Influence of Tensioner Dry

Friction on the Vibration of Belt

Farong Zhu
Robert G. Parker1
e-mail: parker.242@osu.edu
Department of Mechanical Engineering,
Scott Laboratory,
The Ohio State University,
201 W. 19th Avenue,
Columbus, OH 43210-1142
Drives With Belt Bending
Stiffness
A model of dry friction tensioner in a belt-pulley system considering transverse belt
vibration is developed, and the influence of the dry friction on the system dynamics is
examined. The discretized formulation is divided into a linear subsystem including linear
coordinates and a nonlinear subsystem addressing tensioner arm vibration, which re-
duces the dimension of the iteration matrices when employing the harmonic balance
method. The Coulomb damping at the tensioner arm pivot mitigates the tensioner arm
vibration but not necessarily the vibrations of other system components. The extent of the
mitigation varies for different excitation frequency ranges. The critical amplitude of the
dry friction torque beyond which the system operates with a locked arm is determined
analytically. Superharmonic resonances are observed in the responses of the generalized
span coordinates, but their amplitudes are small. The energy dissipation at the tensioner
arm hub is discussed, and the stick-slip phenomena of the arm are reflected in the velocity
reversals near the arm extreme location. Dependence of the span tension fluctuations on
Coulomb torque is explored. 关DOI: 10.1115/1.2775510兴

Introduction culties induced by tensioner arm stick-slip motion. Superharmonic

Since the late 1970s, serpentine belts have been widely used in
the automotive industry to drive vehicle accessories. A tensioning
system, known as a tensioner, plays an important role in serpen-
tine drives by automatically adjusting the belt tension during op-
eration. It consists of an idler pulley 共tensioner pulley兲 at the end
of a rigid arm 共tensioner arm兲 共Fig. 1兲. The arm pivots around a
fixed point and the pulley is pinned at the free end of the arm
关1–6兴. This assembly leads to geometric nonlinearity in the system
model 关1–5兴, and the arm motion causes belt-pulley coupling
关3,7,8兴 by moving the end points of the two adjacent spans.
Barker et al. 关1兴 establish a mathematical model for the entire
serpentine belt drive with a dynamic tensioner. Determination of
the tensioner arm geometric configuration and span tension arising
from belt stretching are addressed in detail. Transient responses to
the engine firing pulsations are discussed. Hwang et al. 关4兴 deter-
mine the equilibria of the system with a tensioner arm, including
geometric nonlinearity. The system dynamic characteristics are
captured from the eigenvalue problem for the equations linearized
about equilibrium. The onset of belt slip at static and dynamic
states is predicted.
The tensioner arm exhibits stick-slip motion during operation
because of dry 共Coulomb兲 friction at its pivot, which is neglected
in the above literature. Dry friction is a controlled design input
that manufacturers use to distinguish their products. It is one of
the few practical ways to introduce system damping to dissipate
vibration caused by crankshaft excitations. Few works focus on
this subject. Kraver et al. 关9兴 develop a complex modal analysis
procedure for a front end accessory drive system. They use
equivalent linear viscous damping to replace the dry friction at the
tensioner arm. Leamy et al. 关6兴 adopt the model in 关1,4兴 but add a
Coulomb damper at the tensioner arm pivot. Runge–Kutta integra-
tion is employed with modifications to surmount numerical diffi-
1
Corresponding author.
Contributed by the Technical Committee on Vibration and Sound of ASME for
publication in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received
September 8, 2006; final manuscript received April 23, 2007; published online
November 12, 2007. Review conducted by Jonathan A. Wickert.
responses are found that are absent from models without the Cou-
lomb damper. Leamy and Perkins 关10兴 utilize the incremental har-
monic balance method to efficiently predict the nonlinear periodic
response. Again, secondary resonances are observed. Cheng and
Zu 关11兴 analyze a belt drive subject to multifrequency excitations
from both driving and driven pulleys. They assume single-term
harmonic response when the arm is slipping and predict the peri-
odic response of the system when the arm either is purely slipping
or sticks once during slipping.
The aforementioned works 关1,4,6,9–11兴 only consider the pul-
ley rotational vibration, which cannot capture the coupling be-
tween pulley rotations and belt transverse motions observed in
practice and in experimental measurements 关3兴. Beikmann et al.
关2,3兴 investigate a prototypical three-pulley system. The belt is
modeled as a continuum string, which shows coupled vibrations
only between the tensioner pulley and its adjacent spans; other
spans remain straight. Adopting this model, Parker 关7兴 presents a
method for calculating the eigensolutions and dynamic response
of the system. Kong and Parker 关8,12兴 consider the belt bending
stiffness in a system model. This beamlike belt model exhibits
new dynamic characteristics of the system and extends the belt
drive model to broader operating conditions. In particular, all span
motions are coupled to pulley rotations.
In this study, dry friction damping at the arm pivot is incorpo-
rated into the three-pulley system with the beamlike belt model
established in 关8兴. The objective is to investigate the effects of dry
friction on the coupled belt-pulley system dynamics. A multiterm
harmonic balance method is used to find the steady-state periodic
response of accessory drives subject to periodic excitation from
the driving pulley. Energy dissipation occurs when the two mating
surfaces at the arm pivot rotate relative to each other, which is
examined by varying the dry friction torque. The dynamic span
tension fluctuations, an important practical consideration, are in-
vestigated for various dry friction torques and other major param-
eters.

Journal of Vibration and Acoustics Copyright © 2008 by ASME FEBRUARY 2008, Vol. 130 / 011002-1

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 rt rt
y 1 = w1 − x1␪t cos ␤1, y 2 = w2 + 共x2 − 1兲␪t cos ␤2, y 3 = w3
l1 l2
共3兲
where x1,2 苸 关0 , 1兴 are the spatial variables along each span, rt is
the length of tensioner arm, and li are span lengths. Note that y i
satisfies trivial boundary conditions, and y 1 and y 2 refer to the
transverse displacements about the span curvatures, which are
formed by the rigid translation of the span equilibrium curvatures
from moving their end points adjacent to the tensioner pulley from
the arm equilibrium position ␪*t to the instantaneous position ␪t.
The differential operators M, K are self-adjoint, and G is skew-
Fig. 1 A prototypical three-pulley serpentine belt system. The
self-adjoint with an appropriate inner product. See 关12兴 for the
tildes on the physical quantities have been dropped for details of these operators and the definition of the inner product. F
simplicity. is a vector composed of dynamic accessory torques. In this work,
zero dynamic accessory torque is assumed; that is, F = 0. The term
H = 兵0, . . . ,0,hc其T 共4兲
in 共2兲 represents the nonlinear dry friction torque applied at the
System Model arm pivot, where hc = Qm sgn共␪˙ t兲 is the dimensionless torque from
共1兲. The dimensionless quantities are
Kong and Parker 关8,12兴 establish a hybrid continuum-discrete
model incorporating belt bending stiffness. Figure 1 depicts this x̃i w̃i l1 + l2 + l3
prototypical serpentine belt drive, consisting of a driving pulley, a xi = , wi = , l= ,
li li 3
driven pulley, a tensioner, and a serpentine belt. The driving pul-
ley 1 and the driven pulley 3 are subject to accessory moments M̃ 1
and M̃ 3. The tensioner arm is spring loaded with rotational stiff-
t = t̃ 冑 P0
␳l2
, ␧2 =
EI
P 0l 2
, ␥=
EA
P0
ness kr at its pivot and is assembled with alignment angles ␤i , i 共5兲
= 1 , 2 relative to the two adjacent spans in the reference state,
which corresponds to a stationary system without bending stiff-
ness 关12兴. The belt spans are modeled as moving Euler–Bernoulli
s=c 冑 ␳
P0
, ks =
kr
P 0r t
, mi =
Ji
␳ r il 2
,

beams with constant speed c. The incorporation of finite belt

bending stiffness induces nontrivial steady-state span deflection, Jt Q̃m M̃ i
in contrast to the string belt models, which result in straight span mt = , Qm = , Mi =
␳ r tl 2 P 0r t P 0r i
equilibria. The nontrivial equilibria lead to the span-pulley cou-
pling, and the steady-state equilibrium curvature determines the where t̃ is time, P0 is the uniform belt tension for the stationary
extent of the coupling. In 关8兴, the governing equations are linear- belt system with zero accessory torques, EI is the bending stiff-
ized about the steady-state configuration. Eigensolutions from the ness, EA is the belt modulus, Ji and Jt denote the rotational iner-
Galerkin discretized equations show characteristics that differ tias of the pulleys and tensioner arm, and ␳ is the belt mass per
from the string belt model 关2,3兴. Unlike string belt models, the unit length.
spans between fixed-center pulleys are coupled with pulley rota- For Galerkin discretization, the extended variable Y is ex-
tion by the nontrivial steady span deflection. panded in a series of basis functions as
In the present work, the hybrid continuum-discrete system p 3
model in 关8,12兴 is adopted, and a dry friction damper is introduced
at the tensioner arm pivot. The dry friction torque is Y= 兺 a 共t兲␺ 共x兲 + 兺 ␪ 共t兲␺
k=1
k k
k=1
k p+k共x兲 + ␪t共t兲␺t共x兲 共6兲

冦 冧
where p = N1 + N2 + N3 and Ni is the number of basis functions for
Q̃m , ␪˙ t ⬎ 0 the ith span. Each ␺i is a global comparison function that de-
h̃c共t兲 = Q̃m sgn共␪˙ t兲 = 共1兲 scribes deflections of the entire system and satisfies all bound-
0, ␪˙ = 0
t ary conditions. For the ith span, the span deflection is the
− Q̃m , ␪˙ t ⬍ 0 superposition of sinusoidal waves. For example, ␺N1+k
= 兵0 , sin k␲x , 0 , 0 , 0 , 0 , 0其T for the second span for k = 1 , . . . , N2.
The magnitude Q̃m of the dry friction torque is controlled through For the discrete pulleys and tensioner arm, the basis functions are,
preload adjustment between rubbing parts on the arm and fixed for example, ␺ p+3 = 兵0 , 0 , 0 , 0 , 0 , 1 , 0其T for pulley 3. Using the in-
ner product in 关8兴, the discretized formulation from 共2兲 is
hub. The velocity of the arm ␪˙ t remains zero for a finite time when
the arm is locked up. 关M兴Z̈ + 关G兴Ż + 关K兴Z + H = 0 共7兲
Similar to 关8兴, the dimensionless governing equations of the
system considering tensioner dry friction is written into the ex- where Z共t兲 = 兵a1共t兲 , . . . , a p共t兲 , ␪1 , ␪2 , ␪3 , ␪t其T are generalized coor-
tended operator form dinates. H is the same as 共4兲.
The excitation of the system is from the engine firing pulsa-
tions, which induces periodic fluctuations in the crankshaft speed
MŸ + GẎ + KY + H = F 共2兲 at the firing frequency ⍀. In this paper, pulley 1 represents the
crankshaft and its vibration is specified based on engine proper-
where Y = 兵y 1 , y 2 , y 3 , ␪1 , ␪2 , ␪3 , ␪t其T, ␪i, i = 1 , . . . , 3, and ␪t denote ties. For specified ␪˙ 1共t兲, all terms involving ␪1 in 共7兲 are moved to
rotational displacement of the pulley and the tensioner arm about the right-hand side as excitation, which results in
the equilibria, respectively. y i is related to the transverse displace-
ment about the equilibrium of each span wi as M0Z̈0 + 共G0 + C0兲Ż0 + K0Z0 + H0 = f

011002-2 / Vol. 130, FEBRUARY 2008 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 冢 冣 冤 冥
] L0
f = − 关M兴i,p+1␪¨ 1 + 关G兴i,p+1␪˙ 1 + 关K兴i,p+1␪1 , i = 1, . . . ,p + 4, q = Lu, L=
共15兲
] L0

i⫽p+1 共8兲 where u = 兵u1 , . . . , u p+2其T. Introducing the 共R + 1兲 ⫻ 共R + 1兲 opera-

tor
Superscript 0 denotes the new matrices and vectors induced by the
A0 = − ⍀2 diag共12,12,32,32, . . . ,R2,R2兲 共16兲
elimination of the 共p + 1兲th row and column from the original
ones. C0 is a damping matrix obtained from modal damping to and a similar operator B0, one has the relations q̇ = LBu and q̈
capture the energy dissipation in the spans and bearings. = LAu, where A and B are block-diagonal matrices constructed
from A0 and B0, similar to forming L from L0 in 共15兲. Further-
more, f0 = L␹, i.e., ␹ consists of Fourier coefficients of f0. Substi-
tution of 共14兲 into 共10兲 generates fa = LKav0, where Ka is a block-
Methods 0
diagonal matrix with M i,p+3 A0 + 共Gi,p+3
0 0
+ Ci,p+3 兲B0 + Ki,p+3
0
I0 as the
We seek steady-state periodic solutions of 共8兲 for the periodic diagonal element for i = 1 , . . . , p + 2. I0 is an 共R + 1兲 ⫻ 共R + 1兲 iden-
driving pulley excitation using the harmonic balance method 关13兴. tity matrix and v0 = 兵vT , . . . , vT其T. Fourier expansion of 共9兲 yields
A special feature exists in 共8兲 that the last equation for ␪t is the
only equation with a nonlinear term, which is similar to the sys- LKlu = L共␹ − Kav0兲, Kl = m̃A + 共g̃ + c̃兲B + k̃ 共17兲
tem analyzed by Leamy and Perkins 关10兴. Adopting the method of
where
incremental harmonic balance 关14兴, they separate the nonlinear

冤 冥
0
equation from the linear ones to obtain periodic solutions. This m011I0 ¯ m1,p+2 I0
approach reduces the dimensions of the iteration matrices. Ac-
m̃ =
共18兲
cording to this spirit, q = 兵a1 , . . . , a p , ␪2 , ␪3其T are collected as linear
unknown variables, and submatrices m0, g0, c0, and k0 are intro- m0p+2,1I0 ¯ m0p+2,p+2I0
duced by eliminating the last row and column of M0, G0, C0, and and g̃, c̃, and k̃ are formed similarly. Equation 共17兲 produces a
K0. The formulation of this subsystem is relationship between the linear subsystem unknown u and the
nonlinear subsystem unknown v
m0q̈ + 共g0 + c0兲q̇ + k0q = f0 − fa 共9兲
u = Kl−1共␹ − Kav0兲 共19兲

冢 冣
] Fourier expansion also leads to
fa = 0
M i,p+3 ␪¨ t + 共Gi,p+3
0
+ 0
Ci,p+3 兲␪˙ t + 0
Ki,p+3 ␪t 共10兲 ␪¨ t = L0A0v ␪˙ t = L0B0v hc = L0d f p+3 = L0␩ 共20兲
] where d 共unknown兲 and ␩ 共known兲 consist of the Fourier coeffi-
0 cients of the periodic nonlinear function hc and the excitation f p+3.
for i = 1 , . . . , p + 2. Elimination of the last element of f yields f . Substitution of 共20兲 into 共11兲 leads to
The last equation of 共8兲 is
E = Kbv + d + Pu − ␩
M 0p+3,p+3␪¨ t + 共G0p+3,p+3 + C0p+3,p+3兲␪˙ t + K0p+3,p+3␪t + hc
Kb = M 0p+3,p+3A0 + 共G0p+3,p+3 + C0p+3,p+3兲B0 + K0p+3,p+3I0 共21兲
p+2

+ 兺 关M
i=1
0
p+3,iäi + 共G0p+3,i + C0p+3,i兲ȧi + K0p+3,iai兴 = f p+3
where E represents the numerical residue and P is derived from
the summation term in 共11兲.
For a given initial guess v, u can be determined from 共19兲. ␪˙ t
共11兲
from 共20兲 and hc from 共1兲 are known as well. The discrete Fourier
where a p+1 = ␪2 and a p+2 = ␪3 are used subsequently for conve- transformation matrix ⌫ evaluated for R harmonics yields the
nience. Fourier coefficients of the nonlinear function hc as d = ⌫hc. The
To apply harmonic balance, the response is assumed to be pe- residue E of 共21兲 is then evaluated for the initial guess of v.
riodic and the unknowns are expanded as the Fourier series trun- Newton-Raphson iteration is employed to determine the unknown
cated to R harmonics. Because the nonlinear friction torque 共1兲 is v such that E ⬇ 0. Similar to the procedure in 关13兴, the Jacobian
an odd function, only odd order harmonics are considered for Newton–Raphson iteration is

ai共t兲 =
R

兺
r=1,3,. . .
共ui,r cos r⍀t + ui,r+1 sin r⍀t兲, i = 1, . . . ,p + 2
J=
⳵E
⳵v
= Kb +
⳵d
⳵v
⳵u
+P ,
⳵v
⳵d
⳵v
= ⌫ diag 冉冏 冏 冊 ⳵hc
⳵␪˙ t ti
L 0B 0

共22兲
共12兲
where ⳵u / ⳵v is determined from 共19兲. The iteration 兩v = vold new

R
− 共J−1E兲兩vold continues until 储vnew − vold储 / 储vnew储 and 储E储 are less

兺
than specified tolerances.
␪t共t兲 = 共vr cos r⍀t + vr+1 sin r⍀t兲 共13兲
r=1,3,. . . The nonlinear function 共1兲 is discontinuous at ␪˙ t = 0. The ten-
sioner arm exhibits stick-slip motions. When the arm is in the
The time domain is then discretized into N intervals as t0 , . . . , tN to sticking phase, the velocity of the arm is seldom identically zero
form an operator L0 such that even though it is small. When sticking is prevalent, the sign of the
velocity, and so that of the dry friction torque, changes rapidly,
a i = L 0u i ␪ t = L 0v 共14兲 which introduces numerical difficulties. When applying numerical
integration, it takes significant time to achieve accurate results and
where ui = 兵ui,1 , ui,2 , . . . , ui,R+1其T, v = 兵v1 , v2 , . . . , vR+1其T, and L0 maintain numerical stability because small time steps are required.
contains cos r⍀t and sin r⍀t evaluated at these discrete times. The When employing the harmonic balance technique, the iteration
time-discretized response vector of the linear subsystem is hardly converges across the sticking phase. Many methods have

Journal of Vibration and Acoustics FEBRUARY 2008, Vol. 130 / 011002-3

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 exist. There is no problem to compute the response to multifre-
quency excitation even if the system is extended to a general
n-pulley system. Most results have been confirmed by numerical
integration, which is much more time consuming than the har-
monic balance method. Because the distinction between the re-
sults yielded by the two methods is negligible, only those from
harmonic balance are shown in the figures for clarity.

Results and Discussion

Fig. 2 The dry friction torque hc„t… „--… and its approximation
hcs„t… „—…
been developed to avoid or solve these difficulties. As recom-
mended in 关15兴, a saturation approximation that removes the dis-
continuity can yield reliable and numerically efficient results. A
saturation function hcs共t兲 that approximates hc共t兲 is introduced as
共Fig. 2兲
Periodic Response. In this section, the response of the three-
pulley belt system to periodic driving pulley motion is presented.
The nominal physical properties are given in Table 1, from which
the dimensionless parameters can be calculated.
The crankshaft speed fluctuation is ␪˙ 1 = A cos ⍀t. In vehicle ap-
plications, the excitation frequency ⍀ is ␣ times the engine oper-
ating speed, where ␣ indicates an engine with 2␣ cylinders; ␣
= 3 is used in the present work. ⍀ = 0 – 12 corresponds to the en-
gine speed varying over 0 – 5880 rpm, which is a range of practi-
cal importance. The amplitude of the crankshaft speed fluctuation
A is typically an estimated percentage ␮ of the engine speed; thus,
A = ␮⍀ / ␣. The value ␮ = 10% is chosen, although this depends
on engine speed in practice. The belt translation speed is c
= r1⍀冑P0 / ␳l2 / ␣, and its dimensionless quantity is obtained by 共5兲
as s = r1⍀ / ␣l. As excitation frequency ⍀ varies, the belt speed,
and consequently, the eigensolutions of the linear system, change.
For small bending stiffness, natural frequencies of the dominantly
span transverse vibration modes decrease quickly with increasing
speed, while those of the dominantly pulley rotational vibration

冦 冧
Qm ␪˙ t ⬎ ␴
Qm␪˙ t
hc ⬇ hcs = 兩␪˙ t兩 艋 ␴
␴ modal damping coefficient to obtain C0 in 共8兲 is ␨ = 5%, and the
− Qm ␪˙ t ⬍ − ␴
where ␴ is a small positive number. The bound of numerical error
from 共23兲 during sticking is O共␴兲 关15兴. Use of 共23兲 improves the
convergence of Newton–Raphson iteration. If divergence still oc-
curs, then the Broyden 共or secant兲 method 关14,16兴 is employed in
place of Newton–Raphson iteration. In addition, 共23兲 alleviates
numerical “chattering” and saves substantial computer time when
numerically integrating. To maintain the accuracy and numerical
efficiency in this work, ␴ varies between 10−5 and 10−4, according
to the convergence of the harmonic balance procedures, and is
fixed at 10−6 for numerical integration. These values of ␴ ensure
satisfaction of the sticking conditions presented in 关15兴. The varia-
tion of ␴ does not affect when the arm sticks or slips, and its
impact on the dynamic response is negligible.
Through the above approach, steady-state period-T responses
are calculated across a wide range of excitation frequency ⍀,
where T = 2␲ / ⍀. Superharmonic resonances are observed if they
modes change slightly. For large bending stiffness, all natural fre-
quencies decrease comparably slowly and nonmonotonically with
the speed because of the strong belt-pulley coupling 关8兴. Unless
共23兲 indicated otherwise, ␧ = 0.01 is used for belt bending stiffness. The
accessory moments M 1 and M 3 are assumed to be zero.
Tensioner dry friction damping is desirable to reduce vibration.
This may not be achievable, however, if the friction torque is so
large that the tensioner arm is essentially locked. The dynamic
belt tension is compensated by the arm’s rotation around its
steady-state equilibrium. A nearly locked tensioner arm prevents
this and may intensify dynamic response and belt tension fluctua-
tions compared to the case of no tensioner dry friction. Two lim-
iting cases of the system are introduced for comparison. The free
arm system is when there is no dry friction applied on the arm
pivot, i.e., Qm = 0. When Qm is so large that the tensioner arm does
not move, a locked arm system results. Both limiting systems are
linear.
The root mean square 共rms兲 responses of the tensioner arm and
pulleys 2 and 3 across the excitation frequency range of practical
importance are shown in Fig. 3 for various dry friction torques. To
quantify the belt vibrations, the belt transverse displacements av-
eraged along each span are evaluated as

Table 1 Physical properties of the example system

Pulley radius r1 0.0889 m Pulley center 共x1 , y 1兲 共0.5525, 0.0556兲 m

Pulley radius r2 0.0452 m Pulley center 共x2 , y 2兲 共0.3477, 0.0572兲 m
Pulley radius r3 0.0270 m Pulley center 共x3 , y 3兲 共0,0兲
Tensioner arm rt 0.0970 m Pulley center 共xt , y t兲 共0.2508, 0.0635兲 m
Rotational inertia J1 0.07248 kg m2 Belt modulus EA 120,000 N
Rotational inertia J2 0.000292 kg m2 Initial tension P0 300 N
Rotational inertia J3 0.000292 kg m2 Belt mass density ␳ m 0.1029 kg/ m
Rotational inertia Jt 0.001165 kg m2 Tensioner stiffness kr 28.25 N m / rad
Span length l1 0.1548 m Alignment angle ␤1 135.79 deg
Span length l2 0.3449 m Alignment angle ␤2 178.74 deg
Span length l3 0.5518 m Tensioner rotation ␪tr 0.1688 rad

011002-4 / Vol. 130, FEBRUARY 2008 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 3 rms response of „a… the tensioner arm, „b… pulley 2, and
„c… pulley 3 for various Coulomb torques for the parameters in
䊊 locked arm
Table 1: ···· Qm = 0, — Qm = 0.1, -- Qm = 1.5, -·- Qm = 5, —
冑冕 1 dominantly span 2 transverse vibration mode at ⍀ = 3.0 replaces
ȳ i共t兲 = y i2共xi,t兲dxi 共24兲
0
The rms responses of ȳ i共t兲 are shown in Fig. 4.
If the modal amplitude of the tensioner arm in the free arm
system is considerable in a specific mode, the nonlinear response
at that modal resonance will be mitigated significantly. For ex-
ample, the tensioner arm motion participates strongly in the pulley
dominated vibration modes at ⍀ = 2.8, 3.3, and 8.7 of the free arm
system, and therefore, the rotational vibrations of the tensioner
arm and pulleys 2 and 3 at these resonances are markedly sup-
pressed by Coulomb damping 共Fig. 3兲. The suppression is even
reflected in the span transverse vibration in Fig. 4. In contrast, the
modal amplitudes of the tensioner arm are small in the dominantly
span transverse modes at ⍀ = 6.3 共span 1兲, and are moderate in
those at ⍀ = 5.7 and 8.0 共span 2兲 and at ⍀ = 2.0, 3.8, 5.5, and 9.4
共span 3兲. Accordingly, the dry friction has less impact on these
resonances.
Continued increase of dry friction torque monotonically de-
creases vibration of the tensioner arm until the arm is fully locked.
This is not always true for other components. With an effective
loss of one degree of freedom, the eigensolutions of the original
system tend to those of the locked arm case for sufficiently large
dry friction torque. The pulley dominant modes shift to new fre-
quencies while the span transverse dominant modes change only
slightly. The nonlinear response of components other than the ten-
sioner arm approach the new natural frequency resonances for
large Coulomb torque. For the system in Table 1, one of the pulley
3 dominant modes of the free arm system shifts from ⍀ = 8.7 to
⍀ = 6.8 in the locked arm system. The responses of pulley 2, pul-
ley 3 共Fig. 3兲 and span 3 共Fig. 4共c兲兲 suddenly increase near the
locked arm resonance at ⍀ = 6.8. For span 1 共Fig. 4共a兲兲, although
the vibration is suppressed for given dry friction torque up to
Qm = 5 near ⍀ = 6.8, for even larger Coulomb torque the vibration
increases because the response tends to the linear locked arm case,
which is higher than that of the free arm case. Furthermore, a
Journal of Vibration and Acoustics
Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use
Fig. 4 rms response of transverse displacement averaged
along each span varies with excitation frequency for the pa-
䊊
rameters in Table 1: ···· Qm = 0, — Qm = 0.1, -- Qm = 1.5, -·- Qm = 5, —
locked arm
the pulley rotational vibration modes at ⍀ = 2.8 and 3.3 of the free
arm system 共Fig. 4共b兲兲. This mode is barely evident even in the
linear locked arm response of span 2, and therefore, the vibration
is small for the nonlinear system.
The system is subject to an odd excitation function 共1兲 at the
tensioner arm hub, which leads to odd order secondary reso-
nances. For the current model, superharmonics are observed in the
periodic response of the span generalized coordinates. Consider
the rms response and waterfall spectra of the displacement ampli-
tude a3 of the sin 3␲x component of span 1 in Fig. 5. For a given
dry friction torque Qm = 0.5, ⍀ = 4.2 共s = 0.35兲 is one-fifth of ⍀15
= 21.0 and ⍀ = 4.5 共s = 0.38兲 is one-third of ⍀13 = 13.5, where ⍀i
are the ith natural frequencies of the free arm system with belt
speed s. The secondary resonances are not large, and when using
共3兲 to yield belt deflections, they only modestly influence the
physical response. No secondary resonance is possible if the arm
is locked with linear response. Higher dry friction torque extends
the frequency range with locked-arm response. For instance, ob-
serving where the curves in Fig. 5共a兲 separate from the locked-
arm case, the arm is locked for ⍀ ⬍ 2.8 for Qm = 0.5 while the
range extends to ⍀ ⬍ 4.3 for Qm = 1.5. As a result, an increase of
dry friction torque extends the range without secondary resonance
to higher frequency. This property holds for systems with higher
bending stiffness. Accordingly, for the current damping ratio ␨
= 5% and even higher values, the dry friction torque does not
introduce harmful secondary resonances to offset its dissipative
benefits. Even when the system is reduced to a purely discrete
model that considers only pulley rotations, there are no pro-
nounced superharmonics in the response.
Impact of Dry Friction Torque. Starting from the linear sys-
tem 共Qm = 0兲, the tensioner arm vibration decreases monotonically
with increasing Qm across the entire frequency range until even-
tually the arm does not move at all 共Fig. 6共a兲兲. The rate of de-
crease depends on which mode is dominantly excited. Dry friction
most impacts those modes having large tensioner arm motion, and
FEBRUARY 2008, Vol. 130 / 011002-5
 Fig. 6 rms response of „a… the tensioner arm, „b… pulley 2, and
„c… pulley 3 with varying Coulomb torque for the parameters in
Table 1: ···· ⍀ = 2.8, — ⍀ = 3.3, -- ⍀ = 5.7, -·- ⍀ = 6.3, —
䊊 ⍀ = 8.0, —
䉭
⍀ = 8.7

some cases, there is an extremum of the curves in Figs. 6共b兲 and

6共c兲 that is outside of this range 共e.g., ␪2 response for ⍀ = 8.0兲.
Energy Dissipation. The dissipated energy ⌬E within a cycle,
which is a direct measure of dry friction effectiveness, is given by

冕
T N

⌬E =
0
Qm sgn共␪˙ t兲␪˙ tdt ⬇ 兺Q
k=1
m sgn关␪˙ t共tk兲兴␪˙ t共tk兲⌬t 共25兲

Figure 7 shows the dissipated energy for various friction torques

Fig. 5 Superharmonics are shown in „a… rms response and „b…
spectra waterfall of the displacement amplitude a3 of the
sin 3␲x component of span 1 for the parameters in Table 1. In
䊊 locked arm;
„a…, ···· Qm = 0, — Qm = 0.1, -- Qm = 0.5, -·- Qm = 1.5, —
„b… Qm = 0.5.
across the practically important frequency range. Values near zero
in Fig. 7 imply a nearly stationary arm. At resonances of modes
with large tensioner arm rotation, large arm velocity causes con-
siderable dissipation. Larger torque undesirably extends the fre-
quency range where the arm gets stuck and no dissipation occurs.
Arm sticking also prevents the tensioner from achieving its main

response in such modes decreases more steeply. For instance, the

arm rotations in the pulley dominant modes at ⍀ = 2.8 and 3.3 are
larger than in the mode at ⍀ = 8.7; thus, the decrease rate is
steeper. For the dominantly span transverse vibration modes near
⍀ = 5.7, 6.3, and 8.0, the arm modal displacement is least for the
mode at ⍀ = 6.3; thus, this mode requires the highest Coulomb
torque to achieve locked arm status.
The response of other system components does not vary mono-
tonically with the friction torque. Moreover, the vibration in-
creases at the frequency range near the resonance of the mode of
the locked arm system involving large pulley amplitudes, such as
the modes at ⍀ = 5.7, 6.3, and 6.8; it is diminished outside these
frequency ranges. Figures 6共b兲 and 6共c兲 show the response of
pulleys 2 and 3 for varying Coulomb torque. The free arm re-
sponse is the starting point of each frequency branch. The flat
portion of each branch begins at a friction torque beyond which
the arm is fully locked. The vibration is suppressed for the fre-
quency branches ⍀ = 2.8, 3.3, 8.0, and 8.7, while it increases for Fig. 7 Energy dissipation varies with excitation frequency for
the branches ⍀ = 5.7 and 6.3. Generally, the nonlinear response the parameters in Table 1. — Qm = 0.1; ···· Qm = 0.5; -- Qm = 3; -·-
falls between the locked and free arm linear responses, but in Qm = 5.

011002-6 / Vol. 130, FEBRUARY 2008 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 9 Time history of ␪˙ t „--… and dry friction torque „—… for the
parameters in Table 1 and ⍀ = 6.3: „a… Qm = 1.0, „b… Qm = 3.5, „c…
Fig. 8 Energy dissipation varies with Coulomb torque for the Qm = 6.1, and „d… Qm = 11
parameters in Table 1. ␯ indicates the number of the velocity
reversals near the arm extreme location in a half cycle for ⍀
= 6.3; ···· ⍀ = 2.8, — ⍀ = 3.3, -- ⍀ = 5.7, -·- ⍀ = 6.3, —
䊊 ⍀ = 8.0. is effectively stuck. The arm velocity is within the interval
关−␴ , ␴兴, and the linear friction torque in 共23兲 applies.
Critical Dry Friction Torque. There exists a critical dry fric-
purpose of dynamic tension compensation as operating conditions tion torque that induces locked arm status according to previous
fluctuate. The energy dissipation is greatest at frequencies where sections. With the arm locked, ␪t = 0, all terms involving the ten-
high velocities occur. sioner arm in 共11兲 vanish and the pivot torque is
The energy dissipation does not vary monotonically with the
p+2
torque for specific frequencies because the dissipation depends on
the product of arm velocity and Coulomb torque, according to Qm = f p+3 − 兺 关M 0
p+3,iäi + 共G0p+3,i + C0p+3,i兲ȧi + K0p+3,iai兴 共26兲
共25兲. Larger torque may induce lower velocity, which may lead to i=1
lower dissipation. For instance, Qm = 3 and 5 dissipate more en- where ai refers to the component response of the locked arm sys-
ergy than Qm = 0.1 and 0.5 for 6.0⬍ ⍀ ⬍ 7.7, where lower veloci- tem from 共9兲 with fa = 0. For the sinusoidally driven locked arm
ties result for the former dry friction cases, but the opposite holds linear system, Qm varies sinusoidally, i.e., Qm = Qm cr
sin共⍀t + ␾兲
outside this range. Figure 8 addresses how the Coulomb torque
where Qm ⬎ 0 is the critical torque. Figure 10 shows how the
cr
affects energy dissipation for given excitation frequencies. With cr
increasing friction torque for a given frequency, the energy dissi- critical Coulomb torque Qm varies with frequency for various belt
pation first increases to an extreme value and then decreases bending stiffness. These are upper bounds that enforce ␪t共t兲 = 0.
gradually until there is no dissipation for a stationary arm. The system can be effectively locked, ␪t共t兲 ⬇ 0, for lower Qm as
cr
The energy dissipation curves in Fig. 8 for each frequency shown in Fig. 8 for specific frequency. For example, Qm = 18.7 for
branch exhibit lobes. The transitions between lobes reflect a ⍀ = 6.3 and ␧ = 0.01, which is larger than the estimated value
change in the number of arm velocity reversals, as indicated by ␯, Qmn = 10.4 in Fig. 8. In the curves near ⌬E ⬇ 0 in Fig. 8, the
near the arm extreme location for each half cycle. Within a half
cycle, for low friction torque, the arm normally moves back and
forth only once because the torques exerted by other system com-
ponents easily overcome the dry friction torque. See ␯ = 0 portion
of ⍀ = 6.3 branch. An example time history of the tensioner arm
and the corresponding dry friction torque for Qm = 1.0 is shown in
Fig. 9共a兲, where the arm vibrates around the equilibrium regularly.
With increasing Coulomb torque, the vibratory torques are insuf-
ficient to surmount the dry friction and the arm tends to stick near
the extreme rotation angle. With increasing Qm, the arm velocity
first reverses once while crossing the sticking phase, which corre-
sponds to the second lobe on the dissipation curve where ␯ = 1.
Figure 9共b兲 illustrates one velocity reversal with Qm = 3.5. After
the sign of the arm velocity changes, a short time later, the arm
fluctuates back to the extreme angle and then continues its normal
trajectory toward the equilibrium. Because the fluctuation is not
large, the angular velocity falls into 关−␴ , ␴兴 and, according to
共23兲, linear dry friction torque is used during calculation.
When further increasing the torque magnitude, the arm velocity
may reverse two or more times at the extreme arm location. Even-
tually, the arm gets fully stuck for large torque. In Fig. 9共c兲 for
Qm = 6.1, on the return toward the equilibrium, the arm fluctuates Fig. 10 Critical Coulomb torque varies with excitation fre-
back to the extreme location twice 共␯ = 2兲 crossing the sticking quency for different belt bending stiffness beyond which the
phase. In Fig. 9共d兲 for Qm = 11, the arm rotational angle varies tensioner arm is entirely locked for the parameters in Table 1:
only slightly around the equilibrium for the entire cycle. The arm ···· ␧ = 0.01, -·- ␧ = 0.05, — ␧ = 0.1, -- ␧ = 0.2

Journal of Vibration and Acoustics FEBRUARY 2008, Vol. 130 / 011002-7

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 energy dissipations are nearly, but not exactly, zero and decreasing
slowly.
Other than the resonant regions, the curves in Fig. 10 show that
the critical Coulomb torque tends to decrease with increasing
bending stiffness. Away from resonance, the torque generated by
the summation term of the right-hand side of 共26兲 is small com-
pared to that from the first term, i.e., the torque f p+3 derived from
the driving pulley rotation dominates the pivot torque. With this
stipulation,
r1
Qm ⬇ f p+3 = ␥ 关sin ␤1 − cos ␤1w*1共1兲兴␪1 共27兲
l1
For the specified system parameters and the driving pulley rota-
tion ␪1, Qm ⬇ f p+3 varies with the equilibrium deflection at the first
span boundary w*1共1兲. From 关12兴, the boundary layers resulting
from small bending stiffness ␧ become less pronounced in the
equilibrium deflections as ␧ grows and the equilibrium deflections
become larger and smoother. For the parameters in Table 1, cos ␤1
and w*1共1兲 are negative. Therefore, f p+3 and so Qm decrease with
increasing bending stiffness. Fig. 11 Maximum dynamic tension in each span varies with
At the resonant regions, the system components exert more excitation frequency for the parameters in Table 1: ···· Qm = 0, —
pronounced torque evaluated by the summation term in 共26兲 and Qm = 0.1, -- Qm = 1.5, -·- Qm = 5
the pulley vibrations dominate the peak values of each bending
stiffness branch. For example, with ␧ = 0.01, the resonant ampli-
tude of pulley 3 induces a peak value at ⍀ = 6.8 in the critical dry
friction torque curve. frequencies of the locked arm system whose modes involve large
pulley vibrations away from the operating frequency range if a
Dynamic Span Tension. Span tension fluctuations are impor- large dry friction torque is used at the tensioner arm pivot.
tant criteria in evaluating the dynamic behavior of the whole sys-
tem. The tensioning system aims to compensate for operating con-
dition changes and dynamic excitations to reduce the dynamic belt Summary and Conclusions
tension amplitudes. The dynamic span tensions Pdi are 关8兴

冉 冊
A mathematical model for belt-pulley systems with a dry fric-

冕
1 tion tensioner considering belt bending stiffness and transverse
r2 r1 rt
Pd1 = ␥ − ␪2 + ␪1 − ␪t sin ␤1 + w1,xw*1,xdx span vibration is established, and the influence of dry friction at
l1 l1 l1 0 the tensioner arm hub is examined. The dry friction torque is

冉 冊
modeled as a signum function with a constant magnitude, the sign

冕
1
r3 r2 rt of which is determined by the arm velocity. The governing equa-
Pd2 = ␥ − ␪3 + ␪2 + ␪t sin ␤2 + w2,xw*2,xdx tions are divided into linear and nonlinear subsystems. The non-
l2 l2 l2 0 linear one addresses the vibration of the tensioner arm subject to

冉 冊
the dry friction torque; the linear one includes the pulley rotations

冕
1
r1 r3
Pd3 = ␥ − ␪1 + ␪3 + w3,xw*3,xdx
l3 l3 0 flected in the periodic fluctuation of the crankshaft speed.
where the quantities with asterisks are from the steady equilibrium
solution. The integrals in 共28兲 are typically small compared to the
dynamic span deflection with the pulley rotations; thus, the
bounding pulley rotational vibrations dominate the tension fluc-
tuations of each span.
The maximum dynamic tension in each span is shown in Fig.
11. On one hand, increasing dry friction torque monotonically
diminishes the tensioner arm rotation until the tensioner is effec-
tively locked. On the other hand, pulleys 2 and 3 vibrate more
near the resonances of the natural modes of the locked arm system
that involve large pulley amplitudes, such as the modes at ⍀
= 5.5, 5.7, 6.3 共dominated by span vibration兲 and at ⍀ = 6.8 共domi-
nated by pulley 3 vibration兲. These competing effects combine to
impact the tensions for the first and second spans when varying
the dry friction torque. Near these resonances, the balance of the
response of the tensioner arm and pulley 2 leads to slight reduc-
tion in the dynamic tension of the first span with increasing dry
friction torque, while the involvement of pulley 3 vibration in-
creases the dynamic tension of the second span. Pulley 3 vibration
independently affects the third span, so large dynamic tension
occurs at those resonances where large Coulomb torque drives the
arm toward the locked arm configuration. Other than the above
resonant regions, higher Coulomb torque produces better tension
reduction.
To avoid large belt tension fluctuation, one may tune the natural
and spatially discretized span motions. The engine firing pulsation
共28兲 is the excitation source driving the entire system, which is re-
The response to periodic excitation is obtained by the harmonic
balance method. Suppression of the tensioner arm vibration in-
creases with the dry friction torque until the arm is essentially
locked. For other system components, this dissipative benefit is
not monotonic. Odd order superharmonics are identified in the
response of the generalized span coordinates, but they have minor
adverse effect on the physical system because of their small am-
plitudes.
With increasing dry friction torque, energy dissipation from the
dry friction first increases until a maximum value is achieved and
then gradually decreases to zero where the arm is locked. Lobes
occur in the dissipation curve because the arm velocity reverses
near the extreme arm location that reveals the arm sticking fea-
ture. More velocity reversals indicate the arm spends more per-
centage of time in the sticking phase of each cycle. The critical
magnitude of the Coulomb torque that makes the arm essentially
locked is determined analytically. The dynamic span tension is
dominated by the vibration of discrete components adjacent to the
span.
Acknowledgment
The authors are grateful to Mark IV Automotive/Dayco Corpo-
ration for its support and to Dr. Lingyuan Kong for help under-
standing his prior work related to the model.

011002-8 / Vol. 130, FEBRUARY 2008 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 References
关1兴 Barker, C. R., Oliver, L. R., and Breig, W. F., 1991, “Dynamic Analysis of Belt
Drive Tension Forces During Rapid Engine Acceleration,” 1991 SAE Interna-
tional Congress and Exposition, Detroit, Feb. 25–Mar. 1.
关2兴 Beikmann, R. S., Perkins, N. C., and Ulsoy, A. G., 1996, “Free Vibration of
Serpentine Belt Drive Systems,” ASME J. Vibr. Acoust., 118, pp. 406–413.
关3兴 Beikmann, R. S., Perkins, N. C., and Ulsoy, A. G., 1996, “Nonlinear Coupled
Vibration Response of Serpentine Belt Drive Systems,” ASME J. Vibr.
Acoust., 118, pp. 567–574.
关4兴 Hwang, S.-J., Perkins, N. C., Ulsoy, A. G., and Meckstroth, R. J., 1994, “Ro-
tational Response and Slip Prediction of Serpentine Belt Drive Systems,”
ASME J. Vibr. Acoust., 116, pp. 71–78.
关5兴 Jha, R. K., and Parker, R. G., 2000, “Spatial Discretization of Axially Moving
Media Vibration Problems,” ASME J. Vibr. Acoust., 122, pp. 290–294.
关6兴 Leamy, M. J., Perkins, N. C., Barber, J. R., and Meckstroth, R. J., 1997,
“Influence of Tensioner Friction on Accessory Drive Dynamics,” 1997 SAE
Noise and Vibration Conference and Exposition, Traverse City, MI, May 20–
22.
关7兴 Parker, R. G., 2004, “Efficient Eigensolution, Dynamic Response, and
Eigensensitivity of Serpentine Belt Drives,” J. Sound Vib., 270, pp. 15–38.
关8兴 Kong, L. and Parker, R. G., 2004, “Coupled Belt-Pulley Vibration in Serpen-
tine Drives With Belt Bending Stiffness,” ASME J. Appl. Mech., 71, pp.
Journal of Vibration and Acoustics
Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 06/11/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use
109–119.
关9兴 Kraver, T. C., Fan, G. W., and Shah, J. J., 1996, “Complex Modal Analysis of
a Flat Belt Pulley System With Belt Damping and Coulomb-Damped Ten-
sioner,” J. Mech. Des., 118, pp. 306–311.
关10兴 Leamy, M. J. and Perkins, N. C., 1998, “Nonlinear Periodic Response of
Engine Accessory Drives With Dry Friction Tensioners,” ASME J. Vibr.
Acoust., 120, pp. 909–916.
关11兴 Cheng, G. and Zu, J. W., 2003, “Nonstick and Stick-Slip Motion of a
Coulomb-Damped Belt Drive System Subjected to Multifrequency Excita-
tions,” ASME J. Appl. Mech., 70, pp. 871–884.
关12兴 Kong, L. and Parker, R. G., 2003, “Equilibrium and Belt-Pulley Vibration
Coupling in Serpentine Belt Drives,” ASME J. Appl. Mech., 70, pp. 739–750.
关13兴 Zhu, F. and Parker, R. G., 2005, “Nonlinear Dynamics of a One-Way Clutch in
Belt-Pulley Systems,” J. Sound Vib., 279, pp. 285–308.
关14兴 Pierre, C., Ferri, A. A., and Dowell, E. H., 1985, “Multi-Harmonic Analysis of
Dry Friction Damped Systems Using an Incremental Harmonic Balance
Method,” ASME J. Appl. Mech., 52, pp. 958–962.
关15兴 Ferri, A. A. and Heck, B. S., 1997, “Analysis of Stick-Slip Motion in Coulomb
Damped System Using Variable Structure System Theory,” 1997 ASME 16th
Biennial Conference on Vibration and Noise, Sacramento, Sept. 14–17.
关16兴 Baker, G. R. and Overman, E. A., 2006, The Art of Scientific Computing,
共unpublished兲.
FEBRUARY 2008, Vol. 130 / 011002-9