Journal of Sound and <ibration (2000) 229(5), 1023}1055

doi:10.1006/jsvi.1999.2449, available online at http://www.idealibrary.com on

DYNAMICS OF A FLEXIBLE BEAM CARRYING A MOVING

MASS USING PERTURBATION, NUMERICAL AND
TIME-FREQUENCY ANALYSIS TECHNIQUES
S. A. Q. SIDDIQUI
National Research Council of Canada, Institute of Aerospace Research,
Ottawa, Ontario, Canada K1A-OR6
M. F. GOLNARAGHI
Department of Mechanical Engineering, ;niversity of =aterloo, =aterloo, Ontario,
Canada N2¸ 3G1
AND

G. R. HEPPLER
Department of Systems Design Engineering, ;niversity of =aterloo, =aterloo, Ontario,
Canada N2¸ 3G1

(Received 21 January 1999, and in ,nal form 20 May 1999)

The dynamic behaviour of a #exible cantilever beam carrying a moving

mass-spring is investigated. This system is an idealization of an important class of
problems that are characterized by interaction between a continuously distributed
mass and sti!ness sub-system (the beam), and a lumped mass and sti!ness
sub-system (the moving mass-spring). Inertial non-linearities form the coupling
between the two, resulting in internal resonance behavior under certain parametric
conditions. The dynamics of the system are described by coupled non-linear partial
di!erential equations, where the coupling terms have to be evaluated at the
position of the moving mass. The equations of motion are solved numerically using
the Galerkin method and an automatic ODE solver. The numerical results are
compared with a closed-form analytical solution obtained using a perturbation
method and a parametric analysis of the system is performed using the
perturbation solution. The spectral behavior of the system is investigated using
time-frequency analysis.
( 2000 Academic Press

1. INTRODUCTION
Investigation of the dynamics of a beam carrying a moving mass has been an area
of research interest for a number of years [1, 2]. Historically, the problem "rst arose
in the design of railway bridges and later in other transportation engineering
structures. There have been numerous investigations in this regard. Some of the
early investigations were by Stokes [1] and Ayre [3]. There are two well-known

0022-460X/00/051023#33 $35.00/0 ( 2000 Academic Press

1024 S. A. Q. SIDDIQUI E¹ A¸.

monographs in this area, one by Inglis [4] and the other by Hillerborg [5]. There
were also some investigations into the e!ect of high-speed moving forces on beams
(see, e.g., references [6, 7]). A more recent book by Fryba [8] includes analyses
under di!erent loading conditions. These earlier studies neglected the inertial e!ect
of the moving mass by considering it as a moving force and the solution techniques
used were generally based on integral transformation or asymptotic expansions.
See, for example, Stanisic et al. [9], where an asymptotic expansion method is used
to obtain an approximate analytical solution.
In transport engineering problems, the traversing mass and a continuous beam
model results in a partial di!erential equation with intertial coupling terms which
depend on the position of the mass. Due to these coupling terms, the mode shapes
of simple beams do not arise as eigenfunctions in the separation of variables
method even if a linear model of the beam is assumed. Hayashikawa and Watanabe
[10] developed a method similar to the dynamic sti!ness approach to obtain
natural frequencies and mode shapes and used it to obtain the response of
multi-span beams with moving forces. Another approach is reported by Stanisic
[11] where a method is developed to obtain mode shapes which account for the
motion of the mass.
A "nite-element-based method was used by Cifuentes [12] where a set of
auxiliary functions were developed to account for the e!ect of the moving mass at
each node as it moves along the length of the beam.
Lin and Tretheway [13] considered a moving mass with a spring and damper
traversing the beam. The damping and the spring sti!ness was assumed to be in the
direction of the beam de#ection. The "nite element method along with
Runge}Kutta time integration was used in obtaining the solution. Internal
resonance behavior of the system was however not considered.
Gbadeyan and Oni [14] considered moving forces and moving masses on beams
and plates by using integral transformations and asymptotic expansions. The beam
and the plate were both assumed to be of the Rayleigh type which includes the e!ect
of rotatory inertia.
Lee [15] analyzed the problem of the moving mass separating from the beam by
monitoring the contact forces, while Michaltsos et al. [16] discussed the e!ect of the
moving mass and other parameters on the dynamic response of the beam. Henchi
et al. [17] developed a dynamic sti!ness matrix for the analysis of beams with
moving masses.
The problem arises in many applications other than in the motion of vehicles on
bridges. Some space structures [18] and systems such as cranes carrying moving
loads [2] exhibit similar behaviors. Some novel applications, like using the moving
mass as a controller to suppress vibrations in the beam, have also been proposed
[19, 20].
The papers cited earlier [1, 2] dealt with a problem where the motion of the mass
was prescribed, and its e!ect on the beam response was studied. In this work, the
focus is on the non-linear interaction between the mass and the beam, and unlike
the frequently used assumption of prescribed motion for the moving mass, the mass
is assumed to move under an applied force. The applied force is assumed to be
proportional to the displacement of the moving mass; hence, conceptually a spring is
 BEAM WITH MOVING MASS 1025

Figure 1. System model.

attached to the mass. The virtual spring serves two purposes: to prevent the mass
from sliding o! the beam and to promote an oscillatory motion for the moving mass.
In this work, a #exible cantilever beam carrying a moving spring}mass system is
considered. This system was studied earlier by Siddiqui et al. [21] where numerical
solutions were obtained using the Rayleigh}Ritz method and the results were
compared with solutions obtained using a semi-analytic}numeric approach based
on the perturbation method of multiple scales. In this paper, this work is extended
to obtain a closed-form analytical solution using a perturbation method. Besides
higher accuracy, the closed-form solution gives the ability to conduct parametric
analysis, which was not possible in reference [21] using the semi-analytic}numeric
approach. In non-linear systems, small changes in the parameters can cause
signi"cant qualitative and quantitative changes in the system response. Parametric
analysis is, therefore, used to identify regions of strong non-linear coupling between
the beam and the moving mass. Also, the spectral behavior of the system is
investigated using time-frequency analysis.
The mass}beam system to be considered here was investigated earlier by Khalily
et al. [22] where numerical solutions were obtained using two mode shapes for the
system. To account for the motion of the mass in the mode shapes, the method
developed in reference [11] was used. The numerical results obtained using these
mode shapes were not satisfactory, as unrealistically large initial values were
required to show the coupling between the mass and the beam.
The system model shown in Figure 1, consists of a cantilever beam carrying
a moving mass which has an attached spring. The equations of motion are a set of
two coupled non-linear partial di!erential equations where the coupling terms have
to be evaluated at the position of the mass. Inertial non-linearities in the system
arise due to the coupling between the mass and the beam. As a result, under certain
conditions, when one of the frequencies becomes an integral multiple of other
frequencies in the system, the phenomenon of internal resonance (IR) occurs. When
the system parameters are close to internal resonance conditions the dynamic
behavior undergoes a remarkable change which is characterized by the motion
undergoing distinctive beats. Understanding IR is therefore an important part in
the study of non-linear coupled systems.
Some of the papers investigating internal resonance behavior in continuous
systems are by Zavodney and Nayfeh [23], Nayfeh et al. [24], Pakdemirli and
Nayfeh [25] and Anderson et al. [26]. In Zavodney and Nayfeh [23], a slender
cantilever beam carrying a "xed lumped mass subjected to base excitation was
1026 S. A. Q. SIDDIQUI E¹ A¸.

considered and Nayfeh et al. [24] focussed on the dynamics of a pressure relief
value. A beam supported by a spring-mass was treated by Padkermirli and
Nayfeh [25] and a base-excited cantilever beam was considered by Anderson et al.
[26].
The solution methodology used in this work is to reduce the partial di!erential
equations to a set of non-linear ordinary di!erential equations (ODEs) using
Galerkin's method. As basis functions the mode shapes of a simple cantilever beam
are used. The choice of the basis functions used in the Galerkin method plays
a signi"cant role in the solvability of the resulting di!erential equations. Increasing
the number of basis functions increases the number of ODEs to be solved and
makes the di!erence between the smallest and the largest eigenvalues larger, thus
making the system &&sti!er''. The e!ect of the large eigenvalues may be insigni"cant
in the response but their presence requires taking very small step sizes to ensure the
stability of the ODE solvers. On the other hand, fewer basis functions may not give
the desired convergence to the solution. For the mass}beam system considered
here, the "rst four mode shapes of a simple cantilever beam are used as the basis
functions. The orthogonality of these functions makes some of the integrations
involved in solving the equations of motion simpler and the integrations can be
carried out analytically. The resulting non-linear ODEs are then solved
numerically using an automatic sti! ODE solver. The results are compared with
a closed-form solution obtained using the perturbation method of multiple scales.
The perturbation solution provides qualitative insight into the system behavior
and in this case allows for a closed-form solution in terms of elliptic functions.
Using the closed-form solution, a parametric study of the system is conducted. The
focus of this analysis is an internal resonance behavior between the moving mass
and the beam.
A numerical solution provides quantitative results and is a necessary step in
investigating more complex dynamic behavior like frequency modulation using
spectral analysis techniques. The numerical solution of non-linear ODEs can be
obtained using automatic solvers or direct time discretization using "nite
di!erences. Gear [27] gives a survey of automatic ODE solvers which can be
applied to problems reduced from PDEs. As the ODEs obtained through spatial
discretization of PDEs are characteristically sti!, the ODE solvers are generally
based on an implicit formulation, which requires solving a system of non-linear
algebraic equations, often many times during each time step. The automatic ODE
solver used in this work is based on an implicit formulation and a generalization of
the fourth order Runge}Kutta}Fehlberg method.
The time response of the system shows the amplitude of the mass and the beam
undergoing modulation due to internal resonance. To examine the evolution of this
behavior in the spectral domain, time-frequency analysis techniques are used. The
power spectrum, which is obtained by taking the discrete Fourier transform of the
time series and computing the power (mean-squared amplitude) at the various
frequencies, gives the averaged behavior for the length of time series. To investigate
the local spectral behavior of the system, a spectrograph is used. The spectrograph
is obtained by "nding the power spectrum of relatively small segments of data and
the results are displayed on time and frequency axes, with time corresponding to
 BEAM WITH MOVING MASS 1027
the centrex of the data segment and the power is shown using a grey scaling
for the whole plot. Increasing the size of the data segment improves the spectral
resolution but at the expense of time localization. For smoother transitions the data
segments are overlapped. To reduce leakage of the power from one frequency bin to
another, the data are windowed using the Hann Window. The numerical,
perturbation and the time-frequency analysis results are compared and studied for
a number of cases.

2. MATHEMATICAL MODELLING
The system and the various parameters used in its modelling are shown in
Figure 1. The beam parameters are length (l), area of cross-section (A), volume
mass density (o), second moment of the area about the z-axis (I ), and the modulus
z
of elasticity (E). The moving mass (m) slides along the length of the beam. The
position of the moving mass is measured by an arclength co-ordinate s and the
de#ection of the beam is given by v(x, t) measured from the undeformed centroidal
axis of the beam. As mentioned earlier, the mass is induced to move by an applied
force in contrast to the frequently used assumption of prescribed motion for the
moving mass. In this work, the applied force is assumed to be proportional to the
displacement of the mass. Hence, in e!ect, a virtual spring of sti!ness k is attached
to the mass.
Only non-dimensionalized parameters are used in the present analysis and they
are de"ned as

sL "s/l, vL "v/l, xL "x/l,

AK "A/l2, mL "m/oAl, IK "I /Al2, (1)

z z

tL "t/J(oAl4/EI ), u"J(ioAl4/(mEI ).
z z

Because we will be using only the non-dimensional parameters (1) and because it
will be convenient, the ( ( )s are dropped from here on.
The equations of motion are obtained from Hamilton's principle using a linear
model of the beam, based on the Euler}Bernoulli assumptions. The non-linearities
in the equations of motion arise due to the inertial coupling between the beam and
the motion mass. These equations were presented in Siddiqui et al. [21] and are not
reproduced here due to space considerations. The non-linear partial di!erential
equations of motion are reduced ordinary di!erential equations using the Galerkin
method. This procedure is detailed in Becker et al. [28]. As this is a relatively
straightforward procedure, only the "nal form of the semi-discretized equations are
presented here and a detailed derivation is availabe in Siddiqui [29]. Using the
following assumed trial function for the de#ection of the beam v(x, t),

v(x, t)"+ a (t)/ (x), (2)

i i
i
1028 S. A. Q. SIDDIQUI E¹ A¸.

where a (t) and / (x) are the time-dependent undetermined parameters and the
i i
spatial basis functions respectively; the following semi-discretized equations of
motion are obtained using the Galerkin method.
s equation:
/ /@y
sK #u2(s!s )#x MaK a N"0. (3)
e i j x/s(t) i j
v equation:

Mm[/ / ] MaK ]#sK [/ /@ ] Ma N#2sR [/ /@ ] MaR N

i j x/s(t) j i j x/s(t) j i j x/s(t) j

CP D CP D
1 1
#sR 2[/ /A ] Ma NN# / / dx MaK N# /A /A dx Ma N"0. (4)
i j x/s(t) j i j j i j j
0 0

In equations (3) and (4), index notation is used where the repeated indices imply
summation over the index. To further clarify the index notation, brackets are also
used; x 2 y denote a row matrix and M2N denote a column matrix whereas [2]
denote a square matrix. The derivatives with respect to t and x are denoted by ( ) )
and ( )@, respectively. Since the beam model is linear the non-linearities are only due
to inertial coupling; hence, there are no non-linear terms for the beam portion
(integral matrices) of the equations of motion (4). The number of equations of
motion depends on the number of basis functions / used for the approximation.
i
The basis functions in Galerkin's method are generally chosen to be as simple as
possible but a desirable property is that these functions should be orthogonal to
facilitate solving the initial value problem by producing well-conditioned system
matrices. The eigenfunctions for the cantilever beam are then a natural choice.
The cantilever beam eigenfunctions for the non-dimensionalized parameters
used here are given by
cos(k )#cosh(k )
/ "cosh(k x)!cos(k x)! i i (sinh(k x)!sin(k x)), (5)
i i i sin(k )#sinh(k ) i i
i i
where for the "rst four modes, the k have the values
i
k "1)8751, k "4)6941, k "7)8548 and k "10)9955.
1 2 3 4
It should be noted here that the eigenfunctions of a cantilever beam do not
account for the e!ect of the moving mass and are not the eigenfunctions of the
complete system, but they do satisfy the natural and the forced boundary
conditions with the mass con"ned to move between the two ends, and are used here
as basis functions in the Galerkin method.
The mass, and the sti!ness matrices in equation (4), [:1 / / dx] and
0 i j
1
[: /A /A dx], respectively, require integration of products of basis functions and
0 i j
their second derivatives over the length of the beam. These integrations are carried
out analytically using symbolic manipulation. The other terms remaining in
 BEAM WITH MOVING MASS 1029
equations (3) and (4) depend on the position of the moving mass and therefore have
to be dealt during the simulation.
The mass}beam system is assumed to carry no external forces. Initial values are
however prescribed, and the time evolution is investigated. The following initial
values are used:

K K
Ls(t) Lv(x, t)
s(0)"s , "0, v(x, 0)"v (x), "0, (6)
0 Lt 0 Lt
t/0 t/0
where v (x) represents the initial de#ection curve of the beam, s is the initial
0 0
position of the moving mass, and the initial velocities are assumed to be zero. The
initial values for the beam de#ection must be selected such that the boundary
conditions are satis"ed. In this work, the scaled "rst mode of a linear beam,

v / (x)
v (x)" t0 1 , (7)
0 2

is used, where v is the prescribed tip de#ection and / (x) is the "rst mode of
t0 1
a linear cantilever beam. The initial values for a are then obtained using the
i
orthgonality of the modes, a "a (0)":1 / (x)v (x) dx, which gives the following
i0 i 0 i 0
values:
a "1 v , a "a "a "0. (8)
10 2 t0 20 30 40

3. PERTURBATION ANALYSIS
A perturbation method is used to obtain qualitative insight into the behavior of
the system, especially the parametric behavior. In order to obtain a solution using
the perturbation method, the equations of motion, equations (3) and (4), are further
simpli"ed by expanding them about their equilibrium position and using only one
basis function / (x). It can be seen from equation (3) that the equilibrium position
1
for the mass is s and for the beam is a "0. Using the Taylor series expansion of
e i
equations (3) and (4) about the equilibrium position and including terms up to the
second derivative, the following equations are obtained for small motions about the
equilibrium positions:
sK #u2s#c aK a "0,
1 1 1
aK #u2 a #mc sK a #2mc sR aR #2mc saK "0, (9)
1 1 1 2 1 2 1 2 1
where the constants c , c , and u are de"ned according to
1 2 1
/ /@ D
c "/ /@ D , c " 1 1 x/se
1 1 1 x/se 2 :1 (/ )2 dx#m(/ )2D
0 1 1 x/se
:1 (/A )2 dx
(u )2" 0 1 . (10)
1 :1 (/ )2 dx#m(/ )2D
0 1 1 x/se
1030 S. A. Q. SIDDIQUI E¹ A¸.

Equations (9) are solved using the method of multiple scales by using a two-term
expansion. In application of this technique, the methodology presented in Nayfeh
and Mook [30] is followed. Begin by de"ning two time scales ¹ and ¹ as
0 1
¹ "t, ¹ "et, (11)
0 1
where e is a scaling parameter. If the non-linear terms are neglected in equation (10),
the system would be two uncoupled linear oscillators with frequencies u and
u . This would be the primary motion on time scale ¹ . The non-linearities
1 0
are expected to have a smaller e!ect and that e!ect will be on the slower time scale
¹.
1
The next step is to assume an asymptotic series solution for s and a . In this case,
1
a two-term expansion is assumed as per

s(t)"es (¹ , ¹ )#e2s (¹ , ¹ ), a (t)"eu (¹ , ¹ )#e2u (¹ , ¹ ), (12)

1 0 1 2 0 1 1 1 0 1 2 0 1
where s (¹ , ¹ ) and s (¹ , ¹ ) are the e and e2 order solutions, respectively, for the
1 0 1 2 0 1
moving mass position, and u (¹ , ¹ ) and u (¹ , ¹ ) and the e and e2 order
1 0 1 2 0 1
solutions respectively for the beam de#ection. The closed-form solution obtained
using the perturbation method corresponds to the e order terms s and u . To
1 1
compare this solution with a numerical solution, the initial values of s (denoted by
1
s ) and u (denoted by u ) are set equal to the initial values of s (denoted by s )
10 1 10 0
and a (denoted by a ) respectively, thus assuming the value of e equal to one.
1 10
The substitution of the asymptotic expansions (12) in the equations of motion (9),
and the elimination of the secular terms under internal resonance conditions was
carried out in an earlier papers by Siddiqui et al. [21]. Here, only the results of this
analysis are presented. The e order solution, s (¹ , ¹ ), is given by
1 0 1
s "P (¹ )e*uT0#PM (¹ )e~*uT0 , u "P (¹ )e*u1T0#PM (¹ )e~*u1T0 , (13)
1 1 1 1 1 1 2 1 2 1
where P and P are complex variables that are, in general, functions of the slower
1 2
time scales. The overbars in equation (13) denote the complex conjugate. The
complex variables P and P are converted to polar form using the relations
1 2
P (¹ )"1 p (¹ )e*u1(T1), P (¹ )"1 p (¹ )e*u2(T1). (14)
1 1 2 1 1 2 1 2 2 1
The following relationship between the frequency of the moving mass u and the
"rst frequency of the beam u , results in internal resonance between the moving
1
mass and the beam:
u"2u #ep, (15)
1
where p is a small detuning parameter. When p is zero, we have a perfect 1 : 2 ratio
between the "rst two natural frequencies of the system. This case is referred to as
1 : 2 IR. Under internal resonance conditions, the elimination of secular terms
 BEAM WITH MOVING MASS 1031
results in the following non-linear di!erential equations:

Lp 1 c p2 u2
1 " 1 2 1 sin(2u !u !p¹ ),
L¹ 4 u 2 1 1
1

Lu 1 c p2 u2
p 1"! 1 2 1 cos(2u !u !p¹ ),
1 L¹ 4 u 2 1 1
1

LP 1 mc (u2!2uu #2u2 )
2 "! 2 1 1 p p sin(2u !u !p¹ ),
L¹ 4 u 1 2 2 1 1
1 1

Lu 1 mc (u2!2uu #2u2 )
p 2"! 2 1 1 p p cos(2u !u !p¹ ). (16)
2 L¹ 4 u 1 2 2 1 1
1 1

In equations (16), p and p are the modal amplitudes and u and u are the
1 2 1 2
corresponding phases.
To determine the initial values for p , p , u and u , equations (13) are "rst
1 2 1 2
expressed in terms of trigonometric functions as

s "p (¹ ) cos(u¹ #u (¹ )), u "p (¹ ) cos(u ¹ #u (¹ )). (17)

1 1 12 0 1 1 1 2 1 1 0 2 1

Using equations (17) and their derivatives, and taking the initial velocities as zero
and setting e to one, the initial values p (0)"p , p (0)"p , u (0)"u and
1 10 2 20 1 10
u (0)"u are obtained by solving
2 20

s "p cos(u ), a "p cos(u ),

10 10 10 10 20 20

0"l p2 sin(2u )!up sin(u ),

1 20 20 10 10

0"l p p sin(u !u )!u p sin(u ). (18)

2 10 20 10 20 1 20 20

The following solution of equations (18) is used as the initial values:

p "s , p "a , u "0, u "0. (19)

10 10 20 10 10 20

As shown in Nayfeh and Mook [30], non-linear di!erential equations of the form
(16) can be solved analytically using elliptic functions. For the most part (equations
(20)}(27) and (30)}(39)) the approach presented in Nayfeh and Mook [30] for
obtaining a closed-form solution to equations (16) is followed. We proceed by
de"ning c as

c"2u !u !p¹ . (20)

2 1 1
1032 S. A. Q. SIDDIQUI E¹ A¸.

Using equation (20), equations (16) can be reduced to the three non-linear
di!erential equations
Lp Lp
1 "l p2 sin(c), 2 "!l p p sin(c),
L¹ 1 2 L¹ 2 1 2
1 1
Lc
p "(!2l p2#l p2 ) cos(c)!pp , (21)
1 L¹ 2 1 1 2 1
1
where l and l are constants given by
1 2
c u2 mc (u2!2uu #2u2 )
l " 1 1, l " 2 1 1 . (22)
1 4u 2 4u
1
Eliminating c from equations (21) and (22) and integrating the result gives
1 2
1 l p2#1 l p2"G, (23)
2 1 2 2 2 1
where G is a constant of integration, to be determined using the initial values for p
1
and p . Equation (23) is an expression of conservation of energy and shows energy
2
being exchanged between p and p .
1 2
Using equation (23) both p and p can be expressed in terms of one variable m as
1 2
1 l p2"Gm, 1 l p2"G(1!m). (24)
2 1 2 2 2 1
Eliminating ¹ between equations (21) and (21) , and rearranging the terms gives
1 1 3
p
!l d(p2 p cos(c))# d(p2 )"0,
1 2 1 2 1
p
!l p2 p cos(c)# p2"¸, (25)
1 2 1 2 1

where ¸ is another integration constant and d(2) implies implicit di!erentiation.

Solving for cos(c) from equations (25) gives
pp2!2¸
cos(c)" 1 (26)
2l p2 p
1 2 1
and using sin2(c)"1!cos2(c) to eliminate c from equation (21) and expressing p
1 1
and p in terms of m gives the di!erential equation
2

A B
1 Lm 2 p2!8l G p(¸l !pG) (¸l !pG)2
"!m3! 2 m2! 2 m! 2 . (27)
8l G L¹ 8l G 4G2l 8l G3
2 1 2 2 2
The problem is thus reduced to solving the single di!erential equation (27). The
solution of m depends on the roots m , m and m of the left-hand side of equation
1 2 3
 BEAM WITH MOVING MASS 1033
(27). The roots are
(2l p !p$D)(2l p #p) l p2
m " 2 10 2 10 , m " 1 20 , (28)
1,2 16Gl 3 2G
2
where the discriminant (D) is given by

D"J(p!2l p )2#16l l p2 . (29)

2 10 1 2 20
In deriving equation (28), the constants G and ¸ are evaluated using the initial
values p , p , u and u . In equation (29), l can become negative for large
10 20 10 20 1
values of p (when p is negative and has a magnitude greater than 2u ); however, in
1
such a case the system would be far away from 1 : 2 IR and this perturbation
analysis is not applicable. The discriminant (29) is therefore real under internal
resonance conditions. When the initial value p and the detuning parameter p are
10
close to zero, the roots m and m approach each other and m approaches unity and
1 2 3
when the other initial value p and p are close to zero the di!erence between m
20 1
and m approaches a maximum value and m approach zero. These cases are
2 3
discussed further in Section 6.
Assuming the roots are ordered such that m (m (m and writing equation (27)
1 2 3
as

A B
1 L 2
m "(m !m)(m!m )(m!m ), (30)
8l G L¹ 3 2 1
2 1
the following transformation is applied to m:

m !m"(m !m ) sin2(s). (31)

3 3 2
Equation (30) then reduces to

A B
1 L 2
s "m !m !(m !m ) sin2(s)
2l G L¹ 3 1 3 2
2 1

A B
(m !m )
"(m !m ) 1! 3 2 sin2(s) . (32)
3 1 (m !m )
3 1
Taking the square root of equation (32) gives

1 Ls
"$Jm !m (1!g2 sin2(s))1@2, (33)
J2l G L¹1 3 1
2
where g is given by

S
(m !m )
g" 3 2. (34)
(m !m )
3 1
1034 S. A. Q. SIDDIQUI E¹ A¸.

Integrating equation (33) gives

x T1
P0 PT d¹1
1
ds"$J2l G(m !m ) (35)
J1!g2 sin2(s) 2 3 1
e

In equation (35), ¹ corresponds to s"0 or m"m . The left-hand side of

e 3
equation (35) is Legendre's elliptic integral of the "rst kind, or the inverse of the
Jacobi elliptic function sn. In terms of sn, equation (35) can be written as

sn~1(sin(s); g)"i(¹ !¹ ), (36)

1 e
where i is given by

i"$J2l G(m , m ). (37)

2 3 1
From equation (36) it follows that

sin(s)"sn(i(¹ !¹ ); g). (38)

1 e
In equation (38), g is the modulus of the elliptic function and its value a!ects the
period of sn. Substituting equation (38) into equation (31), the solution for m is
obtained as
m"m !(m !m )sn2(i(¹ !¹ ); g). (39)
3 3 2 1 e
Using equations (24) and (39) the values of p and p can be easily obtained, and
1 2
then using equation (26), c can be determined. However, to "nd the phases u and
1
u , another equation is needed and is obtained by eliminating cos(c) from
2
equations (16) and (16) , and integrating the result, thus giving
2 4
l p2 u !l p2 u "Q, (40)
2 1 1 1 2 2
where Q is a constant of integration. Substituting the initial values for u and u
1 2
(see equation (19)) in equation (40) makes Q zero and the following relationship
between u and u is obtained:
1 2
l p2
u " 2 1u . (41)
2 l p2 1
1 2
Using equations (41) and (20), u and u can now be determined.
1 2
The beating period (q ) is another important characteristic of the response. It is
b
de"ned as the time elapsed between two successive peaks in the amplitude of
motion, and corresponds to half the period of p or p and hence to half the period
1 2
of the Jacobi elliptic function sn. The half period of sn(i¹ !¹ ) is given by the
1 e
following relationship where the integral is known as the complete Jacobi elliptic
 BEAM WITH MOVING MASS 1035
integral of the "rst kind:

2 n@2
P
1
q" dx. (42)
b i J1!g2 sin(x)2
0

Using the perturbation analysis, the solution is obtained for the modal
amplitudes p , p and the phases u and u . To summarize, the perturbation
1 2 1 2
solution is obtained using the following algorithm.

3.1. ALGORITHM 3.1: ALGORITHM TO OBTAIN THE PERTURBATION SOLUTION

(1) Set the initial value p to the initial position of the mass s , and p to the
10 10 20
"rst mode contribution of the tip de#ection a . Select a value for the
10
detuning parameter p. An algorithm is presented in the next section
(Algorithm 4.1) for selecting a value for p that is used to compare the
perturbation and numerical solutions. Also set the value of the equilibrium
position of the moving mass s .
e
(2) Compute the value of the "rst mode shape of a cantilever beam and its "rst
and second derivatives at the equilibrium position (see equation (5)). Using
these values obtain the constants c , c , and u from equations (10). Also "nd
1 2
the constants l , l , G, and L using equations (22), (23) and (25) respectively.
1 2
Calculate the discriminant D (equation (29)), the roots m , m , and m
1 2 3
(equation (28)) and sort the roots such that m , m , and m are in ascending
1 2 3
order. Find i using equation (37) and the modulus g using equation (34).
Compute the beating period q using equation (42). This involves "nding the
b
complete Jacobi elliptic integral of the "rst kind.
(3) Find the time ¹ . This requires calculating m from the initial values of p and
e 1
p using equation (24) and then "nding sin(s) using equation (31) and "nally
2
obtaining ¹ from equation (36) by substituting for ¹ , the initial time.
e 1
(4) Vary ¹ from the initial time to the desired "nal time ¹ and obtain m from
1 f
equation (39), modal amplitudes p and p from equation (24), c from
1 2
equation (26), and phases u and u from equations (20) and (41) respective-
1 2
ly.

4. COMPARISON BETWEEN PERTURBATION AND NUMERICAL SOLUTIONS

The results obtained using perturbation analysis are now compared with numer-
ical simulation of the ODEs derived using Taylor series expansion (9). The various
models and solution methodologies used in this work are identi"ed in Table 1 and
the designations shown are used for future reference. In this section, comparisons
are made between the perturbation solution (PM1) and the numerical solution
(NM1). The parameters used for all the simulations presented in this section are
given in Table 2. First consider the case where the initial displacement of the
moving mass about its equilibrium position s is 0)00001, the initial beam tip
0
1036 S. A. Q. SIDDIQUI E¹ A¸.

TABLE 1
Model designations

Model Description

PM1 Perturbation solution, "rst mode of a cantilever beam as basis function

NM1 Numerical solution, "rst mode shape of a cantilever beam as basis function
MM2 Numerical solution, four mode shapes of a cantilever beam as basis
functions

TABLE 2
Comparison between perturbation and numerical solution21

Parameter set 1
m"0)1, s "0)5, u using 20 "nite elements"2)891228
e 1
u using one mode of a cantilever beam"2)908776
1
Initial values Initial values
s "0)50001, v "0)1 s "0)53, v "0)00001
0 t0 0 t0
Figure 2, 3 4, 5
Model PM1, NM1 PM1, NM1
p !0)0002 !0)0085
Dt(average) 0)01797 0)0094
Spectrogram
No. of segments 16 32
Segment size 256 256

de#ection is v "0)1, the equilibrium position of the moving mass s "0)5, and the
t0 e
non-dimensionalized mass ratio m"1)0. Figures 2 and 3 show the response for the
moving mass and then beam respectively.
Figures 2(a) and (b) are the analytical results obtained using the perturbation
analysis. It may be noted here that p "0)5v (see equations (19) and (8)).
20 t0
Figure 2(c) is the numerical solution obtained from equation (9) by using a variable
step sti! ODE solver based on the fourth order Rosenbrock method presented in
reference [31]. This technique is based on an implicit formulation and is a general-
ization of the Runge}Kutta}Fehlberg method that uses the parameters presented
by Shampine (for details see reference [31]). Later on, the same ODE solver is used
for solving the more general equations of motion, equations (3) and (4), using four
cantilever beam mode shapes. Figure 2(c) appears darkened because the oscilla-
tions are at a very high frequency and small time steps were required.
Figure 2(d) shows a spectrogram obtained by taking a small data window
consisting of 16 segments of data with 256 data points per segment, of the time
series. A one-sided power spectral density (PSD) is obtained, with the mean squared
 BEAM WITH MOVING MASS 1037

Figure 2. Mass response !1 : 2 IR, m"1)0, s "0)5, s "0)00001, and v "0)1. (a), (b)
e 10 t0
Perturbation solution p"!0)0002, (c) numerical solution, and (d) spectrogram.

amplitude as the measure, and using the fast-fourier transform (FFT) algorithm
[31]. For a detailed treatment on time-frequency analyses see Cohen [32]. The data
window is advanced along the time axis one segment at each step allowing
oberlapping of the previous data segments by the current data window. The
spectrogram (Figure 2(d)) shows the variation in the PSD for the frequencies on the
vertical axis with time on the horizontal axis. The change in the PSD is represented
by the grey scaling with the darker regions representing higher values of PSD and
lighter regions representing lower values of PSD. The empty space at the beginning
of the graph is half the size of the data window. Note that the mean-squared
1038 S. A. Q. SIDDIQUI E¹ A¸.

Figure 3. Tip de#ection !1 : 2 IR, m"1)0, s "0)5, s "0)00001, and v "0)1. (a), (b)
e 10 t0
Perturbation solution p"!0)0002, (c) numerical solution, and (d) spectrogram.

amplitude, the measure used for PSD, represents energy, and the spectrogram
shows the variation in the energy content of the frequencies with time.
Figure 3 shows similar results obtained for the beam. The time responses in
Figures 2(a), 3(a), 2(c) and 3(c) show characteristic beating motion for the mass and
the beam under internal resonance. The numerical solutions (Figures 2(c) and 3(c))
are obtained for u"2u . Under such perfect resonance conditions, the perturba-
1
tion analysis gives a solution where the amplitudes match closely with the numer-
ical solution, but there are small di!erences in the beating periods which may be
attributed to neglecting the higher order terms in the perturbation analysis. To
compare the perturbation and the numerical solutions the approach used here is to
 BEAM WITH MOVING MASS 1039
match the two results by choosing the detuning parameter p. The value of p then
acts as a measure of the di!erence between the two solutions. This approach is
justi"ed because the natural frequency of the beam u , obtained using equa-
1
tion (10) for the simpli"ed model used in the perturbation analysis would in general
be slightly di!erent than u obtained using equations (3) and (4). The method for
1
"nding the value of p is outlined in the following algorithm:

4.1. ALGORITHM 4.1: ALGORITHM TO OBTAIN THE DETUNING PARAMETER p

(1) Find the beating period (q ) for the numerical solution. This is accomplished
b
by taking the Hilbert transform of the time series for the tip de#ection of the
beam, which gives an enveloping curve for the amplitude. Using the ex-
tremum values of this curve the beating periods can be found easily. The
beating periods di!er slightly from one beat to another, therefore an average
value is selected.
(2) Using equation (42) the value of p is determined numerically by varying
p and "nding the beating period which matches ¹ determined in step (1).
b
For Figures 2 and 3 the value of p was determined as !0)0002 which gives the
same beating period as the numerical results.
It can be seen from equations (17) that p and p describe the change in the
1 2
amplitude, and u and u could possibly a!ect the natural frequencies u and u
1 2 1
respectively. This corresponds to both amplitude and frequency modulation. Fig-
ure 2(b) shows that the phases u and u remain constant, except that they
1 2
approach singular positions just before and after the beam reaches a peak, or the
mass amplitude becomes zero. A similar e!ect can be seen in the spectrogram,
Figure 2(d), where the energy corresponding to the frequency u disappears. The
1
energy is in fact transferred to the beam as can be seen from Figure 3(d).
Figures 4 and 5 are obtained for the same parameters as in Figures 2 and 3 but
the initial values are now taken as s "0)03 and tip de#ection v "0)00001. In this
0 10
case, the system is predominantly excited by the moving mass. The value of p for the
perturbation solution was found to be !0)0085. The magnitude of the peaks for
the phases u and u is not similar to that seen in Figures 2 and 3, but seems to
1 2
increase with time.

5. PARAMETRIC ANALYSIS
The response of the system depends on the roots m , m and m and Figures 6 and
1 2 3
7 show the values of these roots for di!erent m and p. Figure 6 is obtained for initial
values of mass position s "0)00001 and tip de#ection v "0)01 and with the
0 t0
equilibrium position for the moving mass s "0)5. Whereas Figure 7 is obtained
e
for initial values s "0)03, v "0)00001 and equilibrium position s "0)5. Where-
0 t0 e
as Figure 7 is obtained for initial values s "0)03, v "0)00001 and equilibrium
0 t0
position s "0)5. Similar to the circular sine function, the amplitude of the elliptic
e
sine function sn varies between !1 and 1. Therefore, from equation (39) it follows
that m oscillates between m and m . The farther apart the roots m and m are, the
2 3 2 3
1040 S. A. Q. SIDDIQUI E¹ A¸.

Figure 4. Mass response !1 : 2 IR, m"1)0, s "0)5, s "0)003, and v "0)00001. (a), (b)
e 10 t0
Perturbation solution p"!0)0085, (c) numerical solution, and (d) spectrogram.

larger is the amplitude of m, which from equation (24) implies that more exchange of
energy occurs between the moving mass and the beam. Both Figures 6 and 7 show
that this region of strong coupling occurs when p is close to zero and when p moves
away from zero the di!erence between the roots m and m decreases, "nally
2 3
becoming zero. As the ratio of the moving mass and the mass of the beam (m)
increases, the range of p for which the roots m and m are distinct and increases.
2 3
This can be seen more clearly in Figure 7(c). When the roots m and m become
2 3
equal, it follows from equation (39) that the solution for m becomes a constant,
equal to one for the case considered in Figure 6 and equal to zero for the case
considered in Figure 7.
 BEAM WITH MOVING MASS 1041

Figure 5. Tip de#ection !1 : 2 IR, m"1)0, s "0)5, s "0)03, and v "0)00001. (a), (b)
e 10 t0
Perturbation solution p"!0)0085, (c) numerical solution, and (d) spectrogram.

The modulus g of the elliptic function sn gives some insight into the type of
response. When g"0 the elliptic function sn becomes the circular sine function and
for g"1, sn becomes tanh. So it follows from equations (34), (39) and (24) that
when m in close to m , p and p appear closer to sinusoidal functions, and when m
2 3 1 2 2
and m are numerically far apart from each other, p and p appear close to
3 1 2
hyperbolic functions (p close to tanh and p close to sech).
1 2
When m and m become equal, the modulus g becomes unity (see equation (34)),
1 2
the beating period q becomes in"nite (see equation (42)), and in equation (39) the
b
elliptic sine function sn becomes tanh, thus once the energy is transferred from the
mass to beam or vice versa it stays there as the period for p and p is in"nite. This
1 2
1042 S. A. Q. SIDDIQUI E¹ A¸.

Figure 6. Roots!1 : 2 IR, s "0)5, s "0)00001, and v "0)1. (a) m , (b) m , and (c) m .
e 10 t0 1 2 3

motion is however unstable and the smallest di!erence between m and m makes
1 2
the beating period "nite. In Figure 7, it appears that m and m are equal for a large
1 2
range of p, there is however a small di!erence between them which is due to the
small non-zero initial value of the tip de#ection v . It may be noted here that when
10
v is zero, there is no coupling between the moving mass and the beam, but even
t0
a small non-zero value of v results in large-amplitude vibrations of the beam as
t0
well illustrated in Figures 4 and 5.
The maximum and minimum values of p and p are indicative of the exchange of
1 2
energy between the mass and the beam. Since m oscillates between m and m , the
2 3
maximum value being m and the minimum value m , using equation (24), the
3 2
 BEAM WITH MOVING MASS 1043

Figure 7. Roots !1 : 2 IR, s "0)5, s "0)03, and v "0)00001. (a) m , (b) m , and (c) m .
e 10 t0 1 2 3

maximum and the minimum values of p and p are given by

1 2

S S
2G(1!m ) 2Gm
p " 2 , p " 3,
1max l 2max l
2 2

S S
2 2Gm
p " (1!m )l , p " 3. (43)
1min 3 2 2min l
1
Figure 8 shows p and p for various values of m and equilibrium position s .
1max 2min e
The initial value p is taken as 0)00001 and p as 0)05 (same as in Figures 2, 3,
10 20
1044 S. A. Q. SIDDIQUI E¹ A¸.

Figure 8. Maximum and minimum amplitude !1 : 2 IR, s "0)5, p"!0)0002, s "0)00001, and
e 10
v "0)1. (a) Maximum amplitude for the mass p , (b) minimum amplitude for the tip of the beam
t0 1max
p .
2min

and 6). Figure 8 shows that when the equilibrium position of the moving mass is
close to the "xed of the beam, the value of p is almost the same as its initial value
2min
p and also the corresponding maximum amplitude of the mass is close to zero
20
indicating a weak coupling between the moving mass and the beam. As the
equilibrium position is moved towards the free end of the beam, the exchange of
energy between the mass and the beam increases as indicated by p decreasing
2min
sharply and approaching zero and p increasing correspondingly. The other
1max
maximum and minimum values p and p are not shown as they do not vary
1min 2max
signi"cantly. The values of p remains close to its initial value p and the value
1max 10
of p remains close to its initial value p . Figure 8 was obtained for a value of
2max 20
p close to perfect 1 : 2 resonance (p"!0)0002). Figure 8 also shows an expected
result that a lighter moving mass (smaller value of m) oscillates with a larger
amplitude than a larger moving mass (larger value of m). Figure 9 shows the
behavior of p and p for the case with initial values p "0)03 and
1min 2max 10
p "0)0000005 (same as in Figures 4, 5, and 7). The values of p and p are
20 1max 2min
not shown as they remain the same as their initial values. For this case, the values of
p and p are less sensitive to changes in s near the "xed end of the beam and
1min 2max e
as s moves away from the "xed end a sharp change in p and p is observed.
e 1min 2max
Figure 10 shows the beating period for di!erent values of m and p for the "rst
case where the initial values are p "0)00001 and p "0)05. For p close to zero the
10 20
beating periods are large and decrease as p increases. The maximum beating period
 BEAM WITH MOVING MASS 1045

Figure 9. Maximum and minimum amplitude !1 : 2 IR, s "0)5, p"!0)0085, s "0)003, and
e 10
v "0)00001. (a) Minimum amplitude for the mass p , (b) maximum amplitude for the tip of the
t0 1min
beam p .
2max

Figure 10. Beating period q !1 : 2 IR, s "0)5, p"!0)0085 s "0)00001, and v "0)1.
b e 10 t0

occurs when m and m are close to each other or when the modulus
1 2
g approaches unity (see equation (42)). In Figure 11, results are shown for the
second case where p "0)03 and p "0)000005. This "gure is plotted in two
10 20
parts (a) and (b) each with di!erent range of values of m to improve the resolution.
For the smaller mass ratio in Figure 11(a), two peaks in the beating period, which
move apart as the mass ratio increases, are observed. From Figure 7 it can be seen
that m and m are close to each other for a range of values of p and the two peaks in
1 2
Figure 11(a) correspond to the end points of this range. It can be seen from
Figure 7 that corresponding to the peaks in the beating period, the values of m and
2
m are close to each other, thus indicating a weak coupling between the beam and
3
1046 S. A. Q. SIDDIQUI E¹ A¸.

Figure 11. Beating period q !1 : 2 IR, s "0)5, s "0)03, and v "0)00001. (a) m"0)1}0)5 and
b e 10 t0
(b) m"0)6}1)0

the moving mass. The beating period in Figure 11 behaves in an opposite manner
to that observed in Figure 10 with the minimum value of q being to p"0.
b
Figures 12 and 13 show the change in the beating period as m and p are varied. In
these "gures, the s axis is broken into two parts to improve the resolution. From
e
Figure 12 (obtained for p "0)00001 and p "0)05 and s "0)5) it can be seen
10 20 e
that as the equilibrium position of the moving mass moves towards the free end, the
beating period decreases. Figure 13 shows a similar result obtained for the other
case (p "0)03 and p "0)000005, and s "0)5) where the larger initial value is
10 20 e
given to the moving mass. For this case, the beating period approaches a peak when
s is closer to the "xed end of the beam, depending on the mass ratio, and decreases
e
as s approaches the free end.
e

6. AMPLITUDE MODULATION
In this section, the equations of motion, equations (3) and (4), are solved
numerically using the four cantilever beam mode shapes, equation (5), as the basis
functions. The automatic ODE solver discussed in section 5 is used to obtain the
solution and time-frequency analysis is performed in the manner outlined earlier.
The results are compared with the perturbation solution for the simpli"ed model
obtained in section 5. To establish internal resonance, the fundamental frequency
of the beam must be known. In the perturbation analysis for the simpli"ed model,
 BEAM WITH MOVING MASS 1047

Figure 12. Beating period q !1 : 2 IR, s "0)5, s "0)00001, and v "0)1. (a) s "0)01}0)3 and
b e 10 t0 e
(b) s "0)3}0)9.
e

Figure 13. Beating period q !1 : 2 IR, s "0)5, p"!0)0085 s "0)03, and v "0)00001.
b e 10 t0
(a) s "0)1}0)5 and (b) s "0)5}0)9.
e e
1048 S. A. Q. SIDDIQUI E¹ A¸.

TABLE 3
Beam frequencies obtained using ,nite elements and the one mode equation

FEM One-mode
m s frequencies frequency
e
1)0 0)9 1)758689 1)763542
19)537843
60)469989
120)762565
1)0 0)5 2)891228 2)908776
14)225427
61)681059
95)28855

equation (10) was used to obtain the beam frequencies. The beam is divided into 20
equally spaced elements and at each node three degrees of freedom are considered;
the de#ection of the beam, the slope of the beam, and the curvature. In general, the
minimum continuity for the de#ection of the beam dictates that only the de#ection
and the slope be used as the degree of freedom at each node. This requirement
makes the sti!ness matrix [:1 /A /A dx] in equation (4) well de"ned. The moving
0 i j
mass, however, imposes an additional requirement that the term in equation (4)
containing [/ /A] remain continuous. This condition requires a higher degree
i j x/s(t)
of continuity in the "nite element discretization, therefore, the curvature is used as
an additional degree of freedom. Table 3 lists the "rst few frequencies for the
mass}beam system obtained using "nite elements for the parameters used in the
simulations. The natural frequency of the beam u , obtained using equation (10) is
1
also shown in the table for comparison.
Figures 14 and 15 show the results obtained for the mass and the beam,
respectively, for m"1)0, s "0)9 and initial values t "0)1 and s "0)90001. The
e d 0
parameters used for the simulations presented in this section are tabulated in
Tables 2 and 4. Figures 14(c)}(e) show the spectrogram where the higher
frequencies are also included and Figure 14(f) shows the power spectrum. The
power spectrum is obtained by applying the Hann window to the time series and
using the FFT to obtain the spectrum. To reduce the variance, FFTs are obtained
for a number of data segments (see Table 4 for the number of segments used) and
the results are averaged. The power spectral density (PSD) is computed by taking
the mean squared amplitude of the transformed data. This same format is used for
all the simulation results presented in this section.
From the power spectrum, Figure 14(f), major peaks are observed at even
multiples of u (e.g. 2u , 4u , 6u 2). As the peaks approach the second
1 1 1 1
frequency of the beam (19)5378), the energy corresponding to these peaks increases
slightly. This is much clearer in the beam plot, Figure 15(f), where the energy
increases as the peaks approach the second and the third frequencies, 19)5378 and
60)470, respectively. For the beam, the major peaks in the power spectrum occur
at odd multiples of u (u , 3u ,2). The spectrograms (Figures 14(c)}(e) and
1 1 1
 BEAM WITH MOVING MASS 1049

Figure 14. Mass response !1 : 2 IR, m"1)0, s "0)9, s "0)90001, and v "0)1. (a)
e 10 t0
Perturbation solution p"!0)0008, (b) numerical solution, (c)}(e) spectrograms, and (f) power
spectrum.

15(c)}(e)) show the time variation of the frequencies. For the moving mass, the
energy corresponding to the primary frequency 2u (Figure 15(e)) becomes quite
1
small when the amplitude reaches the minimum, and the energy in the harmonics of
the moving mass (4u and 10u shown in Figures 14(c) and (d)) crests, indicating
1 1
a change in the frequency content of the mass response when the amplitude is
decreasing. The time variation of 4u (Figure 14(d)) shows dissipation of energy to
1
the side bands when the amplitude of the moving mass reaches a minimum;
1050 S. A. Q. SIDDIQUI E¹ A¸.

Figure 15. Tip de#ection !1 : 2 IR, m"1)0, s "0)9, s "0)90001, and v "0)1. (a) Perturbation
e 10 t0
solution p"!0)0008, (b) numerical solution, (c)}(e) spectrograms, and (f) power spectrum.

however, when such a frequency is close to a higher fundamental frequency of the

beam, like 10u (Figure 15(c)) is to the second frequency of the beam, the
1
dissipation of energy to the side bands does not occur. Such a behavior can be
observed in all the simulations presented in this paper. For the beam the beating
pattern can be seen in u and 3u . However the higher frequency 11u , shown in
1 1 1
Figure 15(c), does not show the beating pattern, as this frequency is close to the
second frequency of the beam which is not in resonance.
Next, for the same parameters as in Figures 14 and 15, di!erent initial conditions
are used (v "0)00001 and s "0)95). Figures 16 and 17 show the results. Since in
t0 0
 BEAM WITH MOVING MASS 1051
TABLE 4
Comparison between perturbation and numerical solutions22

Parameter set 2
m"1)0, s "0)9, u using 20 "nite elements"1)758689
e 1
u using one mode of a cantilever beam"1)763542
1
Initial values Initial values
s "0)90001, v "0)1 s "0)95, v "0)00001
0 t0 0 t0
Figure 14, 15 16, 17
Model PM1, NM2 PM1, NM2
p !0)0008 !0)0097
Dt(average) 0)00442 0)00533
Spectrogram
No. of segments 64 64
Segment size 256 256
Power spectrum
No. of FFTs 2 6
Segment size 65536 65536

this case the system is predominantly excited by the moving mass, the energy
corresponding to the higher frequencies of the beam is notably very low as can be
seen by comparing Figures 15(f) and 17(f). The spectrograms for the moving mass
show that initially the energy of the system is concentrated in the primary frequency
of the moving mass 2u and as the amplitude starts decreasing some of the energy
1
moves into the harmonics of the moving mass, e.g. 4u , but predominantly the
1
energy is transferred to the frequencies in the beam u , 3u and 5u . The power
1 1 1
spectrum for the moving mass and the beam are similar to those observed in
Figures 14 and 15 except that the PSD is signi"cantly lower for the higher
frequencies.

7. SUMMARY
Dynamics of a #exible cantilever beam carrying a moving spring-mass were
investigated using perturbation, and numerical methods. Time-frequency analysis
was also performed. The di$culty in obtaining a numerical solution for non-linear
systems is often one of the motivating reason for perturbation analysis. However,
perturbation methods do not always provide a closed-form solution like the one
obtained in this work and a combined analytic}numeric approach is often
required. Comparison between closed-form analytical and numerical solutions is
therefore uncommon. In this work, the perturbation solution for modal amplitudes
is matched with the numerical solution by selecting a value for the detuning
parameter p. A numerical solution obtained under perfect 1 : 2 resonance
conditions when compared with a perturbation solution under perfect 1 : 2
resonance conditions is expected to re#ect di!erences as a result of the di!erent
models for the two cases and also due to neglecting the higher order terms in the
1052 S. A. Q. SIDDIQUI E¹ A¸.

Figure 16. Mass response !1 : 2 IR, m"1)0, s "0)9, s "0)95, and v "0)00001. (a)
e 10 t0
Perturbation solution p"!0)0097, (b) numerical solution, (c)}(d) spectrograms, and (e) power
spectrum.

perturbation solution. The approach under in this work was to quantify these
di!erences using the detuning parameter p. Comparison of the perturbation and
the numerical solutions show that when the motion is predominantly bi-periodic
(two fundamental frequencies, one u and the other the beating frequency n/q and
1 b
their harmonics) the results match very well. Using the closed-form solution, an
extensive parametric analysis was carried out which identi"es regions of strong
non-linear coupling between the beam and the moving mass and gives the change
in some of the important properties of the solution such as the beating period, and
the maximum and minimum amplitudes with the detuning parameter, initial values
and the equilibrium position of the moving mass on the beam.
An analytical solution provides qualitative results and allows for investigation of
non-linear behavior of amplitude and phase modulation under internal resonance
 BEAM WITH MOVING MASS 1053

Figure 17. Tip de#ection !1 : 2 IR, m"1)0, s "0)9, s "0)95, and v "0)00001. (a)
e 10 t0
Perturbation solution p"!0)0097, (b) numerical solution, (c)}(e) spectrograms, and (f) power
spectrum.

conditions. A numerical solution, on the other hand, provides quantitative results

and is also a necessary step in performing spectral analysis. Non-linear systems can
exhibit changes in frequencies with time. This e!ect was investigated using
time-frequency analysis.

ACKNOWLEDGMENTS
This work was supported by the Natural Sciences and Engineering Research
Council of Canada (NSERCC).
1054 S. A. Q. SIDDIQUI E¹ A¸.

REFERENCES
1. G. G. STOKES 1849 ¹ransactions of Cambridge Philosophical Society 8, 707. Discussion
of a di!erential equation relating to the breaking of railway bridges.
2. D. C. D. OGUAMANAN, J. S. HANSEN and G. R. HEPPLER 1998 Journal of Sound and
<ibration 213, 889}906. Dynamic response of an overhead crane system.
3. R. S. AYRE, L. S. JACOBSON and C. S. HSU 1951 Proceedings of the 1st ;.S. National
Congress on Applied Mechanics, 11}16 June, Chicago, 81}90. Transverse vibration of
one and two-span beams under the action of a moving mass load.
4. C. E. INGLIS 1934 A Mathematical ¹reatise on <ibrations in Railway Bridges. London:
Cambridge University Press.
5. A. HILLERBORG 1951 Dynamic In-uences of Smoothly Running ¸oads on Simply
Supported Girders. Stockholm, Sweden: Kungl Tekniska Hogskolan.
6. A. L. FLORENCE 1965 Journal of Applied Mechanics 351}357. Traveling force on
a timoshenko beam.
7. C. R. STEELE 1967 Journal of Applied Mechanics 111}118. The "nite beam with
a moving load.
8. L. FRYBA 1972 <ibration of Solids and Structures under Moving ¸oads. Groningen, The
Netherlands: Noordho!.
9. M. M. STANISIC, J. A. EULER and S. T. MONTGOMERY 1974 Ingenieur-Archiev 43,
295}305. On a theory concenring the dynamical behavior of structures carrying moving
masses.
10. T. HAYASHIKAWA and N. WATANABE 1981 Journal of the Engineering Mechanics Division,
ASCE 107, 229}246. Dynamic behaviour of continuous beams with moving loads.
11. M. M. STANISIC 1985 Inginieur-Archiv 55, 176}185. On a new theory of the dynamical
behavior of structures carrying moving masses.
12. A. O. CIFUENTES 1989 Finite Elements in Analysis and Design 5, 237}246. Dynamic
response of a beam excited by a moving mass.
13. Y. H. LIN and M. W. TRETHEWAY 1990 Journal of Sound and <ibration 136, 323}342.
Finite element analysis of elastic beams subjected to moving dynamic loads.
14. J. A. BBADEYAN and S. T. ONI 1995 Journal of Sound and <ibrations 182, 677}695.
Dynamic behavior of beams and rectangular plates under moving loads.
15. H. P. LEE 1996 Journal of of Sound and <ibration 191, 289}294. Dynamic response of
a beam with a moving mass.
16. G. MICHALTSOS, D. SOPHIANOPOULOS and A. N. KOUNADIS 1996 Journal of of Sound and
<ibration 191, 357}362. The e!ect of a moving mass and other parameters on the
dynamic response of a simply supported beam.
17. K. HENCHI, M. FAFARD, G. DHATT and M. TALBOT 1997 Journal of of Sound and
<ibration 199, 33}50. Dynamic behaviour of multi-span beams under moving loads.
18. A. MESSAC 1994 AIAA/ASME/ASCE/AHS/ASC 35th Structures, Structural Dynamics
and Materials Conference, 18}20 April, Hilton Heads, SC AIAA-94-1614. Dynamics
modelling of a #exible vehicle moving on the #exible rails of the #exible structure.
19. M. F. GOLNARAGHI 1991 Mechanics Research Communications Journal 18, 135}143.
Vibration suppression of #exible structures using internal resonance.
20. M. F. GOLNARAGHI 1991 Journal of Dynamics and Control 1, 405}428. Regulation of
#exible structures via nonlinear coupling.
21. S. A. Q. SIDDIQUI, M. F. GOLNARAGHI and G. R. HEPPLER 1998 Nonlinear Dynamics 15,
137}154. Dynamics of a #exible cantilever beam carrying a moving mass.
22. F. KHALILY, M. F. GOLNARAGHI and G. R. HEPPLER 1994 Nonlinear Dynamics 5,
493}513. On the dynamic behavior of a #exible beam carrying a moving mass.
23. L. D. ZAVODNEY and A. H. NAYFEH 1989 International Journal of Nonlinear Mechanics
24, 105}125. The nonlinear response of a slender beam carrying a lumped mass to
a principal parametric excitation; theory and experiment.
24. A. H. NAYFEH, J. F. NAYFEH and D. T. MOOK 1992 Nonlinear Dynamics 3, 145}162. On
the methods for continuous systems with quadratic and cubic nonlinearities.
 BEAM WITH MOVING MASS 1055
25. M. PAKDEMIRLI and A. H. NAYFEH 1994 Journal of <ibration and Acoustics 116, 433}439.
Nonlinear vibrations of a beam spring mass system.
26. T. J. ANDERSON, B. BALACHANDRAN and A. H. NAYFEH 1994 Journal of <ibration and
Acoustics 116, 480}484. Nonlinear resonances in a #exible cantilever beam.
27. C.W. GEAR 1977 Formulations and Computational Algorithms in Finite Element Analysis
(K. Bathe, J. T. Oden and W. Wunderlich, eidtors), Chapter 24, 691}717. Boston: The
Massachusetts Institute of Technology. Ordinary di!erential equation techniques for
partial di!erential equations.
28. E. B. BECKER, G. F. CAREY and T. J. ODEN 1981 Finite Elements on Introduction, Vol. 1.
Englewood Cli!s, NJ: Prentice-Hall, Inc.
29. S. A. Q. SIDDIQUI 1998. Ph.D. ¹hesis, Department of Mechanical Engineering, ;niversity
of =aterloo, Ontario, Canada. Nonlinear beam behaviour with a moving mass.
30. A. H. NAYFEH and D. T. MOOK 1979 Nonlinear Oscillations. New York: Wiley.
31. W. H. PRESS, S. A. TEUKOLSKY, W. T. VETTERLING and B. P. FLANNERY 1992 Numerical
Recepies: ¹he Art of Scienti,c Computing. Cambridge: Cambridge University Press,
second edition.
32. L. COHEN 1995 ¹ime-Frequency Analysis. Englewood Cli!s, NJ: Prentice-Hall PTR.