International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

www.elsevier.com/locate/nlm

Non-linear dynamic response of a rotating radial Timoshenko beam

with periodic pulse loading at the free-end
Sunil K. Sinha∗
G.E. Aircraft Engines, M.D. A-412, General Electric Company,1 Neumann Way, Cincinnati, OH 45215, USA

Received in revised form 3 December 2003; accepted 10 May 2004

Abstract
Consideration is given to the dynamic response of a Timoshenko beam under repeated pulse loading. Starting with the
basic dynamical equations for a rotating radial cantilever Timoshenko beam clamped at the hub in a centrifugal force field,
a system of equations are derived for coupled axial and lateral motions which includes the transverse shear and rotary
inertia effects, as well. The hyperbolic wave equation governing the axial motion is coupled with the flexural wave equation
governing the lateral motion of the beam through the velocity-dependent skew-symmetric Coriolis force terms. In the analytical
formulation, Rayleigh–Ritz method with a set of sinusoidal displacement shape functions is used to determine stiffness, mass
and gyroscopic matrices of the system. The tip of the rotating beam is subjected to a periodic pulse load due to local rubbing
against the outer case introducing Coulomb friction in the system. Transient response of the beam with the tip deforming due
to rub is discussed in terms of the frequency shift and non-linear dynamic response of the rotating beam. Numerical results
are presented for this vibro-impact problem of hard rub with varying coefficients of friction and the contact-load time. The
effects of beam tip rub forces transmitted through the system are considered to analyze the conditions for dynamic stability
of a rotating blade with intermittent rub.
䉷 2004 Elsevier Ltd. All rights reserved.

1. Introduction to numerous practical usage such as jet engine blades,

helicopter rotor blades, airplane propellers, satellite
The study of transient dynamics of vibro-impact re- antennas, robotic arms and other turbo-machinery
sponse has applications in several areas of engineer- applications, have been investigated for a long time.
ing mechanics. The need to consider shear deforma- In the aerospace applications, due to on-going need
tion in case of impact on a beam was first discussed to reduce weight of every component, an attempt is
by Flügge [1] and again reemphasized by Timoshenko made to improve the dynamic characteristics of the
et al. [2]. Rotating beams, which have importance due rotating blades by increasing its ‘slenderness ratio d̄’,
which effectively means gaining as much on the mo-
∗ Tel.: +1-513-243-3200; fax: +1-513-243-8091. ment of inertia of the cross-section as possible, with-
E-mail address: [email protected] (S.K. Sinha). out any significant increase in its cross-sectional area
0020-7462/$ - see front matter 䉷 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijnonlinmec.2004.05.019
114 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

which affects its weight. Dynamic characteristics of instability in the blade, which is further complicated
such beams are better represented by the Timoshenko’s due to the non-linear effect of the Coulomb friction
beam formulation, which takes into account of the at the contact location. It is known that the rub forces
transverse shear as well as the rotary inertia; rather can be extremely high causing severe damage, but due
than the simple Euler–Bernoulli’s beam theory which to the short duration involved, determining its true
neglects these two effects. Short and stubby blades in magnitude has been a real challenge. During the rub,
the upper stages of the high-pressure compressors also the peak amplitude of the compressive radial impact
have relatively high ‘slenderness ratio d̄’, where the pulse can be many times more than the Euler’s static
classical Euler–Bernoulli’s beam theory over-predicts critical buckling load Fcr for the similar size beam-
the natural frequencies of the blades by a big margin. column [3]. In short, rubbing of blade tips against the
Almost all the research attention in the area of Timo- stationary outer case is a highly non-linear contact-
shenko beams have been focussed on the eigenvalue impact problem of a rotating Timoshenko beam, and
solutions such as free vibration problems and the sta- until now no one has attempted to solve it analytically,
bility analysis in frequency domain. However, little re- which covers its full dynamic characteristics including
search has been done on the forced vibration problems the Coriolis forces due to the coupling of the flexural
because of its complexity, and no closed-form solu- and axial motions of the beam.
tion of such dynamic system has ever been presented. After the importance of transverse shear and rotary
Obviously, some enormous difficulties exist unavoid- inertia in the beam formulation was shown by Tim-
ably in the study of non-linear dynamic behavior of oshenko, many different aspects of his beam theory
Timoshenko beams with contact-impact applications, have been studied by several authors over the past
especially due to the complex dynamics of a vibro- 40 years. Huang [4] was the first one to determine
impact system with multiple degrees-of-freedom. the frequencies and corresponding mode shapes for
In any high-performance turbo-machinery, rub be- the flexural vibrations of Timoshenko beams for 6
tween the stator structure and the rotor blade is a com- common types of boundary conditions at the ends.
monly occurring vibro-impact problem. For example Hagedorn and Koval [5] studied the parametric sta-
in a jet engine high-pressure compressor and fan, usu- bility of a Timoshenko beam subjected to a periodic
ally, the outer casing ovalization results in a blade or a axial load. Stafford and Giurgiurtiu [6] used a semi-
sector of blades rubbing against the case. Specifically, analytic method for obtaining the frequencies and
in aero-engines, the case distortion may be caused ei- mode shapes for uncoupled lead-lag and flapping vi-
ther by a temperature gradient or due to various ma- brations of a rotating Timoshenko beam. Irie and his
neuvering loads usually during take-off and landing of coworkers [7] used spline function method to show
the aircraft. Depending upon the stabilizing or desta- the dynamic stability of Timoshenko beams with fol-
bilizing nature of the rub, engine may correct itself lower forces. Nelson [8] developed a rotating shaft
and will keep on running or, it may cause extensive finite-element utilizing Timoshenko beam theory for
damage to the rotor blades to a point that the engine establishing the shape functions. Abbas [9] used a
may have to be shut-down. Higher energy-efficiency finite-element method to investigate the effect of ro-
of the engine is achieved by reducing the tip clearance tational speed and root flexibilities on the static buck-
between the rotor and the stator components, but as ling loads and on the regions of dynamic stability of
the clearance is reduced; the probability of rub taking a Timoshenko beam. Sinha [10] used the Galerkin’s
place during small changes in the operating conditions method to analyze the effect of oscillating axial load
also increases. For example in a large jet engine, the on the parametric stability of a simply supported
typical tangential velocity of the blade tip will vary Timoshenko beam. Farris and Doyle [11] applied the
in the range of 400–600 m/s. During the initiation of spectral analysis method to study the impact-induced
local tip rub, the typical contact duration would last wave propagation in a split Timoshenko beam. Datta
anywhere between 0.1 and 0.8 ms, and usually it will and Ganguli [12] used a Ritz solution, that employed
repeat for several revolutions with one pulse per revo- polynomial trial functions, to treat the problem of
lution. The periodic nature of the radial impact pulse rotating Timoshenko beam that had a damaged sec-
along the beam-column axis may initiate parametric tion. Laura and his colleagues [13] provide the most
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 115

comprehensive list of literature published to-date in et al. [29] were the first ones to consider the effect of
the area of Timoshenko beams. Lee and Lin [14,15] centrifugal stiffening in a rotating Euler–Bernoulli’s
studied the free vibration problem of a rotating ra- beam model during rub. Later on, Ahrens et al. [30]
dial Timoshenko beam with attached masses and reported a transfer function approach to measure the
restrained root. Tong and his coworkers [16] carried blade-tip contact force in a rubbing blade. Recently,
out the vibration analysis of Timoshenko beams with Sinha [31] analyzed the tip rubbing of a fully bladed
non-homogeneity and varying cross-section. White and rigid-disk on a flexible rotor shaft, which took
and Heppler [17,18] reported the generalized expres- into account of both centrifugal forces in the blade as
sions for the frequency equations and mode shapes well as the gyroscopic moments due to shaft bending.
of a Timoshenko beam with rigid bodies attached to In that study, the flexible blades are formulated based
its ends. Lueschen and Bergaman [19] used a Green’s upon the Euler–Bernoulli’s beam theory.
function approach to determine the forced response of Moon [32] was first to point out that repeated im-
a Timoshenko beam. Lin [20] also used the Green’s pact situation such as one in the impactor-printer ap-
function approach to obtain the exact solutions for plication is highly likely to result in chaotic vibration.
a circular Timoshenko beam with non-homogeneous After Whiston [33] studied the nature of the strange at-
boundary conditions. Oguamanam and Heppler [21] tractors which occur in vibro-impact systems, a wealth
studied the stiffening effect of rotational speed on the- of information is available analyzing the different as-
forced vibrational amplitude of Timoshenko beams pects of the non-linear characteristics of beam oscilla-
with hub radius equal to zero for a prescribed torque tions. Tang and Dowell [34] studied the effect of higher
and velocity profiles. Lin [22] investigated the in- modes on the chaotic oscillations of a buckled beam
fluence of elastic root restraints on the natural fre- under forced external excitation. They determined the
quencies of the non-uniform Timoshenko beams. Yu threshold force at which the motion becomes chaotic.
and Cleghorn [23] determined the natural frequen- Shaw and Shaw [35] have shown as to how a dynamic
cies and corresponding mode shapes for a spinning response of a two-degrees-of-freedom impacting sys-
stepped Timoshenko beam. Using power function to tem made up of an inverted pendulum can become
represent the deformed shape, Zhou and Cheung [24] chaotic through bifurcation. Wen and Xie [36] have
investigated the free vibrations of a wide range of reported a non-typical route to chaos of a two-degrees-
tapered Timoshenko beams. Hutchinson [25] reported of-freedom vibro-impact system. That is the period-
a new method to determine the Timoshenko’s shear doubling bifurcation and then the system turns into
coefficient
for a number of different cross-sections the quasi-periodic response. Due to multiple degrees-
especially for a tip-loaded cantilever beams. of-freedom dynamical system, the current problem of
In the current work, our main focus is on the non- chaotic vibrations in a rotating Timoshenko beam is
linear transient dynamic response of a rotating Timo- an interesting but a very complex problem to analyze.
shenko beam under pulse loading. As one can see in In the following sections, the present paper analyzes
the above list of the published literature, almost all the the effect of intermittent coulomb friction near the
work in the area of Timoshenko beam is concentrated tip with the flexible Timoshenko beam mounted on
on the natural vibration frequency and the mode shape a rigid disk. Starting from the basic deformation and
determination. Simo and Vu-Quoc [26] discuss the role velocity equations along with the rotary inertia and
of non-linear structural theories in the transient dy- gyroscopic effect terms, a complete set of coupled
namic analysis of flexible structures performing large dynamic equations has been derived for this problem.
motions. Segalman and Dohrmann [27] considered the
effect of all the spin involved dynamic non-linearities
up to the second order by using a system of coupled de- 2. Rotating cantilever Timoshenko beam
formation modes. Choy et al. [28] have studied a true formulation for flexible blades
rotor-case rub phenomena and its effect due to friction
in quite detail, however, they have neglected the effect As discussed in the previous section, the rotating
of stress stiffening of blades due to rotation, which is turbine blade dynamics have been studied by several
a major parameter in pulse buckling during rub. Jiang authors in the past with different levels of complexity
116 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Fig. 1. Rub forces Fa and Fa on a typical Jet Engine Fan Rotor Blade spinning with constant angular velocity vector  with respect to
a stationary global frame of reference and the resulting deformations ,  and .

such as Timoshenko’s beam formulation and large de- ometrically non-linear finite-strain of the mid-surface
formations of the free-end. In this analysis, we con- should be considered; which per Green–Saint Venant
sider that elastically deformable radial blades of outer definition in terms of the current notation is written as
radius ‘R’ with the stagger angle ‘’, which the blade
   2 
chord makes with the engine axis, are mounted on j 1 j 2 j
Spanwise Strain s = + + .
a rigid disk of hub radius ‘r’ (see Fig. 1). The indi- js 2 js js
vidual blades behave like a cantilever beam-column (1)
of span length ‘L’ and are subjected to a centripetal
force field generated due to the rotor spin velocity ‘’ For the analytical derivation, we will use the usual no-
with the blade tip rub load Fa (+ sign: tension and − tations such as, Elastic Young’s modulus ‘E’, Shear
sign: compression) acting along the beam axis. The tip modulus ‘G’, material mass density ‘ ’, Poisson’s ra-
rub force Fa is a dynamic contact load, which is non- tio ‘ ’, and cross-sectional area ‘A’. It should be noted
zero only during the tip travel through the rub zone that the stagger angle  is defined such that, in a tur-
on the stator and will always have a (−) sign due to bine  and  have the same sign, and in a compressor
the contact load being compressive in nature. In ad-  and  have the opposite sign. We define two differ-
dition, the blade is deformed in bending by applying ent local coordinate systems called ‘Axial-Tangential-
bending moment about its chord in such a way that a Radial’ with unit vectors as (êa , êt , êr ) and, ‘Chord-
typical cross-section of the deformed blade produces Normal-Span’ with unit vectors as (êc , ên , ês ) (see
the cross-section rotation ‘’, the lateral deflection ‘’ Fig. 2), respectively.
and axial deflection ‘’. It is assumed that all three With respect to the stationary global cartesian unit
components of deformation are functions of spatial vectors (î, jˆ, k̂), the local unit vectors attached to the
coordinate ‘s’, measured along the beam axis and the beam (êc , ên , ês ), rotating at a constant angular veloc-
temporal parameter time ‘t’. It is also assumed that ity  such that = t, are related to each other as
the minor principal moment of inertia of the blade  

êc − sin  cos cos  sin  sin
cross-section ‘I ’ coincides with the chord direction so
ên = − cos  cos − sin  cos  sin
that under pure bending moment the blade lateral de-
ês sin 0 cos
flection (s, t) takes place in the direction normal to
 
the chord with the neutral surface passing through the î
radial-chord plane. According to Simo and Vu-Quoc × jˆ . (2)
[26] for large transient motions of the beam, the ge- k̂
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 117

Fig. 3. Cross-section of the Timoshenko beam as viewed from the

free-end of the cantilever beam along the local radial direction
(z = distance of a typical bending-induced deformed layer from
the mid-surface). Beam cross- section rotated by an angle  with
respect to the axis of spin.
Fig. 2. Schematic view of the rub zone on the stator with respect to
the stationary global frame of reference, and the relative orientation
of the local coordinate system attached to the beam spinning with
angular velocity vector  during a typical rub (time period of one
The angular velocity vector of the beam-column due
revolution = 2 / second). Rub zone contact load of magnitude
Fa = F (t) active for the contact duration time = tc second. to spin velocity  is

[ cos ]êc + [− sin ]ên + [0]ês . (5)

The deformation vector of a typical point on the beam
with large rotation at a distance ‘z’ from the mid- Thus, the velocity vector of any point on the beam is
surface can be written (see Fig. 3) as the cross-product of the angular velocity vector ‘’
[0]êc + (s, t)ên + [s(cos (s, t) − 1) and its position vector ‘r’:
+ (s, t) − z sin (s, t)]ês .  
j j j
Velocity = v0 + X r = ên + −z ês
In the present analysis, it is considered that during the jt jt jt
transient motion of the beam, the cross-section rotation ês êc ên
term (s, t) is sufficiently small such that, cos  ≈ 1 + 0  cos  − sin 
and sin  ≈ . Thus under the small rotation assump- s + r +  − z 0 z+
tion, the deformation vector of a typical point on the ˙ ˙
= [ − z + (z + ) cos ]ês
beam simplifies as − [(s + r +  − z) sin ]êc
[0]êc + (s, t)ên + [(s, t) − z(s, t)]ês . (3) + [˙ − (s + r +  − z) cos ]ên
Here, the total slope  (s, t) of the lateral deformation = [[˙ − z ˙ + (z + ) cos ] sin
of the beam is the sum of the rotation (s, t) and the + [(s + r+−z)−˙ cos ] cos ].î
rotation of the cross-section due to shear force Q(s, t) − [˙ sin ].jˆ
expressed as −(Q/
AG), i.e. + [[˙ − z ˙ + (z + ) cos ] cos
Q(s, t) − [(s + r +  − z)
 (s, t) = (s, t) − . (4)
− ˙ cos ] sin ].k̂.

AG (6)
118 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

The kinetic energy ‘T ’ of the beam is expressed as The typical Lagrangian equation in terms of dependent
variable (s, t) can be written as
A L
   
T = d j j d j
2 0 − + = q.
 ˙  dt j˙ j ds j
˙ + (z + ) cos ] sin
[[ − z
Thus, using the Lagrangian ‘’ in the Lagrange’s
 +[(s + r +  − z) 
  equation yields the following three coupled equations
 −˙ cos ] cos ]2 + [˙ sin ]2 
×  ds. (7) of motion for dependent variables ,  and , respec-
 +[[˙ − z
˙ + (z + ) cos ] cos 
  tively in a local frame of reference attached to the
−[(s + r +  − z)
−˙ cos ] sin ]2 rotating beam-column with angular velocity ‘’ as
   
j j j
On using the minor principal moment of inertia ‘I ’ EI +
AG −
of the cross-section (see Fig. 3) about the neutral axis js js js
 2 
(axis along the local êc direction) and also the property j  d
− I − cos  −   = 0,
2
(10a)
of the CG of the cross-section as jt 2 dt
  
+d/2 j j j
I= z2 dA = c z2 dz and − (Fa + A cf ) +
AG −
−d/2 js js js
 2
Area
j  d
+d/2 + A − (s + r + ) cos 
z dA = c z dz = 0, jt 2 dt
−d/2

j
Area −2 cos2  − 2  cos  = q(s, t), (10b)
jt
we have the Potential energy ‘U ’ due to bending, trans-  2
verse shear and centrifugal stresses ‘ cf ’ as well as the j2  j  d
E 2= − 2 (s + r + ) +  cos 
axial force ‘Fa ’ given as js jt 2 dt

L
j
1 +2  cos  . (10c)
U= [EI ( )2 +
AG( − )2 + AE(s )2 ] ds jt
2 0
1 L A L In Eq. (10b), q(s, t) accounts for the distributed ex-
− Fa ( )2 ds − cf ( )2 ds. (8) ternal transverse load on the beam-column, if there is
2 0 2 0
any. In the above equations, it should be noted that
The simplified form of Lagrangian ‘’ for this rotating in a rotating beam; its lateral motion (s, t) is cou-
Timoshenko’s beam-column motion and deformation pled with the longitudinal motion (s, t) due to the
is Coriolis effect, which introduces velocity-dependent
skew-symmetric terms in the equations of motion. For
1 L linearization, we have used infinitesimal definition of
=T −U =
2 0 longitudinal strain as s = j/js, and on substituting
  [[˙ −z˙ +(z+) cos ] sin  in place of R − r = L, the we can write the centrifugal
   +[(s + r +  − z  )  stress relation in terms of 2 and the longitudinal
  −˙ cos ] cos ]2  force Fa , which gives
  
 A  +[˙ sin ]2 + [[˙ − z ˙ 
   Fa
  +(z + ) cos ] cos  (s) = +
  cf
A
×   ds.
 −[(s + r +  − z)  (R 2 − r 2 − 2rs − s 2 ) Fa
 −˙ cos ] sin ]2  = 2 + (11)
  A
−[EI ( )2 +
AG( − )2  2
 
 +AE( )2 ]  and also,
s
+(A cf + Fa )( )2 j (s)
= − 2 (s + r) = − 2 (s + R − L). (12)
(9) js
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 119

∞

The external shear force due to Coulomb friction on
the blade tip Fa is acting opposite to the direction (s, t) = Vn (s)Yn (t)
n=0
of motion, which in the local coordinate (êa , êt , êr ) ∞
system would be in the direction of êt . Its component  [1 − cos n s]
= Yn (t), (18b)
along ên would act like a point-load in the beam lateral n
n=0
direction, and as such mathematically with the use of
∞

Dirac’s Delta function (s − L) can be treated like a
continuous external force, which is written as (s, t) = Wn (s)Zn (t)
n=0
∞
q(s, t) = −Fa cos (s − L). (13) [sin n s]
= Zn (t) (18c)
n
The geometric boundary conditions for the Timo- n=0
shenko’s beam under consideration are as follows: (where, n = (2n − 1) /2L).
The first terms in each of the above sequence of trial
(0, t) = 0, (0, t) = 0, (0, t) = 0. (14) displacement functions, corresponding to n = 0, per-
tain to the rigid body displacement of the beam caused
The natural boundary conditions at the free-end of the
by the movement of the clamped end at s = 0. Since
cantilever beam for (s = L) are as follows:
our interest is mainly in the strain-energy related de-
j formation of the beam and not in the rigid body move-
(Bending Moment)at s=L = EI = 0, (15) ment, we will consider the terms in the series starting
js s=L
at n=1. Thus, the substitution of the assumed displace-
(Shear Force ‘Q’)at s=L ment shape functions expressed as Eqs. (18a)–(18c) in
  the equations of motion described by Eqs. (10a)–(10c)
j along with the assumption of constant angular veloc-
= −
AG −
js s=L ity  yields the following set of equations:
  ∞
j 
= −Fa cos  + Fa · (16) [−EI X n Un −
AG(Yn Vn − Xn Un )
js s=L
n=1
j + I (Ẍn Un − 2 Xn Un )] = 0, (19a)
(Axial Force)at s=L = EA = Fa . (17)  
js s=L − A 2 (R 2 −r 2 −2rs−s 2 )
Yn V n
2
 + A2 (r + s)Y V 
Due to the pulse loading nature of the axial force,  n n 
∞  −F (t)Y V 
we will assume the longitudinal load Fa to be time-  n n 
 −
AG[Y V − X U ]  = q(s, t),
dependent and henceforth will be written as F (t). Us-  n n n n 
n=1 
 + A(Ÿn Vn −  cos Yn Vn 
2 2 
ing the Rayleigh–Ritz method [37], the solution of
 −2 cos Ż W ) 
the earlier set of Eqs. (10a)–(10c) can be assumed n n
with a set of displacement shape functions Un (s), −F (t)Yn Vn (s − L)
Vn (s) and Wn (s) such that they satisfy all the geomet- (19b)
ric boundary conditions arbitrarily. Thus, using cor- ∞

responding dimensionless time-dependent generalized [−EZ n Wn + (Z̈n Wn − 2 Zn Wn
coordinates Xn (t), Yn (t) and Zn (t) respectively, we n=1
assume the solution in the form of the following sets + 2 cos Ẏn Vn )] = 2 [s + r]. (19c)
of infinite trigonometric series:
In the above sets of equations, the spatial derivative
∞
 ∞
 terms can be written as a function of a set of differ-
(s, t) = Un (s)Xn (t) = [sin n s]Xn (t), ential operators R. In addition, Eqs. (19a)–(19c) are
n=0 n=0 complete only when all the terms in the infinite se-
(18a) ries for the displacement functions Un (s), Vn (s) and
120 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Wn (s) are considered. In the ideal condition, they may  −EA  L W W ds Z 

satisfy all the geometric as well as force boundary 0 n M n

∞  +EA[W  Ln
(L)Wm (L)]Zn 
conditions of the present problem, however, it is not  
 + A( 0 Wn Wm ds Z̈n 
necessary in general. On applying the Rayleigh–Ritz   
n=1  −2 
L
method with relaxed boundary condition requirements W W
n m ds Z n
0 L
[37] and the assumed displacement functions, one ob- +2 cos  0 Vn Wm ds Ẏn )
tains a set of ordinary differential equations in terms L
of time-dependent variables of which a typical mth set = A2 (s + r)Wm ds + F (t)Wm (L). (21c)
0
of equation for each of the trial function terms viz.
Um (s), Vm (s) and Wm (s) can be written as On replacing the subscript indices in Eqs. (21a)–(21c),
L such that ‘i’ represents the row number and ‘j ’ repre-
R1 (, , )Um (s) ds = 0, (20a) sents the column number, respectively, one obtains
0
 L 
L L −EI ( Uj Ui ds)Xj
R2 (, , )Vm (s) ds = q(s, t)Vm (s) ds, ∞ 
0
L
0 0   +(
AG − I 2 ) 0 Uj Ui dsX j 

(20b)  L  = 0,
 −
AG 0 Vj Ui ds Yj 
j =1 L
L
+ I 0 Uj Ui ds Ẍj
R3 (, , )Wm (s) ds (22a)
0
L
 L 
= 2 (s + r)Wm (s) ds. (20c) − A2 (R 2 −r 2 −2rs−s 2 )
V j Vi dsY j
0 2
0
 2 L
 
 + A 0 (r + s)Vj Vi dsY j 
Since, the assumed displacement functions Un (s),  L 
 − A2 cos2  0 Vj Vi ds Yj 
Vn (s), Wn (s) do not satisfy all the force bound-  L 
∞  
ary conditions necessarily, then with the unbalanced  −(
AG + F (t))( 0 Vj Vi ds 
= 
weighted-residual force and moment terms at the free  −Vj (L)Vi (L))Yj 
j =1  L 
boundary (s = L), we have  
L  +
AG( 0 Uj Vi ds − Uj (L)Vi (L))Xj 
   L 
−EI 0 Un Um ds Xn  −2 A cos  0 Wj Vi ds Żj 
L L

∞  −
AG( 0 Vn Um ds Yn 

 L + A 0 Vj Vi ds Ÿj
 
 − 0 Un Um ds Xn )  = 0, (21a) = −F (t) cos Vi (L), (22b)
 L 
n=1  + I ( Un Um ds Ẍn 
 0L  L 
−2 0 Un Um ds Xn ) −EA( 0 Wj Wi ds
  
∞  −Wj (L)Wi (L))Zj

L 2 2   
− A2 0 (R −r −2rs−s
2)
Vn Vm ds Yn  L 
 L 2
  − A2 0 Wj Wi ds Zj 
 + A2 0 (r + s)Vn Vm ds Yn   L 
j =1  +2 A cos 
 L  V W
j i ds Ẏ j
 −
AG[ V Vm ds Yn  L 0
  0 n  + A 0 Wj Wi ds Z̈j
 L 
 − 0 Un Vm dsX n ]  L
 L  = A2 (s + r)Wi ds + F (t)Wi (L).
∞  −F (t) V V ds Y  (22c)
 n m n 
 + A( L V V ds Ÿ
0 0

 n m n 
n=1  0 L 
 − cos  0 Vn Vm ds Yn
2 2 
  
 −2 cos  L W V ds Ż ) 
 0 n m n  3. Dynamical equations for the Timoshenko
 +
AG[Y V (L)V (L) 
 n n m  beam
 −Xn Un (L)Vm (L)] 
+F (t)Vn (L)Vm (L)Yn The general equation of motion for the Timoshenko
= −F (t) cos Vm (L), (21b) beam under consideration including all the dynamic
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 121

effects in a matrix form can be written as The individual non-zero terms in the equivalent mass
M , damping C , elastic stiffness K and time-varying
M{f¨(t)} + C{f˙(t)} + (K + B(t)){f (t)} = {P (t)}. force-dependent B matrices are as follows:
(23)
L
It can be seen that in Eq. (23) the terms containing [Ki,j ] = − EI Ui Uj ds
0
generalized coordinates in the column-vector {f (t)} L
are X, Y and Z’s, which are dimensionless. Suppose, +
AG Ui Uj ds
we consider ‘N’ number of terms for each of the 0
L
three basic deformation trial functions outlined in Eqs.
− I 2 Ui Uj ds, (26)
(18a)–(18c), then for brevity we can introduce follow- 0
ing notations in lieu of the generalized coordinates
L
{f (t)}: [Mi,j ] = I Ui Uj ds, (27)
Beam cross-section rotations ‘’: 0

f1 (t) = X1 (t), f2 (t) = X2 (t), f3 (t) = X3 (t), L

[Ki,j +N ] = −
AG Ui Vj ds, (28)
f4 (t) = X4 (t), 0
f5 (t) = X5 (t), f6 (t) = X6 (t), ...  L 
fN (t) = XN (t). (24a) [Ki+N,j ] =
AG Vi Uj ds − Vi (L)Uj (L) ,
0
Beam lateral deflections ‘’ : (29)
L
fN +1 (t) = Y1 (t), fN+2 (t) = Y2 (t),
[Ci+N,j +2N ] = −2 A cos  Vi Wj ds, (30)
fN +3 (t) = Y3 (t), fN+4 (t) = Y4 (t), 0
fN +5 (t) = Y5 (t), fN+6 (t) = Y6 (t), ... [Ki+N,j +N ]
f2N (t) = YN (t). (24b)  L 
= −
AG Vi Vj ds − Vi (L)Vj (L)
Beam axial deflections ‘’: 0 (L2 +2rL)  L 
− 0 Vi Vj ds
f2N+1 (t) = Z1 (t), f2N+2 (t) = Z2 (t), 
2
 
 +r 0L sV i Vj ds 
f2N+3 (t) = Z3 (t), f2N+4 (t) = Z4 (t),  
 + 1  L s 2 V V ds 
f2N+5 (t) = Z5 (t), f2N+6 (t) = Z6 (t), ...  i j 
 2 0

f3N (t) = ZN (t). + A2  +r  L Vi V ds , (31)
(24c)  L 0 j 
 
 + 0 sV i Vj ds 
In the above expressions, the subscripts on X, Y and  
Z’s refer to the term sequence number from the ad-  −cos2   L V V ds 
0 i j
missible sets of displacement functions for the beam.
Thus, {f (t)}T ={f1 f2 f3 f4 f5 f6 f7 f8 . . . . . . . . .
[Bi+N,j +N ]
. . . f3N } and (. ) denotes their respective derivatives  L 
with respect to time ‘t’. It is obvious that with the ‘N ’ =− Vi Vj ds − Vi (L)Vj (L) F (t), (32)
number of terms used to represent each of the three 0
displacement functions viz. (, , ), the total number L
of degrees of freedom in the numerical scheme will [Mi+N,j +N ] = A Vi Vj ds, (33)
be ‘3N ’. Thus, terms for the right-hand side force 0
vector {P (t)} turn out to be  L 
  [Ki+2N,j +2N ]=−EA Wi Wj ds−Wi (L)Wj (L)
 {0}N  0
{−F (t) cos Vi (L)}N . (25) L
 L  − A2 Wi Wj ds, (34)
{ A2 0 (s + r)Wi ds + F (t)Wi (L)}N 3N 0
122 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

L p = 2 is the second pulse, p = 3 is the third pulse and

[Mi+2N,j +2N ] = A Wi Wj ds, (35) so on. The general sinusoidal pulse equation shown
0
in Eq. (38) can easily be reduced to a special case of
L
1 pulse per revolution in a rotating system by setting
[Ci+2N,j +N ] = 2 A cos  Wi Vj ds. (36)
0 fp = N/60 = (/2 ). Fig. 4 shows the time-history
of such a special case of half-sine wave pulse with the
The above non-zero terms in the velocity-dependent
pulse frequency ‘fp ’ of 1 pulse per revolution and a
coefficient matrix [Ci,j ] are all due to rotational ve-
load-application time of ‘tc ’. In general, the contact
locity  and represent the gyroscopic effect in the
load-time ‘tc ’ plays an important role in determining
system by being skew-symmetric in nature, which in
the dynamic response of an impacted beam, and it can
the governing equations are shown as CG . The mate-
be seen that in the limit case as tc → 0 , the force-
rial internal damping can be taken into account as a
history becomes perfect impulse function. Some very
function of the non-dimensional factor  of the criti-
important wave propagation type of investigations in
cal damping of the beam material and the spin angular
Timoshenko beam are carried out by sending a single
velocity  . In this situation, the typical terms in the
pulse through the beam and analyzing its dispersion
damping matrix [Ci,j ]D due to the material internal
at different locations [11]. The above multiple-pulse
damping are computed as functions of stiffness matrix
force function shown in Eq. (38) can easily be simpli-
terms [Ki,j ]S containing material parameters Young’s
fied for simulating such single pulse impact by setting
modulus E and shear modulus G as well as shear co-
p = 1, which yields F (t) = 0 for p > 1 , i.e.
efficient
of the beam, which are written as
 
2 t
[Ci,j ]D = [Ki,j ]S . (37) F (t) = −Fmax sin for 0
t
tc
 tc
In general the wave-form of the periodic impact force
and,
F (t) along the axis of the Timoshenko beam can have
many different time-dependent distributions, such as
F (t) = 0 for t  tc . (39)
half-sine wave, rectangular pulse or saw-tooth profile,
etc. Appendix A lists the mathematical representations
In case of large deformation of the beam during the
of commonly used wave-forms to simulate the impact
impact, the radial force along the beam axis should
force distribution in the time-domain with a typical
be corrected for the changing directions of the slope-
pulse frequency of fp -Hz. In the current investigation,
dependent follower forces. This correction may be re-
the following periodic load with half-sine wave distri-
quired in a displacement-controlled external tip load
bution function of frequency fp -Hz, a peak magnitude
also (see Fig. 19), when large radial incursion from the
of (Fmax ) and a load-application time of ‘tc ’ has been
outer case makes it necessary that the right-hand-side
used (see Fig. 4):
column-vector {P (t)} representing the external forces
  
(p − 1) p on the dynamical system be updated according to the
F (t) = 0 for + tc
t
changing slope at the tip  (L, t) as
fp fp
and FLongitudinal = F (t) = Fa cos[ (L, t)]
  
(p − 1) and FTransverse = −Fa sin[ (L, t)].
F (t) = −Fmax sin t−
tc fp
   The current formulation is also directly applicable
(p − 1) (p − 1)
for
t
+ tc . (38) for studying the characteristics of coupled non-linear
fp fp
axial and flexural wave propagation generated by
In the above relationship for the force-history of the sending a single pulse through the Timoshenko beam.
radial impact force, ‘p’ represents the sequence num- For discussion purposes, we will consider a Timo-
ber of the pulse starting from time t = 0 and as such shenko beam of rectangular cross-section of chord
p  1. For example, in Eq. (38) p = 1 is the first pulse, length ‘c’ and depth ‘d’, the structural properties of
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 123

Fig. 4. Periodic nature of the axial pulse F (t) on the free-end of the beam with sinusoidal temporal distribution given by Eq. (38) during
rub (where, pulse sequence number p = 1, 2, 3, 4, 5, etc.) and the frequency of pulse fp = 1 pulse per revolution (non-dimensional contact
load time = t¯c ).

Table 1
Nomenclatures of structural and loading parameters used in the example

Minor principal moment of inertia of the cross-section 1 cd 3

I = 12
Cross-sectional area A = cd
Mass of the beam AL
Shear modulus E
G = 2(1+ )
Shear coefficient shape factor for rectangular cross-section:
= 0.833
(for other shapes, see Hutchinson [25]) 
First flexural mode frequency ( = 3.51602) for the Euler–Bernoulli beam 1 =  EI
L2 A
Pulse frequency (Hz) fp !
2 EI
Euler’s buckling load Fcr =
4L2
Spin velocity (rpm) N
Rotational velocity  = 260N
Special case of 1 pulse per revolution N
fp = 60

which are shown in Table 1. At this point, we will Hence, the matrix terms used in Eq. (23) can be
also introduce a set of non-dimensional parameters broken into the following separate matrices.
with a ‘− ’ on the top as follows, which make it con-
venient to discuss the significance of the numerical M{f¨} + [CD + CG ]{f˙} + ([KS + KC ]
results: + B[F (t)]){f } = {P (t)}, (40)
Non-dimensional peak axial force F̄max : F̄max = where, M is the mass matrix, a function of density
Fmax /Fcr , (symmetric), CD the damping matrix due to the ma-
Non-dimensional
" angular velocity of rotation : terial internal damping =(2/)Ks , CG () the gy-
¯ = [ ( R 2 )/E], roscopic matrix (skew-symmetric, causes coupling of
Non-dimensional time t¯: t¯ = t[/(2 )], axial and lateral motions in the beam), KS the struc-
Non-dimensional contact load time t¯c : t¯c = ture stiffness matrix, a function of E, G and I , etc.
tc [/(2 )], (symmetric), KC () the stress stiffness matrix due to
Slenderness ratio d̄ indicating
" the relative depth of centrifugal stress field, a function of  (symmetric),
the beam cross-section: d̄ = I /(AL2 ), B[F (t)] the time-dependent part of circulatory matrix
Non-dimensional lateral deflection ¯ : ¯ = /L, due to impulsive forces along the axis of the beam-
Non-dimensional axial deflection ¯ : ¯ = /L. column (non-symmetric), and {P (t)} is the column
124 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

vector containing external forces on the dynamical a complex dynamic situation, where instability can be
system. initiated due to the excitation of certain mode shapes
In all the above relations, the inertia matrix M is in the rotor caused by infrequent rubbing of blades.
always symmetric, but the coefficient matrix C for These instabilities which may occur at certain speed
the velocity-dependent-forces is a combination of ranges of the rotor can be either a simple or paramet-
symmetric and skew-symmetric parts. As mentioned ric resonance type of instability in the flexible radial
earlier, the skew-symmetric terms are due to the rotating blade.
Coriolis effect as a result of the coupled lateral and Mathematically speaking, a rigorous definition for
axial movements of the beam. The coefficient matrix the stability of a dynamical equation in a multi-degrees
K for the displacement-dependent-forces can have of freedom system is very difficult to establish. One
time-dependent terms due to contact-impact situation such approach is to use the direct method of Liapunov,
in the longitudinal direction making it non-linear. also called the second method of Liapunov, to prove
In general, in a simple linear structure, it will be a the stability or instability of an equilibrium point of a
symmetric matrix. However, time-dependence part of non-linear system of differential equations. However,
the stiffness matrix B[F (t)] exhibits the interesting in a practical sense if the magnitude of the parameter
non-linear characteristics of the equation of motion. describing the motion grows asymptotically to some
On the right-hand-side of the equation, the column unacceptably high level, then it should be considered
vector {P (t)} contains all the terms due to the pres- unstable. For example, if the lateral movement of a
ence of the external forces on the beam. For example, Timoshenko beam under some kind of excitation or,
corresponding to the generalized coordinates {Y }, it due to the application of a dynamic force grows to an
contains the terms related to lateral forces caused by extent that the outer fiber elastic strain in the beam
the Coulomb friction at the tip. Similarly, correspond- would become in excess of 5%, it may be taken as
ing to the generalized coordinates {Z}, it contains the the threshold of failure based upon the material stress
terms related to axial forces F (t) caused by the con- capability considerations.
tact with the stator structure at the blade tip. If the In the time-domain stability determination, one
rotational speed of the rotor  is not constant then uses the direct-time marching-forward-integration
the time-duration between each pulse is also changing technique, which is a preferred approach in a true
which makes (t) as a function of time as well, and non-linear dynamical problem with clearances and
the corresponding impulse force-history with 1 pulse varying coefficients of friction as a function of speed.
per revolution shown in Eq. (38) becomes: In this method, if the lateral motion of the blade
   at certain locations such as at the tip, is increasing
2 (p − 1) monotonically over a period of time, which can be in
F (t) = −Fmax sin t−
tc (t) the range of milliseconds to seconds, then it is consid-
  
2 (p − 1) 2 (p − 1) ered unstable. The instability is being caused because
for
t
+ tc . (41)
(t) (t) the energy being fed into the dynamical system is
not being dissipated fully by the effective damping
In addition, the terms containing dynamic coefficient present in the dynamical system.
of friction  are non-zero only for the duration the In the frequency-domain approach one solves the
contact-pulse force is active. As a result, all tip rubs are homogenous form of the second-order differential
not necessarily unstable. The dynamic stability of the equations, where the goal is to establish the boundaries
rotating beam can be determined either in a frequency- of the stable and unstable response of the dynamical
domain or in a time-domain. The frequency-domain system as certain parameters such as magnitude of
stability criteria can be used only for linear or quasi- the dynamic load or, the frequency of excitation are
linear type of dynamic problem, where it is possible varied. For solving a non-linear vibration problem in
to write the C, K and B matrices explicitly without frequency-domain, first of all one must simplify the
any reference to the displacements such as effect of contribution of nonlinear terms by using techniques
gaps or clearances in the system. Still, the frequency- such as, perturbation method, asymptotic expan-
domain stability conditions provide a good insight in sion etc. Bolotin [38] describes a method of formal
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 125

Fig. 5. Periodic nature of the axial pulse F (t) on the free-end of the beam with cosine function temporal distribution given by Eq. (42)
during rub (where, pulse sequence number p = 1, 2, 3, 4, 5, etc.) and the frequency of pulse fp = 1 pulse per revolution (Contact load
time tc = 2 / s).

expansion into trigonometric series for solving a gen- ing that tc = 2 /. The net result due to the differ-
eral system of nonlinear differential equations for ences in the two forcing functions is that the amount
dynamic stability of a multiple degrees-of-freedom of energy per revolution being fed to the same dy-
system. The common methods of establishing the namical system is different. Thus, such simplifications,
stability criteria in frequency domain are impossi- which are necessary to make the non-linear contact-
ble to implement in the current problem due to two impact problem amenable to solution in the frequency-
main reasons. First, the discontinuous nature of the domain, may not capture the full dynamics by ne-
dynamic axial load F (t), which is active only during glecting the effects of some important parameters.
the blade-tip travel through the rub-zone on the stator. The time-dependence of the force-history described by
Second, the non-conservative nature of Coulomb fric- Eq. (42) is shown in Fig. 5, which can easily be com-
tion at the blade-tip during the rub cannot be included pared with the one shown in Fig. 4. These approxima-
in the homogeneous form of the equation. As a result tions may be valid in some special situations, such as
of these limitations, in order to obtain a frequency if the spin velocity  is very-very high or, the rub-zone
domain solution for the dynamic stability of a vibro- is circumferentially so large that it is fair to assume al-
impact system with contact-type of non-linearity, one most a 360 ◦ rub on the stator. The underlying assump-
must make some gross simplifying approximations tions are that the pulse-load is being repeated so fast
about the temporal distribution of the dynamic pulse that the inertia of the beam does not allow it to recover
load so that it can be treated as a continuous periodic into undeformed state before the next pulse. On intro-
function. One such approximation about F (t) which ducing the simplifying approximation for F (t) shown
has same peak-to-peak amplitude and frequency of in Eq. (42) in the relationship for B[F (t)] matrix of
the axial load as the true rub load described in Eq. Eq. (40) can be broken in two parts called B1 and B2 ,
(38), but it is a continuous cosine function of time such that the first part accounts for the constant force
‘t’ can be expressed as (see Fig. 5, also Eq. (A.2) in of magnitude Fmax and the second part accounts for
Appendix A) the time-varying component of the axial force on the
beam as
Fmax Fmax
F (t) ≈ − + cos(2 fp t). (42)
2 2 B[F (t)] = B1 (Fmax ) − B2 (Fmax ) cos(2 fp t). (43)
It should be noted that Eqs. (38) and (42) have only
two parameters in common that is, the magnitude of Thus, the equation of motion shown in Eq. (40) in a
the axial load Fmax and the pulse frequency fp . They homogeneous form can be written as
have a completely different temporal distribution and
the function shown in Eq. (42) disregards the effect of M{f¨(t)} + C{f˙(t)}
contact-load time ‘tc ’ altogether, by inherently assum- + (K + B1 − B2 cos(2 fp t)){f (t)} = {0}, (44)
126 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

where, bility for a combination of dynamic parameters where

the vibratory modal oscillations of the beam associ-
C = CD + CG () ated with the respective eigenfrequency i would grow
and exponentially with respect to time ‘t’. The real and
imaginary parts of the eigenvalues r and i enable us
K = KS + KC (). to determine the effective logarithmic decrement as
 
The above equation can easily be identified as a r
Mathieu equation for a multi-degrees of freedom sys-  = −2 . (46)
|i |
tem. Sinha [39,40] has extended the Bolotin’s general
method of solving the dynamic stability characteristic The dynamic system will be stable only when the ef-
in a rotordynamic system for a class of problems, fective logarithmic decrement for each mode is either
which can be reduced in the form of a Mathieu zero or positive. In the dynamic stability problems,
equation. The detailed step-by-step derivation of this the usual interest is in identifying the unstable zones,
method is shown in Appendix B for the sake of com- i.e. to determine the boundary excitation frequencies
pleteness. In this approach, we can assume the general of the “Principal Instability Region”.
solution in the form of a product of two functions: The dynamic stability of a rotor blade subjected
the first is an exponential function with characteris- to pulse-loading with any kind of general discontin-
tic exponent  , and the second in a Fourier series uous distribution over the time-axis can be checked
form of periodic sine and cosine functions with time- in a time-domain solution by direct-time integration
period equal to that of integral multiples of the pulse scheme discussed in the following section.
frequency (1/fp ) as

∞
{b0 }  4. Numerical solution for dynamic stability in
{f (t)} = et  + ({aq } sin(2 qf p t) time-domain
2
q=1

In a non-linear dynamic analysis the transient re-
+ {bq } cos(2 qf p t)) , (45) sponse of the system is of great interest. In order to
solve the present Timoshenko beam-column equations
in time-domain, Eq. (23) is written as
where, {b0 }, {aq } and {bq } are the forcing-function de-
pendent Fourier coefficients. By computing the eigen- {f¨(t)} = M−1 ({P (t)} − C{f˙(t)}
values of Eq. (44) (roots of ), the global conditions
− K{f (t)} − B(t){f (t)}). (47)
of the rotating beam stability with a known value of
internal damping  and excitation frequency fp , can The two most common techniques for integrating the
easily be determined for any combination of parame- equations of motion and computing the displacement
ters along with Fmax . The general solution of this set of and velocity responses of a dynamical system with
equations results in eigenvalues  with complex con- prescribed initial conditions are:
√ roots such that one can write  = r ±ii , where
jugate
i = −1; and eigenvalues ’s have dimensions of per (a) Newmark’s -method and
unit time (rad/s). The positive and negative signs of the (b) fourth-order Runge–Kutta method.
real part of the eigenvalue (r ) indicate the unstable
and stable behaviors of the dynamical system, respec- Dekker and Verwer [41] have done an exhaustive re-
tively. For example, the stability boundaries (r = 0) view of many different variations of the Runge–Kutta
separating the regions of stable and unstable responses schemes for solving the non-linear differential equa-
can be illustrated by plotting the angular velocity of tions and have compared their numerical characteris-
spin  on the x-axis and the peak contact force Fmax tics in terms of stability, uniqueness and convergence
on the y-axis. On this plot, the unstable regions corre- of the solution. Based upon their studies, in this pa-
sponding to r > 0 indicate zones of flutter-type insta- per we have selected a highly accurate sixth order
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 127

Runge–Kutta–Fehlberg [42] integration scheme. The be written as

equation of motion outlined in Eqs. (22a)–(22c) eas-
ily account for time-varying external load {P (t)}, be- yj (t) = {y(t)}
cause in the Runge–Kutta method the contact load at ≡ {X1 , Ẋ1 , X2 , Ẋ2 , . . . , Xk , Ẋk , . . . , XN−1 ,
the blade tip is updated continuously even at the mid- ẊN−1 , XN , ẊN , Y1 , Ẏ1 , Y2 , Ẏ2 , . . . , Yk ,
time-points. Any contact forces are determined by the Ẏk , . . . , YN−1 , ẎN−1 , YN , ẎN ,
penalty method, where all the radial deformations are
Z1 , Ż1 , Z2 , Ż2 , . . . , Zk , Żk , . . . , ZN−1 ,
in the blade, and the case acts like a rigid wall. The
external force column vector {P (t)} contains the term ŻN−1 , ZN , ŻN }. (50)
pertaining to components of forces due to the radial
contact load Fa = F (t) from the rigid case such as, The corresponding entire set of first-order differential
beam free-end tip frictional forces Fa cos Vi (L), equations in a generic form are written as
etc. are in the local normal direction (opposite to the
direction of rotation) and the radial load Fa Wi (L), dyj
= j (y1 , y2 , y3 , y4 , . . . , y6N , t). (51)
etc. are in the local axial direction of the beam, re- dt
spectively. In the numerical integration algorithm of
non-linear equations of motion, the radial load on the The initial conditions at time t = 0 to be used
beam-column F (t) and the corresponding coefficient for computing the transient dynamic response de-
of friction  are updated according to the direction scribed by the above set of equations are ob-
of instantaneous velocity of the tip at the contact lo- tained as the displacements and velocities from
cation. In this sixth order scheme for better accu- the steady-state solution of the rotating beam.
racy, the derivatives of the dependent variables on the These displacement and velocity terms associated
right-hand-side of the equation are computed at sev- with all the generalized coordinates are expressed
eral different mid-points for a given time-step size
t. as, y1 (0), y2 (0), y3 (0), y4 (0), y5 (0), y6 (0), y7 (0), . . . ,
The numerical relationships for combining the values y6N (0), etc. In the sixth-order Runge–Kutta–Fehlberg
at the mid-points are much more cumbersome, but integration scheme [40], the actual numerical values
are briefly discussed here in the generic sense to de- of the coefficients relating the mid-point values of j
scribe the implementation of the sixth order integra- ’s are quite inconvenient and lengthy. For the purpose
tion scheme in the current problem. The equations of of present discussion, the computed value of a typ-
motion shown in Eq. (47) are converted in a set of first- ical variable (yj )t+
t at the time-step (t +
t) in a
order coupled differential equations by introducing a finite-difference form symbolically may be written as
new set of variables called {g(t)} such that, {g(t)} =
{f˙(t)}. This enables us to express Eq. (47) as the (yj )t+
t
following two sets of first-order coupled differential = (yj )t + (
t)
equations:  
(j )t , (j )t+(1/4)
t , (j )t+(3/8)
t ,
× .
(j )t+(1/2)
t , (j )t+(12/13)
t , (j )t+
t
{ġ(t)} = M−1 ({P (t)} − C{g(t)} (52)
− K{f (t)} − B(t){f (t)}), (48)
The numerically stable value of
t for the entire set
is established by computing the highest frequency of
{f˙(t)} = {g(t)}. (49) the following homogeneous equation for undamped
vibration of the beam after dropping the velocity-
Thus, for ‘3N ’ variables described in Eq. (24a)–(24c), dependent coefficient matrix C and the force and the
we would have ‘6N’ number of first-order differen- time-dependent part of the stiffness matrix B[F (t)]
tial equations as shown in Eqs. (48) and (49). At this from Eq. (23), which yields
point, we introduce another new set of variables called
{y(t)}, which in the subscripted index notations can M{f¨} + K{f } = {0}. (53)
128 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Table 2
Details of the parameters considered for the sample beam used in the discussion

M Mass of the beam (blade) 10 kg

R Beam tip radius 150 cm
r Beam root radius 50 cm
A Beam cross-sectional area 22.39 cm2
E Elastic Young’s modulus of the beam material 117 GPa
Beam mass material density 4.466 g/cm3
L Span length of the beam 100 cm
 Beam stagger angle atthe tip −60 ◦
d̄ ‘Slenderness ratio’ = I /AL2 0.010
Fcr Critical Euler buckling load 0.064715 MN
 Angular velocity of rotation (rad/s) 300 rad/s
47.75 Hz = 2864.8 rpm
v Beam tip tangential velocity 450 m/s
# $
tc Beam-column tip pulse-loading duration (s) 0.01 2
 Dynamic coefficient of friction between the blade tip and the inner surface of the outer case 0.0, 0.2
Poisson’s ratio 0.3
F̄max Magnitude of the pulse-load along the beam axis (non-dimensional) 0.8,
# 1.0,
$ 2.0, 5.0, 10.0
t Time (s) t¯ 2
Blade tip running clearance 3 mm.

As opposed to the homogeneous form of the original the cantilever beam-column. In this example problem,
equations of motion given in Eq. (23), the above eigen- we consider the first 10 terms of the trial functions
value problem is very simple to solve with standard for each of lateral deflection  , rotation  and the
techniques such as the Householder method, Jacobi membrane-stretching  deformation of the beam, such
method, Matrix-iteration scheme, etc. It yields 3N that N = 10. Thus, the sizes of each of the M, C and K
numbers of imaginary roots ±ii giving the undamped matrices are 30 × 30. All the 30 eigenvalues are com-
frequencies of the simple dynamical system. Using puted in the current scheme, and the corresponding
the highest eigenvalue of the set called |i |highest ) frequencies expressed in non-dimensional parameter
and taking the 1/8 of the time-period of the highest  and Hz are listed in Table 3. Based upon the time-
mode as step size criteria established in Eq. (53),
t for this
problem turns out to be 2.32391 microseconds. It is
1 2 worth noting that in a Timoshenko beam problem the
Time-step size:
t
. (54)
8 |i |highest numerically stable time-step size
t, to be used in the
explicit method, is controlled by the cross-section rota-
The computer program written for this purpose can tion frequency; which is always much higher than the
simulate any general type of force-history function to typical flexural bending mode frequency of the beam.
represent F (t), which includes all six types of pulse It is obvious from Table 3 that for this beam, mag-
profiles outlined in Appendix A. For the discussion nitudewise the typical axial mode frequencies lie in-
purposes, the sample numerical results are generated between the bending mode frequencies on the lower-
for a typical jet engine fan blade with the parame- bound side and the rotational mode frequencies on the
ters shown in Table 2 . The numerical results show- upper-bound side.
ing linear displacement and deformations are non- In the current numerical algorithm, we also compute
dimensionalized with respect to the span length of the (a) internal energy and, (b) kinetic energy of the
the blade L = 100 cm, and the forces are normalized beam at every time-step for two main deformation
with respect to the critical Euler’s buckling load for modes viz. lateral bending and axial movement, which
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 129

Table 3
Vibrational mode frequencies of the rotating cantilever Timoshenko beam in the numerical example (tip tangential velocity = 450 m/s,
slenderness ratio = 0.01)

Mode number Flexural bending mode Axial stretching mode Cross-section rotation mode

Non-dimensional  (Hz) Non-dimensional  (Hz) Non-dimensional  (Hz)

1 8.3177 67.8565 646.8984 5277.4550 5671.3020 46266.9800

2 28.9632 236.2842 785.3968 6407.3360 5700.8720 46508.2100
3 68.6869 560.3537 817.2732 6667.3870 5751.1120 46918.0900
4 127.7992 1042.5970 1099.5530 8970.2530 5818.7410 47469.8000
5 200.3793 1634.7120 1413.7150 11533.2100 5907.8370 48196.6600
6 290.1916 2367.4090 1727.8760 14096.1600 6010.9200 49037.6200
7 393.9380 3213.7810 2042.0340 16659.0900 6135.0360 50050.1600
8 510.3367 4163.3720 2356.1890 19222.0000 6270.1780 51152.6600
9 637.6240 5201.7920 2670.3530 21784.9800 6424.1020 52408.3900
10 775.0315 6322.7750 2984.5050 24347.8600 6593.2950 53788.6800

are as follows: The full tracking of the time-history of the transfer of

the energy between two deformation modes helps us in
(I E)Lateral bending interpreting the complex non-linear dynamic behavior
(EI ) L of the Timoshenko beam.
= ( )2 ds
2 0

N
2
(EI ) L  4.1. Computing the natural frequencies of the
= (Uk Xk ) ds, (55) Timoshenko beam for different conditions
2 0 k=1
The numerical scheme for computing the mass
(I E)Axial stretching M and stiffness K matrices is verified and checked
(EA) L by determining the eigenvalues of a cantilever radial
= ( )2 ds beam mounted on a rigid disk (see Figs. 1 and 2).
2 0

N 2 Before analyzing the transient response under impact,
(EA) L  we have also investigated the steady-state dynamic
= Wk Zk ds, (56)
2 0 response of the radial blade during the rotating and
k=1
non-rotating conditions. As mentioned earlier, a de-
tailed review of the literature has shown that there is
(KE)Lateral motion a scarcity of published data dealing with the forced
( A) L
= (˙)2 ds response of rotating Timoshenko beams and only re-
2 0 sults related to this topic can be found in terms of

N
2 its natural frequencies. Therefore, it is necessary to
( A) L 
= Vk Ẏk ds, (57) systematically compare the trend of results for some
2 0 limiting cases, such as ‘slenderness ratio’ d̄ → 0, of
k=1
Euler–Bernoulli’s beam formulation which has been
(KE)axial motion analyzed in lot more detail than Timoshenko beams.
( A) L ˙ 2 The current results should serve as a benchmark of
= () ds further reference for researchers and provide useful
2 0

N 2 information to design engineers. With this aim of
( A) L  obtaining a limiting case solution, Eqs. (22a) and
= Wk Żk ds. (58) (22b) are easily reduced to a more familiar form
2 0 k=1 of Euler–Bernoulli’s beam-column formulation for
130 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Fig. 6. Increase in the First Flexural Bending mode frequencies of a Rotating Cantilever Timoshenko beam (slenderness ratio d̄ = 0.01,
0.05, 0.1, 0.15, 0.2, 0.3) under increasing rotational speed.

span-length L as frequency parameter  is defined such that

%
∞ 
 L   (EI )
n = 2 rad/s. (60)
( A) Vm Vn dz Ÿm L ( A)
m=1 0
 L  The computed values of  for the static condition
(EI ) 0 Vm Vn dz
∞
  −(EI )V (L)V (L)  matches well with the closed form solution given
+  m n 
 −F (t)  L V V dz  Ym by Harris and Crede [43]. The graphical data illus-
m=1 0 m n trates the fact that  for the lower slenderness ratio,
+F (t)Vm (L)Vn (L) which in the limit case is equivalent to the classi-
L
= q(s, t)Vn (s) ds. (59) cal Euler–Bernoulli’s beam formulation, moves up
0 rapidly with the increasing rotational speed  ¯ , how-
ever, for higher slenderness ratios (short and stubby
The computed non-dimensional first flexural mode beams), the change is not significant. The effect of the
natural frequency parameter  for varying slenderness speed on all the 10 flexural bending mode frequen-
ratio (d̄ = 0.01, 0.05, 0.1, 0.15, 0.2 and 0.3) are plot- cies of the Timoshenko beam with different slender-
ted for increasing rotational speed up to  ¯ = 0.1 in ness ratios (d̄ = 0.005, 0.01, 0.05, and 0.1) are also
Fig. 6. In majority of rotating beam applications, the shown in Tables 4 and 5, respectively for stationary
upper limit of the tip tangential velocity is a material (¯ = 0.0) and rotating condition ( = 300 rad/s, i.e. ¯
stress-capability limiting parameter, which forces the = 0.0877898) of the beam. The tabulated numerical
designer not to let the realistic tip velocity to exceed results of  are listed here up to four decimal places to
beyond 600 m/s. Corresponding to this limiting value demonstrate the accuracy of the eigenvalue extraction.
of the maximum tip velocity, the non-dimensional an- The first flexural mode frequencies of the beam un-
gular velocity ¯ does not exceed 0.09 in most of the der consideration in the sample example (d̄ = 0.01)
real-life applications. Thus,  ¯ = 0.1 is the practical for the stationary and rotating conditions are: 
limit of the highest angular velocity a design engineer = 3.4809 (28.3977 Hz) and 8.3177 (67.8565 Hz), re-
will ever need to apply from any of the numerical val- spectively. The decays in the value of non-dimensional
ues presented in this paper. Here, the non-dimensional frequency parameter  with increasing slenderness
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 131

Table 4
¯ = 0.0) cantilever Timoshenko
Computed non-dimensional frequency parameter  for the first 10 flexural bending modes of a stationary (
beam as a function of the slenderness ratio d̄

Slenderness ratio d̄ → 0.005 0.01 0.05 0.1

Mode number Non-dimensional  Non-dimensional  Non-dimensional  Non-dimensional 

1 3.5160 3.4809 3.4395 3.2308

2 22.0971 21.9210 19.1603 14.5756
3 61.5388 60.8399 46.8837 31.8319
4 120.3240 117.9210 79.6116 48.4153
5 198.2305 192.1182 114.9608 62.6941
6 293.0804 281.9967 151.3109 68.4167
7 408.1056 384.3208 187.5246 81.0305
8 539.8896 500.4664 220.7516 89.5881
9 688.3948 627.4645 238.7207 102.6107
10 853.3561 764.5730 250.6780 110.3643
 # $ 
A
Non-dimensional frequency Factor  = n L EI , fn = 21 L d̄ E Hz.
2

Table 5
¯ = 0.0877898, tip tangential
Computed non-dimensional frequency parameter  for the first 10 flexural bending modes of a rotating (
velocity = 450 m/s) cantilever Timoshenko beam as a function of the slenderness ratio d̄

Slenderness ratio d̄ → 0.005 0.01 0.05 0.1

Mode number Non-dimensional  Non-dimensional  Non-dimensional  Non-dimensional 

1 15.3513 8.3177 3.7538 3.3121

2 43.7037 28.9632 19.5034 14.6773
3 88.2832 68.6869 47.2828 31.9717
4 150.8615 127.7992 80.0909 48.5743
5 225.0000 200.3793 115.5232 62.8019
6 327.9073 290.1916 151.9531 68.5391
7 443.4582 393.9380 188.2302 81.1678
8 576.4866 510.3367 221.3706 89.7942
9 725.7231 637.6240 238.9201 102.7375
10 891.3798 775.0315 251.1273 110.6600

ratio for the flexural and rotational mode frequencies rotating at 300 rad/s. By comparing these two plots,
are shown in Figs. 7 and 8. The graphical represen- we observe that the rotational speed has significant
tation of  versus d̄ illustrates the overall trend in effect in raising the dynamic buckling load for the
both bending and rotational mode frequencies of the beam. For example, the current beam under consider-
Timoshenko beam as we vary d̄. These results are ation can withstand more than 4 times of the critical
identical to the results recently reported by Zhou and Euler Buckling load (Fmax > 4Fcr ) before collapsing
Cheung [24].The dynamic buckling characteristics of in the first flexural mode under steady-state loading.
the rotating radial beam with its inner radius clamped Under the impact condition depending upon the du-
at the top of the disk in terms of its frequency drop ration of the axial load application, it can deform
versus the compressive axial load are shown in Figs. in some higher modes without actually collapsing.
9a and b. For the purpose of comparing the effect of It can be seen in this plot that all the flexural mode
spin angular velocity, Fig. 9a shows the flexural bend- frequencies drop due to increase in the axial compres-
ing mode frequencies for a stationary beam and Fig. sive load. However, the amount of drop is changing
9b shows the corresponding values for the same beam for different modes with the first and second flexural
132 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Fig. 7. Drop in the First five Flexural Bending (——–) and rotational (- - - - -) mode frequencies of a typical stationary cantilever
Timoshenko beam due to increasing slenderness ratio.

¯ = 0.1) cantilever
Fig. 8. Drop in the First five Flexural Bending (——–) and rotational (- - - - -) mode frequencies of a typical rotating (
Timoshenko beam due to increasing slenderness ratio.

mode frequencies dropping significantly more than pulse. An impulsive load excites all the modes in the
the higher order modes. beam simultaneously, but the higher order modes of
the structure die down quickly leaving only the dom-
4.2. Transient response inant mode active, which happens to be the first flex-
ural mode of the beam. The Runge–Kutta integration
The first transient response of the rotating beam has scheme for the time-marching-forward solution has
been analyzed for a single axial pulse without any fric- been carried out for 60 revolutions of the rotor, which
tional force at the tip. In this simulation, we apply one is 1.26 s. The transient response in terms of the lat-
single longitudinal pulse load of Fmax = 5Fcr lasting eral tip deflection (L, t) has been plotted in Fig. 10.
for t¯c = 0.01, i.e. tc = 0.209 ms. The bar is allowed The stable nature of the practically undamped lateral
to vibrate in an undamped mode due to this single response of ±0.1 mm for 1.26 s shows the robustness
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 133

Fig. 9. (a) Drop in the First six Flexural Bending mode frequencies of a typical stationary cantilever Timoshenko beam (slenderness ratio
d̄ = 0.01) under increasing compressive axial load Fa at the tip. (b) Drop in the First six Flexural Bending mode frequencies of a typical
rotating (angular velocity  =300 rad/s) Timoshenko beam (slenderness ratio d̄ =0.01) under increasing compressive axial load Fa at the tip.

and numerical stability of the integration scheme. It this plot that during the initial impact the beam re-
should also be noted that in the absence of tip fric- sponds at 1286 Hz. It is quickly observed that the lon-
tion ( = 0) , the tip lateral deflection of the beam gitudinal mode response frequency is much lower than
is the result of the coupling of axial motion with the the normal mode eigenvalues calculated from the stiff-
lateral motion caused by the Coriolis effect. As ex- ness and mass matrices. This sub-harmonic response is
pected, the undamped dynamic response of the beam due to parametric resonance in the dynamical system.
is at 67.0 Hz, which matches very well with the first- In the long-time response, the analytically computed
flexural mode frequency of the rotating beam com- time-history of the dynamic deflection illustrates the
puted earlier as 67.8565 Hz. The corresponding tip dis- presence of small amount of numerical damping in the
placement in the axial direction is shown in Fig. 11a. integration scheme, which is inevitable and makes the
The beam tip oscillates about the steady-state deflec- transient response to appear somewhat damped. But,
tion of 2 mm with the initial peak-amplitude in the the most important fact is that there is no numerical in-
range of 1.2 mm
(L, t)
2.8 mm and then even- stability associated with the time-step size
t even for
tually settling down to 1.8 mm
(L, t)
2.2 mm. 50–60 revolutions of the rotor. In order to demonstrate
Fig. 11b shows the zoomed-in view of the short-time the numerical stability and the robustness of the inte-
response in the axial direction. As one can see from gration scheme, we have also looked into the strain-
134 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Fig. 10. Analytically predicted transient lateral deflection of the

tip after a single impulse of F̄max = 5.0 (rotational velocity
 = 300 rad/s).

energy history plot of the dynamical system, which is

shown in Fig. 11c. It is obvious from this plot that
there is no spurious build-up of strain-energy in the
system as the integration marches forward along the
time-axis.
The current time-marching-forward scheme cap-
tures all the complex dynamic characteristics of a
non-linear system very well (see Fig. 12). If the dy-
namical system is undamped, depending upon the
excitation frequency and the duration of the axial load
F (t), the transient response exhibits the characteris-
tics of a typical parametric resonance. Here, due to the
energy transfer between the flexural and axial modes,
the autoparametric vibration instability initiates at
0.1 s and grows to ±4.5 mm in 10 ms. In this exam-
ple, the increasing amplitude modulation frequency
showing the spurt of growing large transverse motions
of the tip of the Timoshenko beam is approximately
1400 Hz. This high-frequency modulation in the beam
response is again the result of coupling between the
longitudinal mode and the flexural mode deflections.
As pointed out earlier, the interaction between the
Fig. 11. Analytically predicted transient axial deflection of the tip
axial motion and lateral motion is caused by the after a single impulse of F̄max =5.0 (steady-state tip deflection due
Coriolis effect, which are coupled through the skew- to rotational velocity of 300 rad/s =2 mm): (a) full-time transient
symmetric terms in the coefficient matrix CG to ac- response, (b) zoomed-in response immediately after impact, and
count for velocity-dependent circulatory forces. This (c) transient axial strain energy in the beam immediately after
impact.
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 135

scheme can handle both the interference-induced load

as well as the longitudinal load momentarily going to
zero due to contact separation. Appendix C outlines
the numerical scheme to determine the magnitude of
the transient tip- contact-load during the interference-
induced rub.

5. Results and discussion

In the analytical discussion of the non-linear tran-

sient response of a Timoshenko beam to a repeated
pulse- loading, we concentrate on the effects of fol-
lowing three parameters:
(i) Dynamic coefficient of friction at the contact
surface between the tip and the outer case ‘’. (ii)
Magnitude of the peak pulse loading i.e., ‘Fmax ’. The
axial load at the blade-tip is varied in steps such that
Fig. 12. Autoparametric vibration instability due to energy transfer F̄max =1.0, 2.0, 5.0, and 10.0, etc. Fig. 13 shows the
between lateral and axial modes (rotational velocity  =300 rad/s). typical load-history of the axial load F (t) applied on
the tip of the rotating beam, which clearly illustrates
the characteristics of repeated pulse. (iii) Duration of
term is directly proportional to the rotational angular rub (effect of increasing contact load-time t¯c ), keeping
velocity . Thus in other words, there is no coupled the peak axial load constant at F̄max =5.
dynamic response in transverse direction due to some In this example, we keep the dimensionless load pa-
prescribed axial motion in a stationary Timoshenko rameter constant at ‘F̄max ’=5, which turns out to be
beam, rather it is a typical dynamic characteristics of Fmax = 0.3236 MN. The resulting lateral deflection at
a rotating beam only, where the interaction effect is the tip (L, t) is plotted against the non-dimensional
directly proportional to the magnitude of rotational time, expressed as the number of revolutions for two
angular velocity . In a stationary beam-column, the- values of the coefficient of friction, i.e. ‘’=0.0 (see
oretically in the absence of dynamic buckling, a pure Fig. 14), and 0.2 (see Fig. 15), respectively. A com-
axial impulse will not produce any lateral movement parison of these two figures highlights the effect of
of the beam along the span. non-linearity in the dynamical system and leads to the
In majority of practical applications of periodic tip- following observations:
pulse problem of a cantilever beam, the blade is repeat-
edly deformed due to local interference with the outer (a) Presence of friction at the blade tip contact sur-
case, when the tip clearance is closed. This problem is face creates an amplitude modulated response cy-
relatively more non-linear than the classical periodic cle for every 16 revolutions of the rotor. The time-
pulse loading situation in the sense that it is a vibro- period of the modulation frequency is practically
contact-impact deformation-controlled scenario rather independent of the magnitude of the blade tip ax-
a known periodic-force-controlled dynamic response ial load. In Fig. 14, this modulation characteris-
of the beam. In an interference-induced deformation- tics is completely absent, when the coefficient of
controlled tip-loading scenario, the magnitude of the friction  = 0. However after 48 revolutions, a bi-
external tip load is not known in priori, rather the load furcation takes place and the dynamic behavior
is dependent upon the amount of deformation. Being changes with sudden jump in the amplitude of os-
a transient contact load, the tip load may even become cillations. This happens for all axial loads of mag-
zero, if the dynamic response of the tip makes it to nitude F̄max  5.0. This chart shows that with a
lose contact with the outer case. The current numerical very low or no friction at all, for lower axial loads
136 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

0.800
Axial Force(-Fa ) in M-Newtons

0.600

Magnitude of Pulse F max = 5(F cr )

0.400

0.200

0.000
0.0 10.0 20.0 30.0 40.0 50.0
Non-dimensional Time (Number of Revolutions)

Fig. 13. Time-history of the typical axial load with repeated pulse loading (1 pulse per revolution, rotational velocity  = 300 rad/s) with
F̄max = 5.0.

up to F̄max = 2.0, there is practically no observ- in the dynamical system. In an eigenvalue solu-
able lateral motion of the blade tip, but for large tion, it is characterized by two distinct roots of i
axial loads F̄max  10.0, the response is extremely separated due to the presence of skew-symmetric
noisy and grows rapidly. terms in the coefficient matrix C affecting the
(b) Fig. 15 shows the response with coefficient of fric- velocity-dependent forces in the system.
tion  = 0.2. Here, with the increasing F̄max , the (c) The timing of the bifurcation point appears to be
modulation frequency remains constant for every a function of the load application time t¯c . For a
16 revolutions of the rotor , but the peak trans- given peak axial load F̄max , the contact load time
verse amplitude of the modulated response in- t¯c has the most significant effect in terms of com-
creases almost linearly with increasingF̄max from plex non-linear dynamic behavior. At some thresh-
±0.25 mm (for F̄max =1) to ±2.90 mm (for F̄max = old value of t¯c , the response quickly becomes
10). By harmonic balance, the typical transverse chaotic. The changing tangential tip lateral move-
dynamic response with amplitude modulation for ments (L, t) are shown in Fig. 16. It can be seen
F̄max = 10 is computed as that for the low values of t¯c = 0.01 when the ax-
ial load is very close to a true impulse function,
(L, t) ≈ 2.90 sin(10t) sin(320t), the dynamic response shows the characteristics of
bifurcation. But, as the t¯c increases, at first the
i.e. magnitude of lateral tip displacement goes down
(L, t) ≈ (2.90/2)[cos(310t) − cos(330t)]. from ±1.0 to ±0.5 mm. Then, it starts increasing
with the motion becoming completely chaotic for
(61)
t¯c =0.2. However, at t¯c =0.5, it becomes again pe-
Thus, the transient dynamic response in the lateral riodic with an amplitude modulated response fre-
direction (L, t) with friction at the tip is a com- quency of 10 revolutions for each modulation cy-
bination of two frequencies close to each other at cle. Determination of threshold t¯c , keeping F̄max
310 and 330 Hz. This particular amplitude modu- constant, where the periodic motion becomes fully
lated response shows typical non-linear character- chaotic is an interesting research topic for further
istics caused by the presence of gyroscopic forces investigation.
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 137

Fig. 14. Transient response of the tip lateral deflection for repeated impulse (1 pulse per revolution, rotational velocity  = 300 rad/s) with
increasing F̄max = 1.0, 2.0, 5.0, 10.0 (coefficient of friction  = 0): (a) F̄max = 1.0, (b) F̄max = 2.0, (c) F̄max = 5.0, and (d) F̄max = 10.0.

The chaotic response of lateral deflection for varying for t¯c = 0.5. The other fact is also observed that while
contact-load time led us to look into the phase-plane maximum axial deflection (L, t) is peaking at 3 mm
trajectory plot in the axial direction (see Fig. 17). The for all cases, the axial velocity is changing from 20 m/s
corresponding ˙ (L, t)) versus (L, t) plots are shown on one extreme (t¯c =0.01) to 0.7 m/s (t¯c =0.5) to other
for t¯c = 0.01, 0.1, 0.3, and 0.5 which illustrates the extreme. It is counter-intuitive that for bigger impulse
typical pattern. One can see the changing of two basins the peak axial velocity of the beam is actually lower
of attraction for t¯c = 0.01 to six basins of attraction than that for the lower impulse.
138 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Fig. 15. Transient response of the tip lateral deflection for repeated impulse (1 pulse per revolution, rotational velocity  = 300 rad/s) with
increasing F̄max = 1.0, 2.0, 5.0, 10.0 (coefficient of friction  = 0.2): (a) F̄max = 1.0, (b) F̄max = 2.0, (c) F̄max = 5.0, and (d) F̄max = 10.0.

6. Concluding remark efficients exists in the differential equations of vibro-

impact systems, the transient response is more suitable
The paper presents an analytical method to illustrate for the exhibition of complicated dynamic behaviors
the non-linear dynamic effect of repeated tip impact than to do the traditional frequency domain eigenvalue
of a rotating Timoshenko beam, which simulates the analysis for the differential equations of the motion.
intermittent rub of radial blades against the rigid outer The biggest drawback in any time-marching-forward
case commonly occurring in a rotating machinery. Be- explicit-integration-scheme is the need to have a very
cause the discontinuity of parameters affecting the co- small time-step size
t to keep the solution process
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 139

Fig. 16. Transient response of the tip lateral deflection for repeated impulse (1 pulse per revolution, rotational velocity  = 300 rad/s) with
increasing contact load-time t¯c = 0.01, 0.1, 0.3,0.5 (coefficient of friction  = 0, F̄max = 5.0): (a) t¯c = 0.01 (circumf. 3.6 ◦ ), (b) t¯c = 0.10
(circumf. 36 ◦ ), (c) t¯c = 0.30 (circumf. 108 ◦ ), and (d) t¯c = 0.50 (circumf. 180 ◦ ).

numerically stable, which makes the computation very been successful in using a reasonable size time-step,
expensive and extremely slow. In an impact and large- so that we can capture all the non-linear character-
deformation problem, the difficulty of numerical in- istics such as bifurcation and chaotic behavior of a
stability is further complicated because the time-step real-life multi-degrees-of-freedom dynamical system
size has to be adjusted as the event progresses. As the for 60 revolutions of the rotor. It is realized that the
structure deforms more severely under impact loads, numerically integrated transient results of any theoret-
the numerically stable time-step size
t keeps on get- ical investigation, even though it is stable, can be open
ting smaller and smaller. In the present paper, with the for discussion until it can be verified with some other
sixth order Runge–Kutta–Fehlberg method we have independent means. The only other way we can ensure
140 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Fig. 17. Phase plane trajectory for chaotic motion of the tip (s = L) in the axial direction subjected to repeated impulse (1 pulse
per revolution, rotational velocity  = 300 rad/s) with increasing contact load-time t¯c = 0.01, 0.1, 0.3,0.5 (coefficient of friction  = 0,
F̄max = 5.0): (a) t¯c = 0.01 (circumf. 3.6 ◦ ), (b) t¯c = 0.10 (circumf. 36 ◦ ), (c) t¯c = 0.30 (circumf. 108 ◦ ), and (d) t¯c = 0.50 (circumf. 180 ◦ ) .

the accuracy of any numerical scheme is to solve some is to impact a rotating blade with a known amount
simplified problem, for which case the closed-form of tip radial incursion (displacement-controlled exter-
solution is either known or, it can be established by nal load) applied by a moving sector of outer case,
comparing the computed results with some test data. and measure its dynamic response using high-speed
Unfortunately, no test data is currently available in camera and dynamic strain gages on the blade. The
public domain with which we can compare our analyt- measured data from their test can be compared with
ical results. Recently, it has been reported that another the analytical predictions using the numerical scheme
research group led by Padova et al. [44] is pursuing a presented in this paper (see Appendix C). The current
separate test program, in which they will actually carry work illustrates that there is a need to do more rigor-
out the impact-rub test in an underground pit. The plan ous research in order to fully understand the complex
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 141

physics of this vibro-impact problem. Specifically, the ways tc . In the case of a single pulse, F (t) is set equal
following observations require further investigations: to zero for all p > 1. The followings are the mathe-
matical relationships for generating a typical periodic
(a) The existence of sub-harmonic resonance re- pulse of magnitude Fmax with an excitation frequency
sponse may be substantiated by solving the prob- of fp -Hz. In the special case of one pulse per revolu-
lem in frequency domain and determining the tion of the rotor spinning at an angular velocity of ,
stability boundaries of the “Principal stability we have, fp = /2 .
zones”. (a) Sine-function pulse (based upon the excitation
(b) The current speculation that there appears to be frequency fp -Hz) (see Fig. 4):
a bifurcation point; which is a function of some   
critical value of the contact load timet¯c needs to (p − 1) p
F (t) = 0 for + tc
t

be analyzed further. At this point it is not known, fp fp
whether it is a Hopf bifurcation, period doubling and
bifurcation or it is a saddle point. Recently, Sinou   
et al. [45] have carried out a Hopf bifurcation sta- (p − 1)
F (t) = −Fmax sin t−
bility analysis of a multiple degrees-of-freedom tc fp
  
disk-type rotor-stator system with Coulomb fric- (p − 1) (p − 1)
tion at the contact surface. for
t
+ tc . (A.1)
fp fp
(c) Another curious observation that the peak axial
velocity is smaller for a larger impulse and larger (b) Cosine-function pulse (based upon the excitation
for small impulse, raises obvious questions about frequency fp -Hz) (see Fig. 5):
the reasons for this interesting phenomena. Fur-   
(p − 1) p
ther work is needed to makes sure that it is not F (t) = 0 for + tc
t

a numerical artifact, and if so what is the physics fp fp
behind this. and
  
The current analysis is limited to a constant rotational Fmax Fmax 2 (p − 1)
F (t) = − + cos t−
2 2 tc f
 p
speed, but it can be extended to a non-constant angular
 
velocity using the approach of Kammer and Schlack (p − 1) (p − 1)
for
t
+ tc . (A.2)
[46] used for solving the Euler–Bernoulli’s beam prob- fp fp
lem. Furthermore, in order to make the results more
realistic and useful in design applications, effect of (c) Symmetric triangular pulse (based upon the ex-
damping on the transient dynamic response should citation frequency fp -Hz):
also be investigated. Analytically predicted magnitude   
(p − 1) p
of undamped response may make numerical results to F (t) = 0 for + tc
t
,
appear much worse and severe than what actual test fp fp
might show.   
2 (p − 1)
F (t) = −Fmax t−
tc fp
  
(p − 1) (p − 1) tc
Appendix A. Mathematical relationships for gen- for
t
+
fp fp 2
erating different forcing functions F (t) to repre-
sent periodic pulse at a rotating contact surface and
  
2 (p − 1)
These relationships for F (t) represent many differ- F (t) = −Fmax tc − t −
tc f
ent pulse-profiles, which can be used in the Transient   p 
Dynamic Analysis of an impacted rotating beam. With (p − 1) tc (p − 1)
for +
t
+ tc .
‘p’ as the pulse sequence number (i.e. p = 1, 2, 3, fp 2 fp
etc.), the load application time for each pulse is al- (A.3)
142 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Fig. 18. Periodic nature of the axial pulse F (t) on the free-end of the beam with rectangular temporal distribution given by Eq. (A.6)
during rub (where, pulse sequence number p = 1, 2, 3, 4, 5, etc.) and the frequency of pulse fp = 1 pulse per revolution.

(d) Saw-tooth pulse (based upon the excitation fre- and

quency fp -Hz): F (t) = −Fmax
  
   (p − 1) (p − 1)
(p − 1) p for
t
+ tc . (A.6)
F (t) = 0 for + tc
t
fp fp
fp fp
The time-history plot of F (t) for a special case of
and fp = /2 has been shown in Fig. (18). This pulse
   profile can be expanded in the form of a cosine series
1 (p − 1) as
F (t) = −Fmax t−
tc fp ∞
   
(p − 1) (p − 1) F (t) = c0 + ck cos(k t), (A.7)
for
t
+ tc . (A.4)
fp fp k=1

where the Fourier coefficients ck and c0 are

(e) Pure Impulse function with varying magnitude
of each pulse ((Fmax )p being the magnitude of pulse  2 /
F (t) cos(k t) dt
sequence number ‘p’): ck = 0 2 /
cos2 (k t) dt
   tc 0
 2 (p − 1) Fmax cos(k t) dt
F (t) = − (Fmax )p  t − = 0 2 /
p
 cos2 (k t) dt
  0
2 sin(k tc )
for 0
t
. (A.5) = Fmax (A.8)
 k
and
(f) Rectangular pulse with step function (based upon  
the excitation frequency fp -Hz) (see Fig. 18): tc
c0 = − Fmax . (A.9)
2
  
(p − 1) p
F (t) = 0 for + tc
t
Thus, in a rectangular shape pulse-profile of
fp fp magnitude Fmax and contact load-time tc with one
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 143

pulse per revolution (see Fig. 18), we can write shown in Eq. (B.1) result in

 

tc {b0 }
F (t) = − (Fmax ) {f˙(t)} = et 
2 2
∞
 
 sin(k tc ) ∞ 
+ cos(k t) . (A.10)  [{aq }−q p {bq }] sin(q p t) 
k +
k=1 +[{bq }+q p {aq }] cos(q p t)
q=1
The above series representation of F (t) enables us (B.2)
to solve the dynamic stability problem with periodic and
pulse impact in frequency domain using the procedure 
outlined in Appendix B. It should be noted that each {b0 }
term of the series (k = 1, 2, 3, 4, . . ., etc.) will yield a {f¨(t)} = et 2
2
different set of solution for the stability zones; and the  [(2 − (q  )2 ){a } 
total solution will be in the form of a sum of individual p q
∞
sets obtained for each value of ‘k’.  −2q p {bq }] sin(q p t) 
+   .
+[(2 − (q p )2 ){bq }
q=1
+2q p {aq }] cos(q p t)
Appendix B. Dynamic stability in frequency (B.3)
domain due to a periodic pulse with its profile We substitute {f (t)}, {f˙(t)}, and {f¨(t)} in the fol-
as a continuous cosine function lowing equation (see earlier Eq. (44)):

For the periodic pulse with a typical frequency of M{f¨(t)} + C{f˙(t)}

fp Hz in the form of a continuous cosine function + (K + B1 − B2 cos(2 fp t)){f (t)} = {0}. (B.4)
distribution, we can use the Bolotin’s [38] method of
For the sake of convenience we will introduce K = K +
harmonic balance to establish the stability conditions
of a dynamic system. At this point, for the sake of
B1 and we will also use previously defined p =2 fp .
Thus, Eq. (B.4) simplifies as
convenience we will introduce a new term called ‘p ’
for the pulse-frequency such that p = 2 fp (rad/s). M{f¨(t)} + C{f˙(t)}
In this scheme, we assume the general solution in the + (K − B2 cos(p t)){f (t)} = {0}. (B.5)
form of a product of two functions: the first being an
exponential function with characteristic exponent , Using the method of harmonic balance we collect the
and the second in a Fourier series form of periodic coefficients of et , et sin(q p t) and et cos(q p t),
sine and cosine terms with time-period equal to that respectively and equate them to zero. This process
of integral multiples of the pulse duration (2 /p ) as yields the following set of homogeneous algebraic
equations:

∞
{b0 }  {b0 } 1
{f (t)} = et  + ({aq } sin(q p t) [2 M + C + K ] − [B2 ]{b1 } = {0}, (B.6)
2 2 2
q=1
 2 M{aq } + [C{aq } − 2(q p )M{bq }]
+ {bq } cos(q p t)) , (B.1) + (K − (q p )2 M){aq } − (q p )C{bq }
1
− B2 ({aq − 1} + {aq + 1}) = {0}, (B.7)
2
where, {b0 }, {aq } and {bq } are the forcing-function
dependent Fourier coefficients. For solving the homo- 2 M{bq } + [C{bq } + 2(q p )M{aq }]
geneous differential equation in a dynamic stability + (K − (q p )2 M){bq } + (q p )C{aq }
problem these Fourier coefficients are considered arbi- 1
trary. Here, the time- derivatives of the function {f (t)} − B2 ({bq − 1} + {bq + 1}) = {0}. (B.8)
2
144 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149


In the above equations, q =1, 2, 3, 4, 5, . . . and {a0 }= M 0 0 0 0
{0}. Thus, the total number of equations represented 0 M 0 0 0
 
by the sets in Eqs. (B.6)–(B.8) is 3N (2q + 1). These R =  0 0 M 0 0 . (B.12)
 
systems of algebraic equations can be rearranged with 0 0 0 M 0
three different matrices P, Q and R as the coefficients 0 0 0 0 M
of ascending powers of  in the following form: Subsequently, the substitution of {Y } = {X} in
[P + Q + 2 R]{X} = {0}. (B.9) Eq. (B.9) yields the following set of homogeneous
algebraic equations:
Here, the column vector {X} contains the unknown
Fourier coefficients such as {b0 }, {a1 }, {b1 }, . . . , etc. It P{X} + Q{Y} + R{Y } = {0}. (B.13)
can be seen that the sizes of the column vector {X} and Using I as a standard unit or identity matrix, Eq. (B.13)
the coefficient matrices P, Q and R are in fact infinite converts into the following set of coupled equations in
in nature, which may be truncated depending upon the terms of {X} and {Y }, which in the “state form” can
number of Fourier expansion terms one is interested be written as
in solving. In the present discussion, the numerical  & ' & '
solution has been shown by limiting the maximum 0 I {X} {X}
= . (B.14)
number of terms of Fourier coefficients ‘q’ to 2. This −R−1 P −R−1 Q {Y } {Y }
implies that the solution of Eq. (B.1) would have only The presence of skew-symmetric terms in the P ma-
two periodic components: one with the time period trix due to terms such as p C is responsible for intro-
(2 /p ) and the second with the period (2 /2p ). ducing instability in the system. The solution of the
The maximum number of instability zones which can above equation results in complex conjugate roots of
be determined by Eq. (B.9) in this manner is equal to eigenvalues , such that one can write  = r ± ii .
2q: that is, for the present problem one can predict up The different signs of the real part of the eigenvalue
to first four instability zones. Here, the transpose of (r ) determine the dynamic stability of the elastic sys-
the column-vector {X} can be written as tem under consideration, which are r > 0 (unstable),
{X}T ={{b0 }T3N {a1 }T3N {b1 }T3N {a2 }T3N r = 0 (critical or oscillatory), and r < 0 (asymptot-
ically stable). Once all the matrices have been for-
× {b2 }T3N . . . . . .}. mulated the eigenvalue problem of Eq. (B.14) can be
In this relationship for {X} each set of Fourier coeffi- solved by using the IMSL-EISPACK routine of Smith
cients have 3N number of terms, which correspond to et al. [47]. The stability boundaries of the “Principal
the dimensions of M, C and K’ matrices derived ear- Instability Region” (r = 0) is determined by taking
lier in Eqs. (26)–(36). The exact expressions for the the subset of the P matrix and solving the following
matrices P, Q and R after a somewhat cumbersome boundary frequency equation (roots of p ) which is
algebraic manipulation are obtained as follows: shown in the determinant form for the present dynamic
 K 0 −B 0 0 
2
 0 (K − 2p M) −p C − 21 B2 0 
 1 
P=
 − 2 B2 p C (K − 2p M) 0 − 21 B2 ,
 (B.10)
 0 − 21 B2 0 (K − (2p ) M)
2
−2p C 
0 0 − 2 B2
1
2p C (K − (2p ) M)
2

Q= system as
 
C 0 0 0 0
(K − 2p M + 21 B2 ) − p C
0 C −2p M 0 0  det . =0.
  p C (K − 2p M − 21 B2 )
0 2 p M C 0 0 ,
  (B.15)
0 0 0 C −2(2p )M
0 0 0 2(2p )M C In a multi-degrees of freedom system with large size
(B.11) M, C and K’ matrices, finding the roots of p by
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 145

numerically solving the determinant is extremely diffi- Eqs. (B.17c) and (B.17d), respectively, we can write
cult and is not an easy task. However, it is much more
convenient by using the following transformation: I{G} = p {E}, (B.20a)
In this scheme of matrix transformation some
lengthy algebraic manipulations are required. On in-
I{H } = p {F }. (B.20b)
troducing dummy column vectors called {S} and {T }
each made up of 3N terms, Eq. (B.15) can be ex-
pressed as a set of homogeneous quadratic equations On combining Eqs. (B.20a), (B.20b) and (B.19) we
arranged in terms of ascending powers of p ; which obtain the following equation in the eigenmatrix form:
in the matrix form are written as  
0 0 I 0
     
(K + 21 B2 ) 0 0 −C  −1 0 0 0 I 
+ p  M ( K + 1 B2 ) 0 0 −M−1 C 
0 (K − 21 B2 ) C 0 2
  & ' & ' 0 M−1 (K − 21 B2 ) M−1 C 0
− M 0 {S} {0}  {E}   {E} 
+2p = . (B.16) 
 
 
 

0 −M {T } {0} ×
{F }
= p
{F }
. (B.21)
 {G} 
   {G} 
 
Now, we introduce a new set of dummy column vec- {H } {H }
tors called {E}, {F }, {G} and {H }, respectively, which
are defined as follows: Since, column-vectors {E} and {F } are in fact arbitrary
by definition, one can multiply the coefficient matrix
{E} = {S}, (B.17a) on the left-hand-side in Eq. (B.21) by a constant with-
out affecting the nature of the solution. Thus, for the
{F } = {T }, (B.17b) purpose of eigenvalue extraction, the coefficient ma-
trix in Eq. (B.21) can be further simplified as
{G} = p {S} = p {E}, (B.17c)
 0 0 I 0 
{H } = p {T } = p {F }. (B.17d)  0 0 0 I 
 
2M−1 (2K + B2 ) 0 0 −2M−1 C
On substituting the above set of column vectors in 0 2M−1 (2K − B2 ) 2M−1 C 0
 {E}   {E} 
Eq. (B.16), we obtain 
 
 
 

{F } {F }
 & ' × = (2p ) , (B.22)
(K + 21 B2 ) 0 {E}  {2G} 
   {2G} 
 
{2H } {2H }
0 (K − 21 B2 ) {F }
 & '
0 −C {G}  0 0 2I 0 
+
C 0 {H } 

0 0 0 2I 

 & ' M−1 (2K + B2 ) 0 0 −2M−1 C
M 0 {G} M−1 (2K − B2 ) 2M−1 C
= p (B.18) 0
 {E/2}   {E/2} 
0
0 M {H }    
   
{F /2} {F /2}
or × = (2p ) . (B.23)
 {G/2} 
   {G/2} 
 
 −1 & ' {H /2} {H /2}
M (K + 21 B2 ) 0 {E}
0 M−1 (K − 21 B2 ) {F } The above Eqs. (B.21), (B.22) or (B.23) can be solved
 −1
& '
0 −M C {G} relatively easily using the IMSL-EISPACK routine
+
M−1 C 0 {H } [47], than solving the determinant form of Eq. (B.15).
& '
{G} This routine yields the eigenvalues as the roots of
= p . (B.19) p or, 2(p ) depending upon whether Eqs. (B.21)
{H }
or (B.22) are solved. These eigenvalues are essen-
On using I as a unit matrix in the expressions for tially the pulse frequencies of the stability bound-
dummy column vectors {G} and {H } outlined in aries in the frequency domain solution of the impacted
146 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

Timoshenko beam. The author [40] has previously expressed as

used the similar method of obtaining the boundary fre-
quencies of stable and unstable zones to establish the (tip )bending-computed
L
general conditions of rotordynamic stability in a ro-
=− [1 − cos( )] ds
tating shaft. 0

)N *
L 
=− 1 − cos V k Yk ds
0 k=1
Appendix C. Calculation of beam tip deflection
) N
*
L 
and resulting dynamic contact forces due to =− 1 − cos sin(k s)Yk ds. (C.2)
interference with the outer case 0 k=1

In most of the practical applications of the peri- Similarly, the blade-tip longitudinal deflection
odic pulse problem, the longitudinal pulse at the free- (tip )membrane due to membrane-stretching effect can
end of the rotating radial beam is generated when the be computed as
moving blade-tip runs into interference due to distor-
(tip )membrane-computed
tion in the outer case. This type of loading condition
N

poses an inverse problem with enforced displacements
at the tip as opposed to a relatively more commonly = ((L, t))computed = Wk (L)Zk
formulated dynamical system which is subjected to a k=1
N 
 
known system of external forces. From the practical sin k L
= Zk . (C.3)
design considerations, one must know the peak mag- k
k=1
nitudes of the dynamic forces for the inverse prob-
lem with displacement constraints. However from an- If the specified tip clearance (gap) is , then for all
alytical considerations, it is not a trivial task to deter- tip > , we will set tip = , ˙ (L, t) = 0, i.e.
mine the dynamic forces generated during the time-
dependent enforced displacement conditions at some (tip )bending-computed +(tip )membrane-corrected = (C.4)
discrete locations of the structure. In this appendix, we
are presenting a numerical scheme, which determines or
all the dynamic forces on the rotating system with en-
(tip )membrane-corrected =−(tip )bending-computed (C.5)
forced time-dependent displacements of the blade tip
caused by the interference with the outer case. or,
The interference in the dynamical system takes
place when the running (steady-state) tip clearance {Zk (t)}corrected
‘’ is used up during the radial movement of the tip, (tip )membrane-corrected
= {Zk (t)}computed (C.6)
henceforth denoted as ‘tip ’. During the sudden inter- (tip )membrane-computed
ference, the contact load at the tip surface pushes the
or
blade tip inward towards the center of spin and the
total tip deflection along the beam-axis tip in the ra- {
Zk (t)} = {Zk (t)}computed − {Zk (t)}corrected (C.7)
dial direction of the rotor can be considered as made
up of two parts (see Fig. 19), which are written as and

tip = (tip )bending + (tip )membrane . (C.1) Interference = [(tip )computed − ]. (C.8)

Thus, we will have the terms in the external load vector

The radial movement of the blade tip ‘tip ’ is computed {P (t)} representing the longitudinal component of the
as follows: external forces modified as follows:
Here, the blade-tip longitudinal deflection
(tip )bending due to beam bending effect can be K {
Zk (t)} = {
Fa W (L)}. (C.9)
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 147

Fig. 19. Blade deformation during the interference-induced rub at the tip due to clearance closure with the outer case.

Increase in the contact force at the tip during at the blade root (s = 0), which loads the retention
system at the hub radius, and can be written as
1

t =
Fa = K{
Zk (t)}. (C.10) (Shear Force ‘Q’)at s=0
Wk (L)
= −Fa cos 
Thus, the new contact force at the time-step  R 
+ A2  d + Fa  (0, t)
(t +
t) = Fa +
Fa . (C.11) r
L
In the integration scheme, the terms corresponding to − A¨ ds.
0
Fa in the external load vector {P (t)} in the right-hand-
side of Eq. (47) are updated accordingly, and we have Thus, the axial component (along the direction of êa )
of the root reaction is expressed as
{f¨} = M−1 ({P (t +
t)} − C{f˙}t   
2R −r
2 2
− K{f }t − B(t +
t){f (t)}t ). (C.12) −Fa cos  + A + Fa  (0, t)
2
L 
In the imposed displacement at the tip, the practical de-
− A ¨ ds sin . (C.13)
sign problem is to determine the dynamic shear forces 0
148 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149

References [20] S.M. Lin, Exact solutions for extensible circular curved
Timoshenko beams with non-homogeneous elastic boundary
[1] W. Flügge, Die Ausbreitung von Biegungswellen in Stäben, conditions, Acta Mech. 130 (1998) 67–79.
Z. Angew. Math. Mech. (ZAMM) 22 (1942) 312–318. [21] D.C.D. Oguamanam, G.R. Heppler, Geometric stiffening
[2] S. Timoshenko, D.H. Young, W. Weaver, in: Vibration of Timoshenko beams, ASME J. Appl. Mech. 65 (1998)
Problems in Engineering, fourth ed., Wiley, New York, 1974, 923–929.
pp. 432–434. [22] S.M. Lin, Dynamic analysis of rotating nonuniform
[3] H.E. Lindberg, A.L. Florence, in: Dynamic Pulse Buckling: Timoshenko beams with an elastically restrained root, ASME
Theory and Experiment, Martinus Nijhoff Publishers, J. Appl. Mech. 66 (1999) 742–748.
Dordecht, 1987, pp. 2–135. [23] S.D. Yu, W.L. Cleghorn, Free vibration of a spinning stepped
[4] T.C. Huang, The effect of rotary inertia and shear deformation Timoshenko beam, ASME J. Appl. Mech. 67 (4) (2000) 839
on the frequency and normal mode equations of uniform –841.
beams with simple end conditions, ASME J. Appl. Mech. 28 [24] D. Zhou, Y.K. Cheung, Vibrations of tapered Timoshenko
(1961) 579–584. beams in terms of static Timoshenko beam functions, ASME
[5] P. Hagedorn, L.R. Koval, On the parametric stability of a J. Appl. Mech. 68 (4) (2001) 596–602.
Timoshenko beam subjected to a periodic axial load, Ing.- [25] J.R. Hutchinson, Shear coefficients for Timoshenko beam
Arch. 40 (1971) 211–220. theory, ASME J. Appl. Mech. 68 (1) (2001) 87–92.
[6] R.O. Stafford, V. Giurgiurtiu, Semi-analytic methods for [26] J.C. Simo, L. Vu-Quoc, The role of nonlinear theories in
rotating Timoshenko beams, Int. J. Mech. Sci. 17 (1975) 719 transient dynamic analysis of flexible beams, J. Sound Vib.
–727. 119 (3) (1987) 487–508.
[7] T. Irie, G. Yamada, I. Takahashi, Vibration and stability of a [27] D.J. Segalman, C.R. Dohrmann, A method for calculating the
non-uniform Timoshenko beam subjected to a follower force, dynamics of rotating flexible structures (Part 1: Derivation,
J. Sound Vib. 70 (1980) 503–512. Part 2: Example calculations), ASME J. Vib. Acoust. 118 (3)
[8] H.D. Nelson, A finite rotating shaft element using Timoshenko (1996) 313–322.
beam theory, ASME J. Mech. Design 102 (1980) 793–802. [28] F.K. Choy, J. Padovan, C. Batur, Rub interactions of flexible
[9] B.A.H. Abbas, Dynamic stability of a rotating Timoshenko casing rotor systems, ASME J. Eng. Gas Turbines Power 111
beam with flexible root, J. Sound Vib. 108 (1) (1986) 25–32. (4) (1989) 652–658.
[10] S.K. Sinha, Dynamic stability of a Timoshenko beam [29] J. Jiang, J. Ahrens, H. Ulbrich, E.M. Scheideler, A contact
subjected to oscillating axial force, J. Sound Vib. 131 (3) model of a rotating rubbing blade, Proceedings of the
(1989) 509–514. Fifth International Conference on Rotor Dynamics, IFToMM,
[11] T.N. Farris, J.F. Doyle, Wave propagation in a split Darmstadt, September 1998, pp. 478–489.
Timoshenko beam, J. Sound Vib. 130 (1) (1989) 137–147. [30] J. Ahrens, H. Ulbrich, G. Ahaus, Measurement of contact
[12] P.K. Datta, R. Ganguli, Vibration characteristics of a rotating forces during blade rubbing, Proceedings of the Seventh
blade with localized damage including the effect of shear International Conference on Vibrations in Rotating Machinery,
deformation and rotary inertia, Comput. & Struct. 36 (6) I Mech E, Nottingham, September 2000, pp. 259–268.
(1990) 1129–1133. [31] S.K. Sinha, Dynamic characteristics of a flexible bladed-rotor
[13] P.A.A. Laura, M.J. Maurizi, R.E. Rossi, A survey of studies with Coulomb damping due to tip-rub, J. Sound Vib. 273
dealing with Timoshenko beams, Shock Vib. Dig. 22 (11) (4–5) (2004) 875–919.
(1990) 3–10. [32] F.C. Moon, in: Chaotic Vibrations, Wiley, New York, 1987,
[14] S.Y. Lee, S.M. Lin, Exact vibration solutions for nonuniform pp. 73–102.
Timoshenko beams with attachments, AIAA J. 30 (1992) [33] G.S. Whiston, On the statistics of chaotic vibro-impact, J.
2930–2934. Sound Vib. 116 (3) (1987) 598–603.
[15] S.Y. Lee, S.M. Lin, Bending vibrations of a rotating [34] D.M. Tang, E.H. Dowell, On the threshold force for chaotic
nonuniform Timoshenko beam with an elastically restrained motions for a forced buckled beam, ASME J. Appl. Mech.
root, ASME J. Appl. Mech. 61 (1994) 949–955. 55 (1988) 190–196.
[16] X. Tong, B. Tabarrok, K.Y. Yeh, Vibration analysis of [35] J. Shaw, S.W. Shaw, The onset of chaos in a two-degrees-of-
Timoshenko beams with non-homogeneity and varying cross- freedom impacting system, ASME J. Appl. Mech. 56 (1989)
section, J. Sound Vib. 186 (1995) 821–835. 168–174.
[17] M.W.D. White, G.R. Heppler, Vibration modes and [36] G.L. Wen, J.H. Xie, Period-doubling bifurcation and non-
frequencies of Timoshenko beams with attached rigid bodies, typical route to chaos of a two-degree-of-freedom vibro-
ASME J. Appl. Mech. 62 (1995) 193–195. impact system, ASME J. Appl. Mech. 68 (4) (2001)
[18] M.W.D. White, G.R. Heppler, Vibration of rotating 670–674.
Timoshenko beams, ASME J. Vib. Acoust. 118 (1996) [37] L. Meirovitch, in: Analytical Methods in Vibrations, The
606–613. Macmillan Company, Collier-Macmillan Limited, London,
[19] G.G.G. Lueschen, L.A. Bergaman, Green’s function for 1989, pp. 225–235.
uniform Timoshenko beams, J. Sound Vib. 194 (1) (1996) [38] V.V. Bolotin, in: The Dynamic Stability of Elastic Systems,
93–102. Holden-Day, San Fancisco, 1964, pp. 272–278.
 S.K. Sinha / International Journal of Non-Linear Mechanics 40 (2005) 113 – 149 149

[39] S.K. Sinha, Stability of viscoelastic rotor-disk system under capability to produce controlled blade tip/shroud rubs at
dynamic axial loads, AIAA J. 27 (1989) 1653–1655. engine speed, Proceedings of IGTI: ASME/IGTI Turbo Expo
[40] S.K. Sinha, On general conditions of rotordynamic stability 2004, Vienna, Austria, Paper No. GT2004-53322, June 14–17,
under combined axial forces and torque, ASME J. Appl. 2004.
Mech. 59 (1992) 225–228. [45] J.J. Sinou, F. Thouverez, L. Jezequel, Non-linear Stability
[41] K. Dekker, J.G. Verwer, in: Stability of Runge–Kutta Methods analysis of a complex rotor/stator contact system, J. Sound
for Stiff Nonlinear Differential Equations, North-Holland, Vib. available on-line February 4, 2004 (38pp).
Amsterdam, 1984, pp. 191–280. [46] D.C. Kammer, A.L. Schlack Jr., Dynamic response of a radial
[42] E. Fehlberg, New high order Runge–Kutta formulas with beam with nonconstant angular velocity, ASME J. Vib. Acoust.
stepsize control for systems of first- and second-order Stress Reliab. Design 109 (1987) 138–143.
differential equations, Z. Angew. Math. Mech. (ZAMM) 44 [47] B.T. Smith, J.M. Boyle, J.J. Dongarra, B.S. Garbow, Y.
(1964) 17–29. Ikebe, V.C. Klema, C.B. Moler, in: Matrix Eigensystems
[43] C.M. Harris, C.E. Crede, in: Shock and Vibration Handbook, Routines—EISPACK Guide, second ed., Springer, NewYork,
second ed., McGraw-Hill Book Company, New York, NY, 1976, pp. 2–21.
1976, pp. 7–14.
[44] C. Padova, J. Barton, M.G. Dunn, S. Manwaring, G. Young,
M. Adams Jr., M. Adams, Development of an experimental