You are on page 1of 19

SIAM J.

APPLIED DYNAMICAL SYSTEMS 


c 2009 Society for Industrial and Applied Mathematics
Vol. 8, No. 4, pp. 1694–1711

The Effect of Codimension-Two Bifurcations on the Global Dynamics of a Gear


Model∗
Joanna F. Mason† and Petri T. Piiroinen‡

Abstract. This paper focuses on examining the global dynamics of a simple model of gear rattle. An impact
map is used to describe a pair of meshing gears, where the nonlinearity arises from the backlash
clearance between the gear teeth. Despite the model’s simplicity, rich and complex dynamics that
lead to gear rattle are observed. To gain insight into the underlying dynamics, a combination of
basin-of-attraction computations, explicit solution construction, one- and two-parameter bifurca-
tion diagrams, and manifold computations is employed. It is found that there is a complex interplay
between both smooth and discontinuity-induced bifurcations, and that two codimension-two bifur-
cations act as organizing centers for the global dynamics. Additionally, computation of the Floquet
multipliers along the grazing curve reveals an interval of robust chaos in between the codimension-
two points.

Key words. nonlinear dynamics, impacting systems, stability analysis, bifurcations, grazing bifurcations, nu-
merical methods, simulation

AMS subject classifications. 34A36, 34C60, 36G25, 37D45, 37G15, 37M05, 70K50

DOI. 10.1137/090759641

1. Introduction. Rattle is a common problem in any geared system. The reason for this is
that gear teeth are typically manufactured with a clearance, known as the backlash [5, 21, 36],
so that they do not jam. Consequently, meshing teeth may repeatedly lose and re-establish
contact and thus rattle. This behavior is also known as backlash oscillation. However, in
quiet operation, the meshing gears are expected to be in permanent contact, where there is
no noisy re-engagement; this regime is therefore highly desirable from a design point of view.
Unfortunately, the manufacturing process gives rise to an additional cause for the gear teeth
to lose contact, namely, eccentric mounting of the gears. This may result in a periodically
varying clearance [18, 27], which adds an additional level of complexity.
Gear systems are often analyzed using finite-element methods [33, 38], but here we will
use a dynamical-systems framework. Gears are thus modeled as a system of second-order
differential equations together with a restitution law at impacts and then rewritten as a system
of first-order differential equations (see [9, 18, 27]). Such systems fall into a general class of
dynamical systems often termed as either piecewise-smooth (PWS) or nonsmooth systems.


Received by the editors May 20, 2009; accepted for publication (in revised form) by J. Meiss October 29, 2009;
published electronically December 23, 2009.
http://www.siam.org/journals/siads/8-4/75964.html

MACSI, Department of Mathematics and Statistics, University of Limerick, Castletroy, Limerick, Ireland (joanna.
mason@ul.ie). The work of this author was supported by the Mathematics Applications Consortium for Science and
Industry (MACSI), funded by the Science Foundation Ireland (SFI) Mathematics initiative 06/MI/005.

Department of Applied Mathematics, School of Mathematics, Statistics and Applied Mathematics, National
University of Ireland, Galway, University Road, Galway, Ireland (petri.piiroinen@nuigalway.ie).
1694

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1695

Many diverse applications found in nature, economics, and engineering can be modeled
as PWS systems. Examples include forest fires [10], floor and ceiling models in economics
[17], railway vehicles [23], magnetic bearing systems [22], passive walkers [34], rocking block
dynamics (with application to the devastating effect of earthquakes) [19], offshore oil-towers
in marine engineering [37], and electronic switches in DC/DC buck converters [11], to name
just a few. For many more examples of applications that can be modeled as PWS systems,
see [12].
PWS systems have the potential for both rich and complex dynamics, and changes in the
dynamical behavior cannot always be described through standard (smooth) bifurcation theory
alone (see [25] for an overview of the theory of smooth bifurcations). It is well known that
PWS systems can exhibit all the dynamical behavior and bifurcations observed in smooth
systems. In addition, phenomena exhibited by PWS systems include grazing and sliding
bifurcations that can lead to instantaneous jumps to chaos (without going through saddle-
node bifurcations or period-doubling cascades) [16, 14]. Extensive analysis of these nonsmooth
phenomena has led to a new theory of discontinuity-induced bifurcations (DIBs) [39, 12]. The
classical example of a DIB is the grazing bifurcation in impacting systems, which occurs when
a limit cycle approaches a discontinuity surface tangentially [29]. See [13] for a comprehensive
review of bifurcations in nonsmooth systems.
The gear model we consider here was originally derived in [18, 27] and takes the form of
a low-degree-of-freedom oscillator, with a nonlinearity arising from the backlash between the
gear teeth. Despite the simplicity of this model, complex dynamics with rich and delicate
structure have been observed [28]. The purpose of this paper is twofold. First we will show
how bifurcation analysis, basins of attraction, and manifold computations can be used to gain
a better understanding of the global dynamics of the system. Second, using the introduced
numerical methods, we will show that the main cause of all irregular behavior is due to two
codimension-two bifurcations where standard (smooth) bifurcations and DIBs coalesce.
The remainder of this paper is organized as follows. We begin in section 2 by introducing
the PWS mathematical model of a symmetric pair of meshing spur gears and also the simpler
impacting-contact model of backlash. In section 3 we introduce notation to describe the vector
field, the phase space, the impact map, and periodic orbits. This is followed in section 4 by a
description of the various numerical methods. Basins of attraction for a range of parameter
values are presented in section 5 alongside one-parameter bifurcation diagrams. In section 5.2
we describe the important role that the period-one solutions play in the underlying dynamics.
The existence bounds of these solutions are illustrated in a two-parameter bifurcation dia-
gram, which reveals two codimension-two bifurcations that act as organizing centers for the
underlying dynamics. Finally, in section 6 we provide some concluding remarks and outline
areas for future work.
2. The gear model. Our nondimensionalized equation of motion (EOM) for a symmetric
1:1 pair of meshing spur gears (see [18, 27]) takes the form

(2.1) Φ + δΦ + 2κB(Φ) = 4πδ − 4π 2 ε cos(2πt) − 2πδε sin(2πt),

where Φ, Φ , and Φ , respectively, denote the relative rotational displacement, velocity, and
acceleration of the gears, and δ and κ denote the nondimensionalized damping and stiffness

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


1696 J. F. MASON AND P. T. PIIROINEN

coefficients, respectively. Moreover, ε describes the nondimensional amplitude of an external


forcing effect caused by the eccentric mounting of the gears, which acts periodically at a
frequency equal to the gross rotation rate of the gears.
Here, the driving torque is held constant and is represented by the term 4πδ. The restoring
torque between the gear pair can be modeled as the PWS backlash function

⎨ Φ − β, Φ ≥ β,
(2.2) B(Φ) = 0, −β < Φ < β,

Φ + β, Φ ≤ −β,

where β > 0 and 2β is the nondimensional backlash width. The backlash function B in (2.2)
is a typical example of a PWS continuous function with discontinuities at Φ = ±β, where it is
not differentiable. However, a realistic large stiffness, κ  1, motivates an impacting-contact
model of backlash [18] with a coefficient of restitution equal to 1 instead of a PWS model. As
an alternative to (2.1) and (2.2) we may thus analyze the somewhat simplified EOM

(2.3) Φ + δΦ = 4πδ − 4π 2 ε cos(2πt) − 2πδε sin(2πt), −β < Φ < β,

with perfectly elastic impact events at times timp , where

(2.4) Φ (t+  −
imp ) = −Φ (timp ) when |Φ(timp )| = β.

Here Φ (t−  +
imp ) and Φ (timp ) are the velocities immediately before and after impact, respectively.
Perfect elasticity (corresponding to a restitution coefficient equal to 1) implies the absence
of chattering (essentially, an infinite number of impacts in a finite time; see further [6, 30, 32]),
which significantly simplifies our analysis. This means that the only way in which quiet motion
can be achieved is through energy losses in the free-flight part of the dynamics; see further
section 3.1.1.
In [18, 28] the original equations, (2.1) and (2.2), for which permanent contact is a physical
reality, were compared with (2.3) and (2.4). Trajectories under the impacting-contact model
which repeatedly impact the Φ = β boundary with very low velocities were identified to behave
akin to permanent linear contact solutions. Therefore, for convenience, we shall exclusively
consider the impacting-contact model in this paper, given by (2.3) and (2.4).
3. Dynamics.
3.1. The dynamical system. To analyze the dynamics of the relative gear displacement
Φ, we start by solving (2.3) to obtain

(3.1) Φ(t) = c1 + c2 e−δt + ε cos 2πt + 4πt,

and thus the velocity is

(3.2) Φ (t) = −δc2 e−δt − 2πε sin(2πt) + 4π,

where the constants c1 and c2 are given by the initial conditions. An important feature of the
dynamics is that Φ > 0 for all parameter values under investigation in this paper. Therefore,

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1697

Φ Φ Φ
β β β

t t̂ t t̂ t
t0 t1 t0 t1 t0 t1

−β −β −β
(a) (b) (c)
Figure 1. Sketches of the three different types of trajectory. From left to right, (a) illustrates a trajectory
which leaves Φ = β and returns to Φ = β without interacting with Φ = −β, (b) a trajectory which leaves
Φ = β, impacts Φ = −β with a negative velocity, and then returns to Φ = β, and (c) a trajectory which leaves
Φ = β and then grazes Φ = −β before returning to Φ = β.

the motion dynamics between impacts cannot have local maxima (see further [28]). The
implication of this is that there are only three possible scenarios for trajectories starting at
Φ(t0 ) = β, which are depicted in Figure 1.
If we now let
 T
(3.3) x = (x1 , x2 , τ )T = Φ, Φ , (t + τ0 ) mod 1 and S = R × R × [0, 1),

the two impact boundaries, or discontinuity surfaces, introduced in section 2 can be defined
as
Σ+ = {x ∈ S | x1 = β} and Σ− = {x ∈ S | x1 = −β},

respectively. The two surfaces naturally divide the space S into the three regions

M + = {x ∈ S | x1 > β},
M 0 = {x ∈ S | −β < x1 < β}, and

M = {x ∈ S | x1 < −β};

see Figure 2(a). Since we will consider only elastic impacts, both M + and M − are forbidden
regions for a trajectory to visit. However, as we will see below, these regions are still important
for the understanding of the overall dynamics.
With these new definitions we can rewrite (2.3) and (2.4) as

(3.4) ẋ = f (x), x ∈ S ⊂ R3 ,
(3.5) x+ −
imp = R(ximp ), x− + − +
imp , ximp ∈ Σ ∪ Σ ,

where x− +
imp is the state at impact, ximp is the state after the impact, and

 T
f (x) = x2 , 4πδ − 4π 2 ε cos(2πτ ) − 2πδε sin(2πτ ), 1 ,
R(x) = (x1 , −x2 , x3 )T .

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


1698 J. F. MASON AND P. T. PIIROINEN

1
00
x1 x2 τ
M+
β Σ+ I+

00 M0 1 Γ
τ

−β Σ− I−
M−
(a) (b)
Figure 2. (a) The division of the phase space projected onto the x1 -τ surface by the discontinuity surfaces
Σ+ and Σ− . (b) The division of initial conditions x ∈ Σ+ projected onto the x2 -τ plane that will lead to
different uses of the map Π (see further Figure 3).

3.1.1. Quiet dynamics. The impacting-contact model has the disadvantage that it does
not admit solutions with permanent contact (quiet solutions) due to chattering [6, 30]. There-
fore, we will here make a simple analysis to show that the system can approach a permanent-
contact solution for long simulations due to energy losses in the free-flight dynamics.
If we assume that a trajectory leaves the Φ = β boundary with velocity Φ = −v (where
0 < v << 1) at time t∗ , we can solve (3.1) and (3.2) for the integration constants c1 and c2 .
An expansion of Φ and Φ about t = t∗ gives
 
∗ δv  
(3.6) Φ = β − v(t − t ) + − δπε sin(2πt ) + 2δπ − 2π ε cos(2πt ) (t − t∗ )2 + O (t − t∗ )3
∗ 2 ∗
2
and
 
(3.7) Φ = −v + δv − 2δπε sin(2πt∗ ) + 4δπ − 4π 2 ε cos(2πt∗ ) (t − t∗ )
 2 
δ v  
+ − + δ πε sin(2πt ) − 2δ π + 4π ε sin(2πt ) (t − t∗ )2 + O (t − t∗ )3 .
2 ∗ 2 3 ∗
2
If we then then look for the next impact of Φ = β at time t = t̄ and ignore cubic and higher-
order terms, we can solve Φ(t̄) = β, which is now quadratic in (t̄ − t∗ ). The nontrivial solution
may be expressed as
2v
(3.8) t̄ − t∗ = .
−δv + 2δπε sin(2πt ) − 4δπ + 4π 2 ε cos(2πt∗ )

This can then be substituted into (3.7) to give


f 2
Φ (t̄) = v + v ,
g
where
 
δ2 v 2 ∗ 2 3 ∗
(3.9) f =4 − + δ πε sin(2πt ) − 2δ π + 4π ε sin(2πt ) ,
2
 2
(3.10) g = −δv + 2δπε sin(2πt∗ ) − 4δπ + 4π 2 ε cos(2πt∗ ) .

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1699

Since g will always be positive, we are concerned only with the sign of f . The minimum value
of f occurs when sin(2πt∗ ) = −1, and therefore

(3.11) fmin = 2v 2 (−δ2 (v + 4π) − 2πε(δ2 + 4π 2 )) < 0,

and the maximum value of f occurs when sin(2πt∗ ) = 1, and therefore

(3.12) fmax = 2v 2 (−δ2 (v + 4π) + 2πε(δ2 + 4π 2 )) > 0

provided

δ2 (v + 4π)
(3.13) ε> ,
2π(δ2 + 4π)

which will always be satisfied for the parameter values that we investigate. The implication
of this is that motion with small impact velocities will sometimes gain energy and sometimes
lose energy from one impact to the next. Numerical simulations indicate that the system on
average loses slightly more energy per forcing period than it gains, particularly if the system
stays in free flight for long periods of time and does not approach a nonsilent attractor. This
means that in theory the motion slowly approaches permanent contact. However, in practice,
when we talk about permanent contact, we actually have sustained motion with low-velocity
impacts.
3.2. Impact map. In this section we will take the necessary steps to define an impact
map Π that maps points on Σ+ with x2 < 0 back to Σ+ again with x2 < 0. To analyze the
dynamics of the gear model (see (3.4) and (3.5)) we can exploit the properties that the system
is smooth between impacts and that no local maxima can occur since ẋ2 > 0 for all x ∈ S.
First, if we let x(t) be the solution to (3.4) with initial condition x(0) ∈ Σ+ at t = 0, then
we can define a map ΠT : Σ+ → S as

(3.14) ΠT (x(0)) = x(T − τ0 ), x(0) = (β, −v, τ0 )T ∈ R+ × R− × [0, 1),

where v > 0 and T is a positive constant such that T − τ0 > 0. Since ẋ2 > 0 for all t and
x2 (0) = −v < 0 in (3.14), we know that ΠT (x(0)) will have a unique minimum in M 0 ∪Σ− ∪M −
(see Figure 2(a)) for some T . Therefore, we can divide the physically valid initial conditions
into the sets

I + = x ∈ Σ+ | x2 < 0, Πt̂ (x) = (x̂1 , 0, τ̂ )T ∈ M 0 ,


I − = x ∈ Σ+ | x2 < 0, Πt̂ (x) = (x̂1 , 0, τ̂ )T ∈ M − ,

and

(3.15) Γ = x ∈ Σ+ | x2 < 0, Πt̂ (x) = (−β, 0, τ̂ )T ∈ Σ−

for t̂ > τ0 . This division naturally separates all physically valid initial conditions on Σ+ into
initial conditions for which the trajectory will and will not reach Σ− (see Figure 2(b)). As
we will see, this will form an important part in the understanding of the dynamics. Also, in

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


1700 J. F. MASON AND P. T. PIIROINEN

(a) x2 (b) x2 (c) x2


Π+ Π+
xa1 R xa2 xa3 xa2
R xa1 R
Π0 R x1 x1 x1
xb xa1 xb xb
Π− xa Π− xa
xa
Σ− Σ+ Σ− Σ+ Σ− Σ+

Figure 3. Sketches of the three different forms of the map Π(x). In (a) xb = Π(xa ) corresponds to a
trajectory with initial condition xa ∈ I + that leaves Σ+ and returns to xa1 ∈ Σ+ without interacting with Σ− ,
whereafter it impacts to reach xb ∈ Σ+ . In (b) xb = Π(xa ) corresponds to a trajectory with initial condition
xa ∈ I − which leaves Σ+ , impacts at xa1 ∈ Σ− with a negative velocity to reach xa2 ∈ Σ− , and then returns to
xa3 ∈ Σ+ , whereafter it impacts to reach x3 ∈ Σ+ . In (c) xb = Π(xa ) corresponds to a trajectory with initial
condition xa ∈ Γ which leaves Σ+ and grazes the other boundary at xa1 ∈ Σ− before returning to xa2 ∈ Σ+ ,
whereafter it impacts to reach xb ∈ Σ+ .

what follows we will refer to Γ as the grazing curve as it corresponds to the initial conditions
that will lead to a trajectory with a minimum on Σ− , i.e., one that grazes the surface Σ− .
Next we define three flow maps that, respectively, map points from Σ+ to Σ− , from Σ−
to Σ+ , and from Σ+ back to itself as (cf. Figure 1)
Π− : Σ+ → Σ− , Π+ : Σ− → Σ+ , and Π0 : Σ+ → Σ+ .
We use (3.5) as the map for the restitution law at impact, i.e., R(x) = (x1 , −x2 , τ )T . Now we
have everything we need to define the impact map as

R ◦ Π0 (x), x ∈ I +,
(3.16) Π(x) =
R ◦ Π+ ◦ R ◦ Π− (x), x ∈ I − ∪ Γ.
The map Π will form the basis of the analysis of the gear-system dynamics. It is well known
that impact maps corresponding to dynamics that experience grazing impacts can locally be
described by square-root maps, and thus we have here that Π ∈ C 0 (see [29]). In Figure 3 the
three possible scenarios for the map Π are depicted. Depending on the initial conditions, a
trajectory can (i) leave Σ+ and return to Σ+ without interacting with Σ− (see Figure 3(a)),
(ii) the trajectory can leave Σ+ , impact Σ− with a negative velocity, and then return to Σ+
(see Figure 3(b)), or (iii) the trajectory can leave Σ+ , graze (impact with zero velocity) Σ− ,
and then return to Σ+ (see Figure 3(c)). Notice that scenario (c) naturally separates scenarios
(a) and (b), and thus Γ naturally separates the initial conditions in I − and I + .
3.3. Periodic orbits. As reported in [28] the system in question reveals a plethora of
different dynamical behavior. To identify the many different periodic solutions x∗ for the
impact map (3.16) we need to solve one or more of the equations
Π(x∗ ) = x∗ , Π2 (x∗ ) = Π(Π(x∗ )) = x∗ , Π3 (x∗ ) = x∗ , etc.
These equations give rise to implicitly defined solutions, and they can thus be solved nu-
merically. In some special cases analytical solutions can be found (see further [18, 27] and
section 4).

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1701

Since fixed points or periodic orbits of the impact map and periodic orbits in the stro-
boscopic sense of the original ordinary differential equation do not necessarily represent the
same solutions, we need a systematic way in which the different types of periodic solutions
can be identified. To classify the periodic solutions we use the notation introduced in [18].
We let P (m, n+ , n− ) denote a periodic solution of forcing period m ∈ N, where n± ≥ 0 denote
the number of times the orbit impacts the Σ± boundaries, respectively. This notation cannot
identify every type of periodic orbit uniquely, but it is sufficient for our needs.
4. Numerical methodology. To get a handle on the underlying dynamics of the gear
model, we will employ several different numerical methods, namely, basins of attraction, bi-
furcation diagrams, and manifold computations. In this section we give a short introduction
to these methods.
For systems with many different solutions of different periodicity it can be beneficial to
find the basins of attraction for all coexisting attractors. It also gives a visual measure of the
complexity of the dynamics and how sensitive solutions are to perturbations. To compute the
basins of attraction we apply the method of cell-to-cell mapping [20] to the impact map Π
described in section 3.2. For the purpose of visualization each basin is colored according to
its itinerary, i.e., the pattern of impacts that occur with both backlash boundaries. Shorter
itineraries are represented by shorter wavelength (bluer) colors. At the two extremes of the
color scale, dark blue denotes behavior akin to permanent contact and red denotes chaotic
behavior. This results in a uniform color scale across the results presented; the same color
in more than one picture denotes the same solution type. In each case the grazing curve Γ
is overlaid in white, and the scale is chosen so that transitions on either side of the grazing
curve can be observed.
To analyze the different attractors observed in the basin-of-attraction plots, we compute
one-parameter bifurcation diagrams for the impact map (3.16). In [18] and [27] the procedure
for explicitly constructing periodic solutions of type P (1, 1, 0) and P (1, 1, 1) is described. We
employ these techniques to construct the stable and unstable branches of these solutions while
they remain period-one. For more complex behavior we use brute-force simulations.
To further analyze the basins of attraction, we calculate stable and unstable manifolds
for saddle-point solutions. Period-one saddle-point solutions can be constructed, as discussed
above, and these saddle points then enable numerical calculation of their stable (and unsta-
ble) manifolds using DsTool (Dynamical Systems Toolkit) [3, 24] with the extension package
Man1D [15].
The main result in this paper is the understanding of how grazing bifurcations and smooth
bifurcations merge in codimension-two bifurcations (in δ and ε). Branches of grazing bifurca-
tions are calculated by explicitly constructing P (1, 1, 1) solutions (as described in [18] and [27]),
which are both in-phase and out-of-phase with the forcing, where ε is considered as an un-
known parameter and the impact velocity at Σ− is set to 0. For prescribed values of δ,
corresponding ε values are found numerically, and these pairs of values can then be plotted in
the δ-ε plane. Branches of saddle-node bifurcations at which P (1, 1, 1) solutions are born and
branches of period-doubling bifurcations are found in a similar way. In each case P (1, 1, 1)
and P (2, 2, 2) solutions, respectively, are constructed, ε is set as an unknown parameter, and
the values of ε for which there is a zero eigenvalue are calculated for two different values of δ.
A secant-predictor with pseudoarclength parameterization of the branch [4] is built on top of

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


1702 J. F. MASON AND P. T. PIIROINEN

MATLAB’s “fsolve” function and then applied to generate the bifurcation curves in the δ-ε
plane.
It is also possible to calculate the bound for the existence of the period-one solutions of
type P (1, 1, 0). This boundary may be identified as saddle-node bifurcations (see [18]) and
can be expressed as the analytical expression
 
δ
(4.1) ε > 2 − δ coth .
2

5. Numerical results.
5.1. Periodic orbits and codimension-one bifurcations. As shown in [18, 27, 28] the gear
system under analysis shows a remarkable complexity of dynamical behavior. To highlight
the mechanisms behind the complex dynamics in this section and section 5.2 we will use the
different numerical analysis techniques described in section 4.
First, in Figure 4 we show basins of attraction for fixed damping δ and half-backlash width
β and nine different values of the eccentricity ε, varying between 0.056 and 0.1. It is clear
from this figure that the number of attractors decreases with decreasing eccentricity as well
as the size of their basins of attraction. When ε reaches a certain value all other attractors
apart from the permanent-contact solution disappear, as shown in Figure 4(a) (ε = 0.056).
We observe what look like fractal basin boundaries separating a period-four (yellow) and
period-eight (dark orange) solution in Figure 4(g) (ε = 0.089). We also notice in Figure 4(i)
(ε = 0.1) that the basins appear to be fragmented, which indicates that for some parameter
values there are chaotic attractors present (see further Figure 5 and [28]).
Second, to understand how attractors are born and how they disappear, we compute a
one-parameter bifurcation diagram for the impact map Π where the eccentricity ε is varied
between 0.05 and 0.15. Projections of the bifurcation diagram onto the ε-x2 and ε-τ planes
are displayed in Figure 5(a) and (b), respectively. The bifurcation diagrams are zoom-ins of
the region in the basins-of-attraction plots where the impact velocity x2 at Σ+ is between
3.7 and 4.9, as we are concerned with how attractors in this region are affected by varying ε.
The reason for displaying both these projections is that they reveal the relationships between
the attractors and the variables x2 and τ . For instance, in Figure 5(a) we see that the impact
velocities along the branches A (unstable) and C (stable) do not change at all while the
impact velocities along the branches B (unstable) and D (stable) change significantly. We
can now compare the bifurcation diagrams in Figure 5 with the basin-of-attraction plots in
Figure 4(a). For small values of ε (cf. Figure 4(a)), all initial conditions result in behavior
akin to permanent contact, and there are no P (1, 1, 0) or P (1, 1, 1) solutions present. As
the eccentricity ε increases, the P (1, 1, 0) solutions, A and C, are born at the saddle-node
bifurcation SN1 (ε ≈ 0.0596), and in Figure 4(b) the basin corresponding to C is observable in
light blue. At the saddle-node bifurcation SN2 (ε ≈ 0.0687) the P (1, 1, 1) solutions, B and D,
are born, and in Figure 4(d) a very small basin corresponding to D emerges in light orange.
The branch D undergoes a period-doubling bifurcation at PD1 (ε ≈ 0.076), and in Figure 4(e)
the light orange basin has turned dark orange and now corresponds to a P (2, 2, 2) solution.
As ε is increased further, the part of the attractor born at PD1 undergoes further period-
doubling and grazing bifurcations, and eventually a chaotic attractor appears at ε ≈ 0.096

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1703

0 0 0

7 7 7
0
(a) ε = 0.056 1 0
(b) ε = 0.0615 1 0
(c) ε = 0.067 1

0 0 0

7 7 7
0 (d) ε = 0.0725 1 0 (e) ε = 0.078 1 0 (f) ε = 0.0835 1

0 0 0

7 7 7
0 1 0 1 0 1
(g) ε = 0.089 (h) ε = 0.0945 (i) ε = 0.1
τ

Figure 4. Basins of attraction for the impacting contact model, (3.4) and (3.5), for δ = 0.6, β = 0.6 with
eccentricity varying between ε = 0.056 and ε = 0.1. Each plot has impact phase τ on the horizontal axis and
the initial-velocity parameter v on the vertical axis. In case (a) all initial conditions result in behavior akin
to permanent linear contact, and in all other cases there is more than one attractor present. In each figure a
regular grid of 1000 × 1000 cells has been used in the basin-of-attraction calculations, where each cell represents
a different initial condition given by (x1 , x2 , τ )T = (β, −v, τ )T .

and disappears at ε ≈ 0.1109 (see further [35]). The branches B and C coalesce at ε ≈ 0.1445
in a grazing bifurcation denoted G1 .
Third, in Figure 6 we show a blow-up of the basins of attraction in Figure 4(i), where we
have plotted the P (1, 1, 0) (denoted A with symbol  × ) and P (1, 1, 1) (B,  × ) saddle-point
solutions and their corresponding stable manifolds, W s (A) and W s (B), in magenta and green,
respectively. We see that the stable manifolds W s (A) and W s (B) encapsulate the basins of
attraction, as expected. In addition, we have overlaid the P (1, 1, 0) attractor (C,  + ), the
P (1, 1, 1) saddle (D,  × ), and the grazing curve Γ in white. The dark region, denoted ΔL ,
at the top part of the figure corresponds to initial conditions (x1 , x2 , τ )T = (β, −v, τ )T that
will lead to quiet solutions that will only impact the Σ+ boundary with low velocities. For
slightly larger v values there is a pale region which is the basin of attraction ΔC for the
P (1, 1, 0) solution C, and the fragmented red region ΔCh corresponds to a chaotic attractor
(see further [28]).

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


1704 J. F. MASON AND P. T. PIIROINEN

4.9
(a)
x2
PD1
4.6 D

SN2

4.3

4.0
SN1 B G1

A,C
3.7
0.05 0.07 0.09 0.11 0.13 0.15
ε

1.0
(b)
τ C

B
0.8
G1

D
0.6 SN1
PD1 A
SN2

0.4
0.05 0.07 0.09 0.11 0.13 0.15
ε

Figure 5. Bifurcation diagram of (a) impact velocity and (b) impact phase at Σ+ , against eccentricity for
the impacting-contact model, (3.4) and (3.5), for δ = 0.6 and β = 0.6. In both panels stable solutions are shown
as solid lines and unstable solutions as dashed lines.

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1705

0
ΔL

ΔC
W s (A)

ΔCh A C
B
Γ D

W s (B)

7
0 0.5 τ 1

Figure 6. A blow-up of Figure 4(i) showing the basins of attraction for the impacting-contact model given
by (2.3), (2.4) for δ = 0.6, β = 0.6, and ε = 0.1. The stable manifolds W s (A) and W s (B) of the out-of-phase
P (1, 1, 0) saddle at A ( ×) and the in-phase P (1, 1, 1) saddle at B ( ×) are overlaid in magenta and green,
respectively. The in-phase P (1, 1, 0) attractor is also overlaid at C ( +) and the out-of-phase P (1, 1, 1) saddle
at D ( ×). The grazing curve Γ is overlaid in white. The three different basins, permanent contact (quiet)
solutions, period-one solutions, and chaotic solutions, are labeled as ΔL , ΔC , and ΔCh , respectively. The
manifolds were generated with DsTool [3, 15, 24]. In this figure a regular grid of 1000 × 1000 cells has been
used in the basin-of-attraction computations, where each cell represents a different initial condition given by
(x1 , x2 , τ )T = (β, −v, τ )T .

5.2. Codimension-two bifurcations. It is clear from Figures 4–6 that the period-one solu-
tions A, B, C, and D play key roles in the underlying complex dynamics of this gear model. To
understand how their births and deaths are interlinked, we have calculated a two-parameter
bifurcation diagram, in δ and ε, which is depicted in Figure 7. In this figure we have plot-
ted branches of the saddle-node bifurcations SN1 and SN2 , the period-doubling bifurcation
PD1 , and the grazing bifurcations G1 . These bifurcations can all be seen in the bifurcation
diagrams in Figure 5, corresponding to the vertical dashed line in Figure 7. Furthermore,
we have included a second branch of grazing bifurcations, G2 , where the periodic solutions A
and D coalesce (not seen in Figure 5).
The two-parameter bifurcation curves depicted in Figure 7 divide the parameter space
(in δ and ε) into five separate regions (I–V), where the dynamical characteristics of the four
period-one solutions A–D can be classified as follows. In region I no period-one solutions exist;

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


1706 J. F. MASON AND P. T. PIIROINEN

0.30
ε
0.25 G1 G2

0.20
IV V

0.15
COD22

0.10
PD1
III
0.05 COD21
SN1 I
II SN2
0
0.4 0.5 0.6 0.7 0.8 0.9 1.0
δ
Figure 7. A two-parameter bifurcation diagram in δ and ε. The branch SN1 corresponds to saddle-
node bifurcations of the P (1, 1, 0) solutions A and C. Similarly, the branch SN2 corresponds to saddle-node
bifurcations of the P (1, 1, 1) solutions B and D. The branch PD 1 represents the period-doubling bifurcation
of D, and G 1 and G 2 denote grazing bifurcations. SN2 and PD 1 meet G 1 in a codimension-two point COD21
at (δ, ε) ≈ (0.7590, 0.09550), and SN1 meets G 1 and G 2 in another codimension-two point COD22 at (δ, ε) ≈
(0.7701, 0.09789). The dashed vertical line indicates the range of ε examined in the one-parameter bifurcation
diagrams in Figures 5(a) and (b), and the dashed rectangle corresponds to the region of magnification around
the two codimension-two points shown in Figure 8.

in region II stable (C) and unstable (A) P (1, 1, 0) solutions exist; in region III stable (C) and
unstable (A) P (1, 1, 0) solutions and stable (D) and unstable (B) P (1, 1, 1) solutions exist; in
region IV stable (C) and unstable (A) P (1, 1, 0) solutions and two unstable P (1, 1, 1) solutions
(B and D) exist; and finally, in region V, an unstable P (1, 1, 0) solution (A) and an unstable
P (1, 1, 1) solution (D) exist. It is now relatively easy to compare the observed transitions
across the bifurcations SN1 , SN2 , PD1 , and G1 in Figures 4 and 5 with the two-parameter
bifurcation diagram in Figure 7. However, the two-parameter bifurcation diagram does not
tell us anything about other bifurcations or the births and deaths of other periodic and chaotic
attractors.
When analyzing Figure 7 it would initially appear that all five branches of bifurcations
coalesce at the same point. However, a magnification of this region (corresponding to the
overlaid rectangle in Figure 7) uncovers two codimension-two points which are very close to
each other; see Figure 8. At this scale we observe that the curves PD1 and SN2 coalesce tan-
gentially with G1 at COD21 ((δ, ε) ≈ (0.7590, 0.09550)), and that SN1 coalesces tangentially
with G1 and G2 at COD22 ((δ, ε) ≈ (0.7701, 0.09789)). This scenario is very similar to the one

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1707

0.100
G2
ε

0.098 COD22

0.096
COD21

0.094 G1 SN1

SN2
PD1
0.092
0.74 0.75 0.76 0.77 0.78
δ
Figure 8. An enlargement of the two-parameter bifurcation diagram in Figure 7 around the two
codimension-two points. At this scale it is clear that the curves PD 1 and SN2 coalesce with G 1 at COD21
and that SN1 coalesces with G 1 and G 2 at COD22 . The dashed vertical line corresponds to the range of ε
examined in the one-parameter bifurcation diagrams in Figure 9.

observed in a model for impact microactuators [8]. In [7] a classification and analysis of various
codimension-two bifurcations is made, and there the COD22 point is classified as a border-
fold bifurcation. Further, the dashed vertical line in Figure 8, located between COD21 and
COD22 , corresponds to the one-parameter bifurcation diagrams for the impact map Π shown
in Figures 9(a) and (b). In the bifurcation diagrams the P (1, 1, 0) solutions A (unstable) and
C (stable) are born in a saddle-node bifurcation at SN1 (ε ≈ 0.09585). For increasing ε there
is a grazing bifurcation at G1 (ε ≈ 0.09606) where the unstable P (1, 1, 1) solution D is born,
the stable solution C disappears, and a chaotic attractor is born. We notice in Figure 9 that
the chaotic attractor grows in size from G1 until it disappears at ε ≈ 0.09742, where it collides
with the unstable solution A. This is different from what we saw in Figure 5, at δ = 0.6, where
the chaotic attractor suddenly appears at a grazing bifurcation. Some explanation of this is
given in the following.
From the bifurcation diagrams in Figure 5 we notice that the grazing bifurcations G1
and G2 act as nonsmooth folds, where a solution of type P (1, 1, 0) and a solution of type
P (1, 1, 1) meet nontangentially and annihilate each other; see [12, 13]. It is also interesting to
note that along G1 (for increasing δ and ε; see Figure 8) up to the point COD21 the solutions
B and C disappear at G1 . However, for δ and ε values along G1 in between COD21 and
COD22 the solution B never appears, C still disappears, and D is created at G1 . The reason

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


1708 J. F. MASON AND P. T. PIIROINEN

(a) (b)
x2 τ
C
G1
D D

A
SN1 G1
A SN1
A,C
ε ε

Figure 9. A local bifurcation diagram of (a) impact velocity and (b) impact phase at Σ+ against eccentricity
for the impacting-contact model, (3.4) and (3.5), δ = 0.762, and β = 0.6. In both panels stable solutions are
shown as solid lines and unstable solutions as dashed lines.

for this is that for these δ values SN2 never exists, and thus B never has the possibility to
appear, but D has to exist above G1 , and thus D is born at G1 (instead of SN2 ). For δ and ε
values larger than at COD22 the solutions A and D are born in G2 and exist above G2 .
To gain a better understanding of the transitions along G1 and G2 , we compute the
nontrivial Floquet multipliers λ1 , λ2 of the monodromy matrix, and thus the stability, for the
grazing P (1, 1, 0) solutions along G1 and G2 (see [2, 1, 26, 12]). We find that λ1 λ2 = e−δ , i.e.,
that the stability of the period-one solutions (at grazing) depends solely on δ. For increasing
values of δ (which are less than the value of δ at COD21 ), we have a pair of complex conjugate
Floquet multipliers with moduli less than 1. At COD21 ((δ, ε) ≈ (0.7590, 0.09550)), the
δ
Floquet multipliers turn real, with λ = λ1,2 = e− 2 ≈ 0.6842. Increasing δ and ε further
along G1 (between COD21 and COD22 ), we have that the largest in magnitude Floquet
multiplier is λ ∈ (2/3, 1). Since we have grazing bifurcations along G1 , the dynamics can
locally be described by square-root maps [29]. For a one-dimensional square-root map it
is shown that if the Floquet multiplier λ ∈ (2/3, 1) is on the non–square-root side of the
discontinuity, we should expect robust chaos without any periodic windows (see [31, 12]).
This is indeed what we observe in the bifurcation diagram in Figure 9 (for δ = 0.762 and
varying ε) at the bifurcation point G1 , where λ1 ≈ 0.8349 and λ2 ≈ 0.5590. Increasing δ
and ε further along G1 until reaching COD22 , we find that λ1 = 1 (and λ2 = e−δ ), which we
should have as SN1 and G1 coalesce. For values of δ and ε larger than at COD22 we have
λ1 > 1, and thus all solutions are unstable at grazing and in the region above G1 and G2
(region V in Figure 7).
If we assume that the dynamics of our two-dimensional system can be described by a one-
dimensional square-root map in the vicinity of grazing, then the largest δ value that would
lead to robust chaos is given by λ = λ1,2 = 2/3 ⇒ δ ≈ 0.8109. (Observe that δ ≈ 0.8109 is
greater than the δ-value at COD22 , and thus we will always have robust chaos at the grazing
bifurcations along G1 between COD21 and COD22 for the given parameter values.)
As mentioned earlier, no solutions other than the permanent-contact solutions are present
for δ and ε values that lie below the SN1 and G2 curves (region I in Figure 7). The practical

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1709

implication of this is that if we could somehow eliminate the two codimension-two points,
then all nonquiet solutions would disappear since the stable manifolds W s (A) and W s (B)
would never appear, and this is something we observe for small values of ε (see Figure 4(a)).
However, for gear manufacturers this limit is hard (expensive) to achieve and thus not an
attractive option (see further [18, 27, 28]).
6. Discussion. We have used a combination of analytical and numerical techniques, name-
ly, bifurcation-diagrams, basin-of-attraction computations, and manifold calculations, to gain
a better understanding of the complex dynamics of a simple model of gear rattle. We found
stable and unstable periodic solutions, and, for some parameter regimes, chaotic attractors.
Some of these solutions correspond to quiet solutions in which the gears remain permanently in
contact, but often they coexist with stable solutions that cause rattle. This paper reveals that
the main cause for the rattling solutions is due to two codimension-two bifurcations. From
a practical perspective it is insightful to note that in the two-parameter bifurcation diagram
in Figure 7 the two codimension-two bifurcations act as organizing centers for the dynamics
of this gear model. The stable manifold W s (A) in Figure 6 engulfs a basin corresponding
to chaotic behavior, which is the main source of the rattle problem. This means that if
we can destroy COD22 , then A disappears, and so does the basin corresponding to chaotic
behavior.
The one-parameter bifurcation diagrams in Figures 5(a),(b) and 9(a),(b) reveal a rich
structure of both smooth and discontinuity-induced bifurcations (grazing bifurcations). How-
ever, there is a transition that we have not examined in this paper. At ε ≈ 0.1109 in Figure 5
and ε ≈ 0.09742 in Figure 9 the chaotic attractor suddenly disappears. Initial investigations
reveal that this is not due to a grazing bifurcation, as one might first expect, but a boundary-
crisis bifurcation (see [25]) where the stable and unstable manifolds, W s (A) and W u (A), meet
tangentially. Preliminary computations indicate that this homoclinic tangency is also born in
the codimension-two point COD22 (see Figure 7 and [35]). This scenario is currently under
investigation.
Acknowledgments. The authors are grateful to the anonymous referees for providing
constructive suggestions which have helped to improve this paper. J. Mason would also like
to thank Drs. R. E. Wilson and M. E. Homer, University of Bristol, both for introducing her
to the gear-rattle problem and for many useful discussions.

REFERENCES

[1] J. Adolfsson, H. Dankowicz, and A.B. Nordmark, 3D passive walkers: Finding periodic gaits in
the presence of discontinuities, Nonlinear Dynam., 24 (2001), pp. 205–229.
[2] M.A. Aizerman and F.R. Gantmakher, On the stability of periodic motions, J. Appl. Math. Mech.,
22 (1958), pp. 1065–1078.
[3] A. Back, J. Guckenheimer, M. R. Myers, F. J. Wicklin, and P. A. Worfolk, DsTool: Computer
assisted exploration of dynamical systems, Notices Amer. Math. Soc., 39 (1992), pp. 303–309.
[4] D.A.W. Barton, Numerical methods for fold- and flip-bifurcation locations in PWS systems. Personal
communications, 2008.
[5] G.W. Blankenship and A. Kahraman, Steady state forced response of a mechanical oscillator with
combined parameter excitation and clearance nonlinearity, J. Sound Vibration, 185 (1995), pp. 743–
765.

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


1710 J. F. MASON AND P. T. PIIROINEN

[6] C. J. Budd and F. Dux, Chattering and related behaviour in impact oscillators, Phil. Trans. Roy. Soc.
London A, 347 (1994), pp. 365–389.
[7] A.C. Colombo and F. Dercole, Discontinuity induced bifurcations of non-hyperbolic cycles in non-
smooth systems, http://arxiv.org/abs/0905.3361 (2009).
[8] H. Dankowicz and X. Zhao, Local analysis of co-dimension-one and co-dimension-two grazing bifurca-
tions in impact microactuators, Phys. D, 2 (2005), pp. 238–257.
[9] S.L.T. de Souza and I.L. Caldas, Basins of attraction and transient chaos in a gear-rattling model, J.
Vib. Control, 7 (2001), pp. 849–862.
[10] F. Dercole and S. Maggi, Detection and continuation of a border collision bifurcation in a forest fire
model, Appl. Math. Comput., 168 (2005), pp. 623–635.
[11] M. di Bernardo, C.J. Budd, and A.R. Champneys, Grazing, skipping and sliding: Analysis of the
nonsmooth dynamics of the DC/DC buck converter, Nonlinearity, 11 (1998), pp. 858–890.
[12] M. di Bernardo, C. J. Budd, A. R. Champneys, and P. Kowalczyk, Piecewise Smooth Dynamical
Systems: Theory and Applications, Springer-Verlag, London, 2008.
[13] M. di Bernardo, C. J. Budd, A. R. Champneys, P. Kowalczyk, A. B. Nordmark, G. Olivar
Tost, and P. T. Piiroinen, Bifurcations in nonsmooth dynamical systems, SIAM Rev., 50 (2008),
pp. 629–701.
[14] M. di Bernardo, P. Kowalczyk, and A. Nordmark, Sliding bifurcations: A novel mechanism for
the sudden onset of chaos in dry friction oscillators, Internat. J. Bifur. Chaos Appl. Sci. Engrg., 13
(2003), pp. 2935–2948.
[15] J. P. England, B. Krauskopf, and H. M. Osinga, Computing one-dimensional stable manifolds of
planar maps without the inverse, SIAM J. Appl. Dyn. Syst., 3 (2004), pp. 161–190.
[16] M.H. Fredriksson and A.B. Nordmark, Bifurcations caused by grazing incidence in many degrees of
freedom impact oscillators, Proc. Roy. Soc. London Ser. A, 453 (1997), pp. 1261–1276.
[17] M. Gallegati, L. Gardini, T. Puu, and I. Sushko, Hick’s trade cycle revisited: Cycles and bifurca-
tions, Math. Comput. Simulation, 63 (2003), pp. 505–527.
[18] C. K. Halse, R. E. Wilson, M. di Bernardo, and M. E. Homer, Coexisting solutions and bifurcations
in mechanical oscillators with backlash, J. Sound Vibration, 305 (2007), pp. 854–885.
[19] S.J. Hogan, On the dynamics of rigid-block motion under harmonic forcing, Phil. Trans. Roy. Soc.
London A, 425 (1989), pp. 441–476.
[20] C. S. Hsu, Cell-to-Cell Mapping. A Method of Global Analysis for Nonlinear Systems, Springer-Verlag,
New York, 1987.
[21] K. Karagiannis and F. Pfeiffer, Theoretical and experimental investigations of gear-rattling, Nonlin-
ear Dynam., 2 (1991), pp. 367–387.
[22] P.S. Keogh and M.O.T. Cole, Rotor vibration with auxiliary bearing contact in magnetic bearing
systems Part 1: Synchronous dynamics, Proceedings of the Institute of Mechanical Engineers, Part
C: Journal of Mechanical Engineering Science, 217 (2003), pp. 377–392.
[23] C. Knudsen, R. Feldberg, and H. True, Bifurcations and chaos in a model of a rolling railway
wheelset, Phil. Trans. Roy. Soc. London A, 338 (1992), pp. 455–469.
[24] B. Krauskopf and H. M. Osinga, Investigating torus bifurcations in the forced Van der Pol oscillator,
in Numerical Methods for Bifurcation Problems and Large-Scale Dynamical Systems, E.J. Doedel
and L.S. Tuckerman, eds., Springer-Verlag, New York, 2000, pp. 198–208.
[25] Y. A. Kuznetsov, Elements of Applied Bifurcation Theory, Springer-Verlag, New York, 2003.
[26] R.I. Leine and H. Nijmeijer, Dynamics and Bifurcations of Non-Smooth Mechanical Systems, Lecture
Notes in Applied and Computational Mechanics 18, Springer-Verlag, Berlin, Heidelberg, 2004.
[27] J. F. Mason, M. E. Homer, and R. E. Wilson, Mathematical models of gear rattle in Roots blower
vacuum pumps, J. Sound Vibration, 308 (2007), pp. 431–440.
[28] J. F. Mason, P. T. Piiroinen, R. E. Wilson, and M. E. Homer, Basins of attraction in nonsmooth
models of gear rattle, Internat. J. Bifur. Chaos Appl. Sci. Engrg., 19 (2009), pp. 203–224.
[29] A. B. Nordmark, Non-periodic motion caused by grazing incidence in impact oscillators, J. Sound
Vibration, 2 (1991), pp. 279–297.
[30] A. B. Nordmark and P. T. Piiroinen, Simulation and stability analysis of impacting systems with
complete chattering, Nonlinear Dynam., 58 (2009), pp. 85–106.
[31] H. Nusse, E. Ott, and J. Yorke, Border-collision bifurcations: An explanation for observed bifurcation
phenomena, Phys. Rev. E (3), 49 (1994), pp. 1073–1076.

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


GLOBAL DYNAMICS OF A GEAR MODEL 1711

[32] G. Osorio, M. di Bernardo, and S. Santini, Chattering and complex behavior of a cam-follower sys-
tem, in ENOC-2005 Fifth Euromech Nonlinear Dynamics Conference, Eindhoven, The Netherlands,
2005.
[33] R.G. Parker, S.M. Vijayakar, and T. Imajo, Non-linear dynamic response of a spur gear pair:
Modelling and experimental comparisons, J. Sound Vibration, 237 (2000), pp. 435–455.
[34] P.T. Piiroinen and H.J. Dankowicz, Low-cost control of repetitive gait in passive bipedal walkers,
Internat. J. Bifur. Chaos Appl. Sci. Engrg., 15 (2005), pp. 1959–1973.
[35] P.T. Piiroinen and J.F. Mason, Smooth and nonsmooth tangencies in a gear model with impacts,
accepted for publication in IFAC-PapersOnline, 2009.
[36] S. Theodossiades and S. Natsiavas, Non-linear dynamics of gear-pair systems with periodic stiffness
and backlash, J. Sound Vibration, 229 (2000), pp. 287–310.
[37] J.M.T. Thompson and R. Ghaffari, Chaotic dynamics of an impact oscillator, Phys. Rev. A, 27 (1982),
pp. 1741–1743.
[38] H. Vinayak and R. Singh, Multi-body dynamics and modal analysis of compliant gear bodies, J. Sound
Vibration, 210 (1998), pp. 171–214.
[39] G.S. Whiston, Singularities in vibro-impact dynamics, J. Sound Vibration, 152 (1992), pp. 427–460.

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.


Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

You might also like