. (1999) investigated free out-of-plane vibration of a rotating beam with a
nonlinear elastomeric constraint. Al-Bedoore and Hamdan (2001) developed a mathematical method
for a rotating flexible arm undergoing large planar flexural deformations. Bazoune et al. (2001) used
finite element method to study the dynamic response of spinning tapered Timoshenko beams.
Vibration of a rotating damped blade with an elastically restrained root was investigated by Lin et al.
† PhD Student
‡ Professor, Corresponding author, E-mail: khadem@modares.ac.ir
2 S. A. A. Hosseini and S. E. Khadem

(2004). The comparison between Bernoulli-Euler and Timoshenko beam theories in a composite
rotating beam is performed by Jung et al. (2001) and Chandiramani et al. (2003).
Aforementioned studies are limited to vibration of rotating beams with deterministic properties.
However, in reality, properties of structures and mechanical systems are random. The first study in
vibration analysis of a system with random properties was carried out by Boice and Goodman
(1964) using perturbation method. Singh (1985) studied the turbine blade reliability with random
properties. Iwan and Haung (1996) presented a procedure for obtaining the dynamic response of
nonlinear system with parameter uncertainty. Chakraborty and Dey (1998) proposed a stochastic
finite element method in the frequency domain for analysis of structural dynamic problems
involving uncertain parameters. Ishida (2001) studied the eigenvalue problem of uniform and
optimum beam with uncertain cross-sectional area. Lin (2000) evaluated the influence of the
random parameter changes to the dynamic behavior of rotating Timoshenko beam. Also, Lin (2001)
investigated the probabilistic behavior of rotating Timoshenko beam. The stochastic finite element
method (SFEM) was developed for analysis of a system with random properties by Combou (1975)
and Handa and Anderson (1981). An important procedure in SFEM is to discretize random fields,
e.g. midpoint method (Handa and Anderson 1981), local average method (Vanmarcke and Grigoriu
1983), and interpolation method (Liu et al. 1986).
In this paper, free vibration of rotating beam with random properties is studied. This work is
based on the Lin’s paper (2000, 2001). He considered the free vibration of rotating beam with
uncertain material, geometry properties and rotating speed. In his papers, these parameters are
modeled as random variables, and used analytical methods for investigation of the free vibration of
a rotating beam. In our work the cross-sectional area, elasticity modulus, moment of inertia, shear
modulus and density are modeled as random fields and the rotational speed as a random variable.
To study uncertainty, stochastic finite element method based on second order perturbation method is
applied. To discretize random fields, the three methods of midpoint, interpolation and local average
are applied and compared.

Fig. 1 Timoshenko rotating beam schematic

 Free vibration analysis of rotating beams with random properties 3

2. Finite element formulation of Timoshenko rotating beam

Fig. 1 shows a uniform beam with hub radius R0 and length L, rotating about the Z-axis with
constant angular velocity Ω. The origin of X, Y, Z coordinate axes is at the hub center. The origin of
x, y, z coordinate axes is at the hub periphery connected to the neutral axis. The x-axis is along the
beam length and coincident with the X-axis, the y-axis is along the beam width and the z-axis is
along the thickness. The beam midplane makes angle Ψ with the rotational plane called setting
angle. One may divide the beam into m elements (l = L/m) as shown in Fig. 1. Using the
Timoshenko beam theory, strain energy in element level is:
U e = 1--- ∫ l EI ⎛⎝ -------
∂ϕ-⎞ dx ′ + 1--- l KGA ⎛ -------
∂w ⎞ 1 l ⎛ ∂ w-⎞ dx ′
⎝ ∂ x-′ – ϕ⎠ dx ′ + --2- ∫ Nx ⎝ -------
2 2 2

(1)
2 ∂ x ′⎠
0 2∫ 0 ∂ x ′⎠ 0
′

The first, second and third terms are bending, shear and centrifugal strain energies, respectively.
The centrifugal force N is calculated as follows:
x'

N x = A ∫xl ρ Ω ( R + nl + x ′ ) dx ′ + A ∫ mln + l ρ Ω ( R + x ) dx =
′
′
2

0
( 1)
2

ρ A Ω l ----- ( m – n ) + 1--- ( m – n ) – ⎛⎝ ----- + n⎞⎠ ⎛⎝ x----′⎞⎠ – 1--- ⎛⎝ x----′⎞⎠

R R 2
2 2 0 2 2 0
(2)
l 2 l l 2 l
and the kinetic energy is calculated as:
T e = 1--- ∫ l ρ A ⎛⎝ ∂------
w-⎞ dx ′ + 1--- l ρ I ⎛ ∂ϕ
------⎞ dx ′
2 2

(3)
2 ∂ t⎠ 2∫ ⎝ ∂ t⎠
0 0

due to translational and rotational kinetic energy contributions.

Due to centrifugal force F that effects in Z direction, the potential energy is obtained as:
z'

W e = 1--- ∫ l F z wA dx ′ (4)
2 0
′

where F = ρ Ω2wsin2ψ. Using two-node elements with 8 degrees of freedom for beam discretization,
z'

the transverse displacement w and bending rotation ϕ in each element length are approximated by
Hermite shape functions with respect to nodal displacements:
w = { f , 0, f , 0, f , 0, f , 0 } { q } = { N w } T { q }
1 3 5 7

ϕ = { 0, f , 0, f , 0, f , 0, f } { q } = { Nϕ }T { q }
2 4 6 8
(5)
{ q }T = { wi, ϕi, wi′, ϕi′, wj, ϕj, wj′, ϕj′ }
where ( ' ) shows derivative with respect to spatial variable and
f = f = 1 – 3( ξ) + 2ξ f = f = ξ – 2ξ + ξ 2 3 3 3

1 2 3 4
(6)
f = f = 3ξ – 2ξ
5 6
f = f = –ξ +ξ
2 3

7 8
2 3

where ζ = x----′ . Hence, the stiffness and mass matrices are obtained:
l
4 S. A. A. Hosseini and S. E. Khadem

[ Kb ]e = ∫ l { Bb } EI { Bb } T dx ′
0

[ Ks ]e = ∫ l { Bs } KGA { Bs }T dx ′
0

[ Kc ]e = ∫ l Nx { Bw } { Bw }T dx ′
0
′
(7)
[ Mt ]e = ∫ l { Nw } ρ A { Nw }T dx ′
0

[ Mr ]e = ∫ l { Nϕ } ρ I { Nϕ }T dx ′
0

Where
{ Bb } ∂ -{N }
= ------- and ∂ {N } – {N } = {B } – {N }
{ Bs } = -------
∂ x′ ϕ
∂x′ w ϕ w ϕ

Using the usual assemblage of matrices, one may obtain stiffness and mass matrices. Finally,
using the Hamilton principle, the following equation is obtained for vibration of rotating beam:
( [ K ] – Ω sin ψ [ Mt ] ) { q } + [ M ] { q·· } = { 0 }
0
2 2
(8)
Where
[ M ] = [ Mr ] + [ Mt ] and [ K ] = [ Kb ] + [ Ks ] + [ Kc ] 0

For a harmonic motion, the following eigenvalue equation is obtained:

( [ K ] – λ [ M ] ) { q }* = { 0 } (9)
Where [K] = [K ] – Ω sin ψ [Mt ] , and λ is the eigenvalue of rotating beam and {q}* is the
0
2 2

vibration amplitude.

3. Free vibration of rotating beam with random properties

In this paper, we assume that the rotating beam has uncertainty in its properties, i.e. elasticity
modulus, shear modulus, moment of inertia and density which are modeled as random fields and
the rotational speed as a random variable. If α , i = 1 ... 6 is a random field then using α =
i i

α i (1 + ε ) where α i is expectation of α , the random field ε can be obtained with zero expectation.
i i i

To consider for uncertainty in Eq. (9), the second-order perturbation method is applied. In this
method, all random elements of an equation are expanded about expectation of random variables
using the Taylor series up to second order terms. If this method is applied to Eq. (9), one gets:
[ M ] = [ M ] + [ M ] p εp + 1--2- [ M ] ps εp εs
0 , ,

[ K ] = [ K ] + [ K ] p εp + 1--2- [ K ] ps εp εs
0 , ,

p, s = 1 … N (10)
[ q ] = [ q ] + [ q ] p εp + 1--2- [ q ] ps εp εs
0 , ,

[ λ ] = [ λ ] + [ λ ] p εp + 1--2- [ λ ] ps εp εs
0 , ,
 Free vibration analysis of rotating beams with random properties 5

where N is total number of random variables. Substituting Eq. (10) in Eq. (9):
{[K] – λ [M] }{q} = {0}
0 0 0 0

{[K] 0
– λ [M]
0
}{q} p = ([K] p – {λ} p[M] – {λ} [M] p){q}
0 , , , 0 0 , 0

(11)
{[K] – λ [ M ] }{q } = ( [ K] – { λ } [ M ] – 2λ [ M ] – λ [ M] ){ q }
0 0 0 ps ps
, ps p s ps , , 0 , , 0 , 0

+ 2( [ K ] – λ [ M ] – λ [ λ ] ){ q }
p, p , p0 s 0 , ,

Note here, the summation law for repeated indices is applied. The comma shows partial
derivative, and [·]0 and {·}0 show deterministic value of matrix and vector. Eqs. (11) should be
solved successively to obtain the final response.

4. Calculation of first and second derivative of eigenvalues and eigenvectors

It is seen from Eqs. (11) that to solve the above equations, the first and second derivatives of
eigenvalues and eigenvectors should be calculated. By applying eigenvalue properties, the following
equations can be obtained (Kleiber and Hien 1992):
{ q } T ( [ K ] p – λ [ M ] p ) { q }-
0 ,

λ p = -------------------------------------------------------------------
,
0 , 0

(12)
{q} [M] {q}
T 0 0 0

{q} T([K] ps – 2λ p[M] s – λ [M] ps){q}- +

0 , , ,

λ ps = --------------------------------------------------------------------------------------------
,
0 , 0

{q} T[ M] {q} 0 0 0

{ q } T ( [ K ] p – λ [ M ] p – λ p [ M ] ) { q } -s
2-------------------------------------------------------------------------------------------------
0 , 0 , , 0 ,

(13)
{q} T[M ] {q} 0 0 0

Superscript “T ” shows transpose. It is necessary to note that Lin (2000, 2001) used the finite
difference method to compute the derivatives of eigenvalues. Two methods are proposed for
calculation of derivatives of eigenvectors (Kleiber and Hien 1992). In the first method, the
derivative of the eigenvector is obtained from linear combination of all eigenvectors. In the second
method, the derivative of the eigenvector is obtained by additional relations found by derivation of
orthogonal relations with respect to random variables. If the second method is applied, we have:
{q} p ,
= –( [ A ] [ A ]
T –1
[ A ]T [ B ] ) { q } 0

where
[A] =
[K] – λ [M] ,
0 0 0

[B] =
[K] p – λ p[M] – λ [M] p
, , 0 0 ,

(14)
2 { q } oT [ M ] 0
{ q }oT [ M ] p ,

Upon application of expectation and variance operators to Eq. (10) and using only the first two
moments of random variables, one gets:
E [ λ ] = λ + 1--- λ ps E [ ε p ε s ] 0 ,

2 (15)
Var [ λ ] = λ p λ s E [ ε p ε s ] , ,
6 S. A. A. Hosseini and S. E. Khadem

Using the above equations, one may calculate expectation, variance and coefficient of variation
(C.O.V.) of first eigenvalue of rotating beam.

5. Sensitivity analysis of rotating beam

The purpose of sensitivity analysis is to determine that with variation of a parameter, how much
variation one gets in response. One of the important advantages of sensitivity analysis is to
determine the parameters, which have the most effects on the response characteristics. On the other
hand, this analysis determines the random parameters that have the least effects on deviation of
response; thus one can assume them as deterministic parameters.
In this paper following the sensitivity analysis of Lin (2000, 2001), the effect of uncertainty of
random variables on response deviation is represented by the following equation:
∂ Sλ Sε
ηi = --------
- -----i
(16)
∂ Sε Sλ
i

Here Sε and Sλ are standard deviation of response and that of random variables respectively. The
parameter η is defined as relative sensitivity factor. If Eq. (15) is used, η will be equal to:
i

i i

S2

η i = ( λ i ) ------ε- (17)
, 2
i

Sλ
2

6. Discussions
In this paper, shear modulus, G(x), cross-sectional area, A(x), elasticity modulus, E(x), moment of
inertia, I(x), and density, ρ (x), are modeled as random fields and rotational speed, Ω , is modeled as
a random variable. It is necessary to note that Lin (2000, 2001) used the parameters “T ” and “H ”
(thickness and width of the cross section) instead of “I(x)” and “A(x)”. To discretize random fields,
three methods of midpoint, interpolation and local average are applied and results are calculated
using these methods. Here, the following property values are used (Lin 2001):
E [ ρ ( x ) ] = 7830 ( kg/m ), E [ E ( x ) ] = 165.44 ( Gpa ), E [ G ( x ) ] = 63.44 ( Gpa )
2

E [ A ( x ) ] = 6 × 10– ( m ) , E [ I ( x ) ] = 8 × 10– ( m ) , L = 0.8 ( m )

4 2 8 4

R = 0.08 ( m ) , K = 0.85
0

Where E[.] is expectation operator and the correlation function of random fields is defined as:
ρ ( r ) = exp ⎛⎝ – a-------r-⎞⎠ (18)
L
Where r is the length between two arbitrary points of the beam and “a” is a constant. The ratio L/a
is called “correlation length”. If a = 0, two points in random field are fully correlated and if a = ∞,
the two points will be fully uncorrelated. Variance vector and correlation function matrix of random
fields are defined as:
 Free vibration analysis of rotating beams with random properties 7

⎧ σε ⎫
⎪ ⎪
2

⎪ σε ⎪ ρ (r) ρ (r) ρ (r) ρ (r) ρ (r)

⎪ ⎪
11 12 13 14 15

(r) (r) (r) (r) (r)

2

ρ ρ ρ ρ ρ
⎪ ⎪,
2

{ Var } [ρ(r)]
21 22 23 24 25

= ⎨ σε ⎬
2
= ρ (r) ρ (r) ρ (r) ρ (r) ρ (r) (19)
⎪ ⎪3 31

(r)
32

(r)
33

(r)
34

(r)
35

(r)
⎪ σε ⎪ ρ ρ ρ ρ ρ
⎪ ⎪
2
41 42 43 44 45

(r) (r) (r) (r) (r)

4

⎪ σε ⎪ ρ ρ ρ ρ ρ
⎩ ⎭
2 51 52 53 54 55

Subscripts 1, 2, 3, 4, 5 show ρ ( x ), I ( x ), E ( x ), A ( x ), G ( x ) respectively. Correlation matrix of

random fields is obtained as:
[R(r)] = { Var } [ ρ ( r ) ] { Var }
Rotational speed is defined as random variable α6.
In Figs. 2-5 coefficient of variation (C.O.V.) of first mode eigenvalue is plotted versus rotational
speed. The variance of all random fields equals to 0.01 and random fields are assumed independent
and a = +∞, so:
⎧ 0.01 ⎫ A 0 0 0 0
⎪ 0.01 ⎪ 0 B 0 0 0
{ Var } = ⎪⎨ 0.01 ⎪⎬, [ ρ ( r ) ] = 0 0 C 0 0 , Var ( Ω ) = 0.01 F (20)
⎪ 0.01 ⎪
⎪ ⎪ 0 0 0 D 0
⎩ 0.01 ⎭ 0 0 0 0 E
In these figurers, four curves are plotted:
Curve (a): all properties are assumed random (A = B = C = D = E = F = 1)
Curve (b): only material properties are assumed random (A = C = E = 1 and the rest are zero)
Curve (c): only geometry properties are assumed random (B = D = 1 and the rest are zero),
Curve (d): only rotational speed is assumed random (F = 1 and the rest are zero).
The accuracy of results is investigated by Monte-Carlo method. Simulation is carried out by
10000 samples and is compared with second-order perturbation method. It is realized that there is a
good agreement between second-order perturbation method and Monte-Carlo method, which
exhibits accuracy of results for all rotational speed range.
Fig. 2 shows C.O.V. for Timoshenko type, where the setting angle is equal to zero. It is seen that
if rotational speed is random (curve (d)), C.O.V. curve is ascending with respect to rotational speed
and if material (curve (b)) and geometry (curve (c)) properties are random, C.O.V. curves are
descending. If all properties are random (curve (a)), the C.O.V. curves are descending due to having
random of material and geometry and then due to rotational speed they are ascending. If rotational
speed increases, the effect of randomness of this parameter (Ω) increases, and consequently, the
C.O.V. curve of this paraeter will be ascending. On the other hand, if rotational speed increases, the
mean value of the first mode eigenvalue increases. But the variance of eigenvalue remains constant
(the variances of geometrical and material random properties are constant) and consequently the
C.O.V (variance over mean value) curves of material and geometry will be decreasing with respect
8 S. A. A. Hosseini and S. E. Khadem

Fig. 2 C.O.V. of first eigenvalue of Timoshenko rotating beam versus rotating speed, setting angle is 0 degree

Fig. 3 C.O.V. of first eigenvalue of Timoshenko rotating beam versus rotating speed, setting angle is 90
degree

to rotational speed. The next point to consider is the close similarity between the curves of
geometry and material. To justify the similarity between curves (b) and (c) in Fig. 2, we simply
assume Bernoulli-Euler rotating beam with zero rotational speed for which (Eqs. (8) and (9)):
[ Kb ] { q } *
= λ [ Mt ] { q } *
(21)
 Free vibration analysis of rotating beams with random properties 9

By factoring “EI ” and “ρA” in Eq. (7) we have

[ Kb ] = EI [ K a ] , [ Mt ] = ρ A [ Ma ] (22)
Then one can rewrite Eq. (21) as
EI [ K a ] { q } = λρ A [ M a ] { q }
* *
(23)
Above equation shows that the “order of magnitude” of geometrical properties (A, I ) and material
properties (ρ, E ) are the same. So, with having identical variation in geometrical and material
properties, then one will have identical variation in eigenvalues. This leads to the same C.O.V.
curves.
The properties of Figs. 3 and 2 are similar except that the setting angle in Fig. 3 is assumed to be
90 degree. It is seen from Fig. 3 that with the increase of setting angle, curves become smoother
and the randomness effect of rotational speed decreases. To justify this, we rewrite the centrifugal
stiffness matrix (Eqs. (2) and (7)) as:
[ Kc ] = Ω [ Kγ ]2
(24)
then Eq. (8) will be
( [ Kb ] + [ Ks ] + Ω [ Kγ ] – sin ψ [ Mt ]) ) { q }
2 2
*
= λ[M]{q} *
(25)
Since the translational mass matrix [M ] is a positive definite matrix and the expression
( [ Kγ ] – sin ψ [ Mt ] ) is multiplied by Ω2, if Ψ = 0, then the variation in rotating speed has the largest
t
2

effect on variation in the first mode eigenvalue.

Fig. 4 C.O.V. of first eigenvalue of Bernoulli-Euler rotating beam versus rotating speed, setting angle is 0
degree
10 S. A. A. Hosseini and S. E. Khadem

Fig. 5 C.O.V. of first eigenvalue of Bernoulli-Euler rotating beam versus rotating speed, setting angle is 90
degree

Figs. 4 and 5 show curves of Bernoulli-Euler rotating beam. It can be seen that aforementioned
figures are Similar to corresponding Figs. 2-3 for the Timoshenko beam. Consequently, C.O.V. of
first mode eigenvalue of Timoshenko and Bernoulli-Euler rotating beams are approximately
identical. The reason is the first mode eigenvalue of Timoshenko and Bernoulli-Euler rotating beam
are nearly identical and consequently the variation of corresponding eigenvalues and finally their
C.O.V. curves are nearly same.
Figs. 6 and 7 show the C.O.V. of rotating Timoshenko beam versus rotational speed with different
variances. All the curves are plotted by assuming that all properties are random and variances are
identical and equal to 0.0025, 0.01, 0.0225, 0.04. Fig. 6 is plotted with setting angle equal to zero.
Monte-Carlo simulation with 10000 samples is plotted to be compared with second-order
perturbation method. It is seen that, agreement is very good. Also, it is realized that if variance of
random properties increases, the C.O.V. increases accordingly. The properties of Fig. 7 are similar to
those of Fig. 6 except that the setting angle is assumed to be 90 degree. Theses curves are smoother
than the curves of Fig. 6. To justify this, Using Eq. (25), since the translational mass matrix [M ] is
a positive definite matrix, and the expression ( [ Kγ ] – sin ψ [ Mt ] ) is multiplied by Ω2, when Ψ = 0,
t
2

the variation in rotating speed has the largest effect on variation in eigenvalue which creates sharp
changes in the curve pattern, while when Ψ = 90, this effect is the smallest, so, the curve pattern
will be smoother.
Fig. 8 shows the C.O.V. of first mode eigenvalue as a function of standard deviation of random
properties with different rotational speeds. It is realized that with increasing of standard deviation of
random properties, the C.O.V. increases. Also, with increasing of rotational speed, the slope of
curves decreases. The relation between C.O.V. and standard deviation is linear.
The aforementioned results are obtained with the assumption that each random field of a random
property is fully-correlated. If random fields are semi-correlated, the random fields as well as beam
should be discretized. Consequently, three methods for discretization of random fields may be applied.
 Free vibration analysis of rotating beams with random properties 11

Fig. 6 C.O.V. of first eigenvalue of Timoshenko rotating beam versus rotating speed with different variances

Fig. 7 C.O.V. of first eigenvalue of Timoshenko rotating beam versus rotating speed with different variances

To discretize random fields, methods of midpoint, interpolation and local average are applied.
ρ (x) and E(x) are assumed random and their variances are identical and equal to 0.01. As a result,
the correlation matrix has 4 nonzero elements:
[ρ(r)] 3,3
= [ρ(r)] 5,5
= ρ (r),
0
[ρ(r)] 5,3
= [ρ(r)] 3,5
= b*ρ ( r )
0
(21)
12 S. A. A. Hosseini and S. E. Khadem

Fig. 8 C.O.V. of first eigenvalue of Timoshenko rotating beam versus standard deviation of random properties
with different rotating speed.

Fig. 9 C.O.V. of first eigenvalue of Timoshenko rotating beam versus number of elements with different
discretization methods for correlated and uncorrelated random fields

Where ρ ( r ) = exp ( –a r / L ) , and b is a constant that defines the correlation between random
0

fields ρ (x) and E(x).

Fig. 9 shows C.O.V. versus the number of elements for the three methods of discretization. The
figures represent correlated and uncorrelated random fields. The rotational speed is zero. It is seen
that compared to uncorrelated case the correlated case has large C.O.V. Also, if the value “a” is
large or if correlation length is small, the convergence rate is lower and more number of elements is
 Free vibration analysis of rotating beams with random properties 13

Fig. 10 C.O.V. of first eigenvalue of Timoshenko rotating beam versus number of elements with different
discretization methods and rotating speed

Fig. 11 C.O.V. of first eigenvalue of Timoshenko rotating beam versus number of elements with different
discretization methods and three correlation functions

necessary for convergence of final response.

It is realized that in correlated case (a = 4.29, b = 1) the convergence rate of interpolation and
local average methods is higher than that midpoint method. As it is seen from Fig. 9, the midpoint
and interpolation methods converge to the solution from upper limit where the local average method
approaches the solution from the lower limit. The curves of uncorrelated case (a = 10, b = 0) are
14 S. A. A. Hosseini and S. E. Khadem

also investigated. It is seen again, that the convergence rate of interpolation and local average
methods is higher than that of midpoint method. Consequently, in general, the convergence rate of
interpolation and local average methods is higher than that of midpoint method.
Fig. 10 shows the C.O.V. versus the number of elements for three different rotational speeds. The
most important result of the figure is that with the increase of rotational speed, the C.O.V.
decreases. It can be concluded that the change of rotational speed has a small effect on convergence
rate. In summary, the method of discretization has significant effect on the convergence rate of
response. The number of elements necessary for a good convergence rate has direct relationship
with correlation length. The smaller correlation length, the larger the number of elements. On the
other hand, if the number of elements is large, the stiffness and mass matrices will also be large.
This creates problems in the numerical calculation of these matrices.
Fig. 11 shows the effect of correlation function on convergence rate. The curves are plotted versus
the number of elements and for three correlation functions of ρ (r) = exp(–(a r /L) ) , ρ (r) = 2

exp(–a r /L) , ρ (r) = (1 + a r /L)exp(–a r /L) namely “Gaussian”, “first order autoregressive” and
0 0

“second order autoregressive” respectively. It can be inferred that the correlation function has a
small effect on convergence rate.
Fig. 12 shows the C.O.V. versus a = L/(Correlation Length) for three different rotational speeds
(b = 0). It is apparent here that for small rotational speed as the parameter “a” increases, the C.O.V
decreases. In other words, when the correlation length decreases, the C.O.V. also decreases.
Likewise, as the rotational speed increases, the slope of curves decreases. It is necessary to note that
in rotational speed Ω = 400 rad/s the curve is not monotonic.
Fig. 13 is plotted assuming the random fields E(x) and ρ (x) are correlated (b = 1). It is seen that
as the rotational speed increases the C.O.V. decreases. Likewise the C.O.V increases with respect to
“a” in range of 0 < a < 3.5 and if a > 3.5 the C.O.V decreases. This feature is approximately
apparent for all rotational speeds.

Fig. 12 C.O.V. of first eigenvalue of Timoshenko rotating beam versus parameter “a = 0.8/(Correlation Length)”,
for uncorrelated random fields
 Free vibration analysis of rotating beams with random properties 15

Fig. 13 C.O.V. of first eigenvalue of Timoshenko rotating beam versus parameter “a = 0.8/(Correlation
Length)”, for correlated random fields

Fig. 14 C.O.V. of first eigenvalue of Timoshenko rotating beam versus number of elements for correlated and
uncorrelated random fields

Fig. 14 shows the C.O.V. versus rotational speed using 14 elements. It should be taken into
account that here only the material properties are random (E(x), ρ (x), G(x)) . Curves are plotted
for correlated and uncorrelated random fields and setting angle 0 and 90 degrees (a = 1). The
C.O.V. corresponding to uncorrelated random field is larger than the correlated random field. For
correlated and uncorrelated cases, the C.O.V. corresponding to setting angle of 90 degree is higher
16 S. A. A. Hosseini and S. E. Khadem

Fig. 15 C.O.V. of first eigenvalue of Timoshenko rotating beam versus number of elements for correlated and
uncorrelated random fields

Fig. 16 Relative sensitivity coefficient of first eigenvalue of Timoshenko rotating beam versus rotating speed,
setting angle is 0 degree

than setting angle of 0 degree. The reason might be the existence of term ( [ Kγ ] – sin ψ [ M t ] ) in
2

stiffness matrix (Eq. (25)) and the fact that where the setting angle increases the randomness effects
of ρ (x) increase (matrix [M ] is function of ρ (x)).
t

Fig. 15 is plotted with properties of Fig. 14, with the exception that the geometry properties (A(x),
I(x)) are assumed to be random. The figure is similar to the previous one. Also, in this figure the
C.O.V. corresponding to uncorrelated random field is larger than the correlated random field.
Likewise, because matrix [M ] is function of A(x), where the setting angle increases the randomness
t
 Free vibration analysis of rotating beams with random properties 17

effects of A(x) increase.

For sensitivity analysis of random properties on response, parameter “η” (relative sensitivity
coefficient) is defined. This parameter determines the effect of randomness of each property on
deviation of response. The advantage of the analysis is that it determines the random weight of each
random property. As a result, the random properties with small η can be assumed to be
deterministic. For this aim, Figs. 16 to 19 are plotted. Here the random fields are assumed
uncorrelated and any field is fully-correlated and the variance is 0.01. In Fig. 16 “η”, is plotted

Fig. 17 Relative sensitivity coefficient of first eigenvalue of Timoshenko rotating beam versus rotating speed,
setting angle is 90 degree

Fig. 18 Relative sensitivity coefficient of first eigenvalue of Bernoulli-Euler rotating beam versus rotating
speed, setting angle is 0 degree
18 S. A. A. Hosseini and S. E. Khadem

Fig. 19 Relative sensitivity coefficient of first eigenvalue of Bernoulli-Euler rotating beam versus rotating
speed setting, angle is 90 degree

versus rotational speed for Timoshenko beam and the setting angle is 0 degree. It is seen the effect
of randomness of parameter “G(x)” is approximately zero. It means that without any error we can
assume G(x) to be a deterministic property. In the first mode eigenvalue of a rotating beam, the
effects of shear deformation is very small and because the shear deformation is directly proportional
to shear modulus G(x), then the variation of shear modulus has very small effects on variation of
first mode eigenvalue. On the other hand, the value of parameters A, ρ, I, E which have medium η
will decrease with an increase of rotational speed. The curves corresponding to rotational speed is
ascending i.e. this parameter has high η in high rotational speed. The setting angle in Fig. 17 is 90
degree. The Figure is similar to previous one with the exception that here, the curves are smoother.
The justification is the presence of term –Ω Sin Ψ [ M t ] in stiffness matrix, will change the
2 2

stiffness as ψ changes. Consequently, the randomness of properties will bear different results.
Figs. 18 and 19 are plotted for Bernoulli-Euler beam with setting angle of 0 and 90 degrees. Here,
one gets the same results as the ones for Timoshenko beam and there are not much differences
between the parameter η of these two types.

7. Conclusions
In this paper, free vibration of rotating beam with random properties is studied. Cross-sectional
area, elasticity modulus, moment of inertia, shear modulus and density are modeled as random
fields and the rotational speed as a random variable. To study uncertainty, stochastic finite element
method based on second order perturbation method is applied. To discretize random fields, the three
methods of midpoint, interpolation and local average are applied and compared. The most important
results of the paper as follows:
1. If rotational speed is random, C.O.V. curve is ascending with respect to rotational speed and if
 Free vibration analysis of rotating beams with random properties 19

material or geometry properties are random, C.O.V. curves are descending. If all properties are
random, The C.O.V. curves are first descending due to randomness of material and geometry
and then due to rotational speed they are ascending.
2. C.O.V. of first mode eigenvalue of Timoshenko and Bernoulli-Euler rotating beams are
approximately identical.
3. If variance of random properties increases, the C.O.V. increases accordingly.
4. Compared to uncorrelated random fields the correlated case has larger C.O.V. value.
5. If correlation length is small, the convergence rate is lower and more number of elements are
necessary for convergence of final response.
6 The convergence rate of interpolation and local average methods is higher than that midpoint
method.
7. The midpoint and interpolation methods converge to the solution from upper limit where the
local average method approaches the solution from the lower limit.
8. With the increase of rotational speed, the C.O.V. decreases.
9. The rotational speed has fairly small effect on convergence rate.
10. The correlation function has fairly small effect on convergence rate.
11. When the correlation length decreases, the C.O.V. also decreases.
12. The C.O.V. corresponding to uncorrelated random field is larger than the correlated random
field.
13. The effect of randomness of parameter “G(x)” is approximately zero. It means that without
any error we can assume G to be a deterministic property. On the other hand, the parameters
A, ρ, I, E have medium η that with the increase of rotational speed their value decrease.
14. There are not much differences between the parameter η of Timoshenko and Bernoulli-Euler
types.

References
Abbas, B.A.H. (1986), “Dynamic analysis of thick rotating blades with flexible roots”, Aeronaut. J., 89, 10-16.
Al-Bedoor, B.O. and Hamdan, M.N. (2001), “Geometrically nonlinear dynamic model of a rotating flexible
arm”, J. Sound Vib., 240(1), 59-72.
Bazoune, A., Khulief, Y.A., Stephan, N.G. and Mohiuddin, M.A. (2001), “Dynamic response of spinning tapered
Timoshenko beams using modal reduction”, Finite Elements in Analysis and Design, 37, 199-219.
Boyce, E.W. and Goodwin, B.E. (1964), “Random transverse vibration of elastic Beams”, SIAM J., 12, 613-629.
Chakraborty, S. and Dey, S.S. (1998), “A stochastic finite element dynamic analysis of structures with uncertain
parameters”, Int. J. Mech Sci., 40.
Chandiramani, N.K., Shete, C.D. and Librescu, L.I. (2003), “Vibration of higher-order-shearable pretwisted
rotating composite blades”, Int. J. Mech. Sci., 45, 2017-2041.
Combou, B. (1975), “Application of first order uncertainty analysis in the finite element method in linear
elasticity”, Proc. of 2nd Int. Conf. on Application of Statistics and Probability in Soil and Structure
Engineering, England, 67-68.
Handa, K. and Anderson, K. (1981), “Application of finite element method in the statistical analysis of
structures”, Proc. of 3rd Int. Conf. on Structural Safety and Reliability, Norway, 409-417.
Ishida, R. (2001), “Stochastic finite element analysis of beam with stochastical uncertainties”, AIAA J., 39, 2192-
2197.
Iwan, W.D. and Haung, C.T. (1996), “On the dynamic response of non-linear systems with parameter
uncertainty”, Int. J. Non-Linear Mech., 31, 631-645.
Jung, S.N., Nagaraj, V.T. and Chopra, I. (2001), “Refined structural dynamics model for composite rotor blades”,
20 S. A. A. Hosseini and S. E. Khadem

AIAA J., 39(2), 339-348.

Kleiber, M. and Hien, T.D. (1992), The Stochastic Finite Element Method, Chichester, Wiley.
Lee, S.Y. and Lin, S.M. (1994), “Bending vibrations of rotating non-uniform Timoshenko beams with an
elastically restrained root”, J. Appl. Mech., ASME, 61, 949-955.
Lin, S.C. (2000), “Sensitivity of dynamic behavior of random rotating Timoshenko beams to system parameter
changes”, J. Aerospace Engeering, 214, 247-257.
Lin, S.C. (2001), “The probabilistic approach for rotating Timoshenko beams”, Int. J. Solids Struct., 38(6),
7197-7213.
Lin, S.M., Lee, S.Y. and Wang, W.R. (2004), “Dynamic analysis of rotating damped beams with an elastically
restrained root”, Int. J. Mech. Sci., 46, 673-693.
Liu, W.K., Belytschko, T. and Mani, A. (1986), “Random field finite elements”, Int. J. Num. Meth. Engng., 23,
1831-1845.
Pohit, G., Mallik, A.K. and Venkatesan, C. (1999), “Free out-of-plane vibrations of a rotating beam with
nonlinear elastomeric constraints”, J. Sound Vib., 200, 1-25.
Singh, M. (1985), “Turbine blade dynamics- a probabilistic approach”, The Tenth Biennial Conference on
Mechanical Vibration and Noise, Ohio, 41-48.
Vanmarcke, E.H. and Grigoriu, M. (1983), “Stochastic finite element analysis of simple beams”, J. Engng.
Mech., ASCE, 109(5), 1203-1214.
Yang, S.M. and Tsao, S.M. (1997), “Dynamic of a pretwisted blade under nonconstant rotating speed”, Comput.
Struct., 62, 643-651.
Yokoyama, T. (1988), “Free vibration characteristics of rotating Timoshenko beam”, Int. J. Mech. Sci., 30(10),
743-755.