Journal of Sound and Vibration 330 (2011) 955–975

Contents lists available at ScienceDirect

Journal of Sound and Vibration

journal homepage: www.elsevier.com/locate/jsvi

Chaos in brake squeal noise

S. Oberst, J.C.S. Lai 
Acoustics & Vibration Unit, School of Engineering and Information Technology, The University of New South Wales at the Australian Defence Force Academy,
Canberra, ACT 2600, Australia

a r t i c l e in fo abstract

Article history: Brake squeal has become an increasing concern to the automotive industry because of
Received 11 January 2010 warranty costs and the requirement for continued interior vehicle noise reduction. Most
Received in revised form research has been directed to either analytical and experimental studies of brake squeal
3 September 2010
mechanisms or the prediction of brake squeal propensity using finite element methods.
Accepted 9 September 2010
By comparison, there is a lack of systematic analysis of brake squeal data obtained from
Handling Editor: H. Ouyang
Available online 12 October 2010 a noise dynamometer. It is well known that brake squeal is a nonlinear transient
phenomenon and a number of studies using analytical and experimental models of
brake systems (e.g., pin-on-disc) indicate that it could be treated as a chaotic
phenomenon. Data obtained from a full brake system on a noise dynamometer were
examined with nonlinear analysis techniques. The application of recurrence plots
reveals chaotic structures even in noisy data from the squealing events. By separating
the time series into different regimes, lower dimensional attractors are isolated and
quantified by dynamic invariants such as correlation dimension estimates or Lyapunov
exponents. Further analysis of the recurrence plot of squealing events by means of
recurrence quantification analysis measures reveals different regimes of laminar and
random behaviour, periodicity and chaos-forming recurrent transitions. These results
help to classify brake squeal mechanisms and to enhance understanding of friction-
related noise phenomena.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction

In automotive industry brake squeal has become an important cost factor because of customer dissatisfaction. In North
America, up to one billion dollars p.a. were spent on noise, vibration and harshness (NVH) issues [1] while friction material
suppliers allocated more than half their budgets to deal with NVH problems [2]. According to a J.D. Power survey
conducted in 2002, 60 percent of warranty claims concerning the brake corner are due to brake squeal [3]. Up to 5 percent
of the USA’s gross national product can be accounted for by losses due to friction and wear, which includes brake noise
related problems [4]. Despite almost 80 years of research and a good deal of progress having been made, the underlying
mechanisms of noise generation are still not fully understood. Recent reviews cover acoustics of friction [1], mechanisms of
squeal [5] and the application of numerical methods such as the complex eigenvalue analysis (CEA) or the time domain
analysis to study brake squeal propensity [6] and design in terms of structural vibrations [7]. As Oberst and Lai [8] have
pointed out, most of the research undertaken over the last two decades has focused primarily on exploring the mechanism
of brake squeal using simplified analytical models or applying the finite element method (FEM) complemented by
experiments to determine unstable vibration modes in a brake system. It was stated that most likely, no complete and

 Corresponding author.
E-mail address: [email protected] (J.C.S. Lai).

0022-460X/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jsv.2010.09.009
956 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

practical solution is in sight in the foreseeable future. Therefore, as was concluded in Hoffmann and Gaul [9], brake squeal
as a friction-induced nonlinear and highly sensitive problem could probably only be solved by incorporating uncertainties
in modelling even though the problem is deterministic.
Deterministic chaos as an expression of nonlinearity which is extremely sensitive to changes in initial conditions
accounts for the loss of long-term predictability [10]. In an utmost stage a seemingly random behaviour can be observed
(e.g., turbulence). A good introduction to chaos theory is presented by Schuster and Just [11] in which the four main routes
to chaos are described: the Feigenbaum, the Manneville–Pomeau, the Ruelle–Takens–Newhouse and intermittency I–III
routes to chaos. A handy definition of chaos is presented in Linz and Sprott [12]: Physical flows need to (i) form a strange
attractor, with underlying non-periodic long time behaviour, (ii) have a sensitive dependence on initial conditions,
(iii) have a dimension greater than two with (iv) some nonlinearity in the vector field.
Tables 1–5, containing 45 references since 1938, present a chronological overview about research work on friction
oscillators, pin-on-disc systems, beam-on-disc systems and laboratory brake set-ups which indicate, that deterministic
chaos is observed in these systems and that conventional methods do not provide adequate insight. The following main
observations can be made by analysing the papers listed in Tables 1–5.
Chaos in simplified models: Analytical models listed in Table 1 are mostly 1-dof systems but display very rich nonlinear
and chaotic behaviour. For the forced 1-dof dry friction oscillator, although it is governed by a second-order differential
equation (see Eq. (A.1) in Appendix A), with displacement and velocity as state variables, a third state variable 2pft arises
from the sinusoidal driving force term [13,14], so that the system can only be fully described in a phase space dimension of
at least 3. Further an additional equation (Eq. (A.2) in Appendix A) has to account for the beginning and end of slip. This
equation allows switching between two systems of differential equations [15–18]. The evolution from periodic to quasi-
periodic and finally chaotic regimes (i.e., from torus in phase space to intermittency and strange attractors) for a
harmonically driven system has been discussed [19–21]. The one degree-of-freedom system of a dry friction oscillator
represents a minimal model of a physical flow of a friction system able to exhibit exorbitant complexity studied by Popp
and Stelter [19], Feeny and Moon [20,21] and later by Hinrichs et al. [22]. Shin et al. [23,24] showed that a forced 2-dof, dry
friction analytical model with negative friction-velocity gradient characteristic developed chaotic pad motion when

Table 1
Works on oscillators.

Author Approach Findings

Popp and Stelter [19] Analytical, 1-dof, (1990)  Higher periodic motions
 Intermittent and toroidal chaos

Feeny and Moon [20] Analytical, 1-dof, (1992)  Friction highly nonlinear
 Known maps differ from chaotic models

Feeny and Moon [21] Experimental and analytical, 1-dof, (1994)  Mass overtakes belt (overshooting)
 State variable friction law
 Stick–slip attractors: same family

Hinrichs et al. [22] Experimental and analytical, 1-dof, (1997)  Rich bifurcational behaviour
 Experimental results match well

Thomson [32] Analytical, 1-dof, (1999)  High-frequency external excitation can

effectively cancel the effect of the
negative friction-velocity curve

Feeny and Moon [59] Experimental, 1-dof, (2000)  Chaos controlled by quenching stick–slip

Shin et al. [23,24] Analytical, 2-dof, (2002)  Chaos in reduced order brake model
 Pad and disc couple in sliding motion
 Damping can have negative effects

Herve et al. [57,58] Analytical 2-dof, (2009)  Limit cycle and quasi-periodic motion
 Destabilisation paradox
 Iso-damping: max growth of amplitudes,
but best configuration against instability
 Gyroscopic effects are not negligible

Paliwal et al. [61] Analytical, 2-dof, (2002)  Very sensitive to coupling-stiffness

 Instability increases limit cycle diameter

Popp [45] Experimental and analytical, 1-dof, (2005)  Bifurcation followed by mode coupling
 Active control of self-excited vibrations

Hetzler et al. [62] Analytical, 1-dof, (2007)  Hopf-bifurcations in sliding friction

 Basin of attraction inside limit cycle
 Linear CEA does not consider limitation
Ouyang [63,64] Analytical 4-dof, (2008/09)  State-feedback control of vibration
 Asymmetric system matrix challenging
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 957

Table 2
Works on pin-on-disc/plate systems.

Author Approach Findings

Bowden and Tabor [65] Experimental (1938)  Stick–slip on metal/metal contact

 Friction is a discontinuous process

Bowden and Tabor [66] Experimental (1942)  Vibration signal intermittent with jerks
 Velocity dependency of friction

Earles and Soar [67,68] Experimental (1971/78)  Confirmation of sprag-slip

Earles and Badi [69] Experimental (1984)  Abrupt changes in acceleration

 Squeal fugitive and high sensitivity to parameters

Earles and Chambers [70] Experimental (1987)  Damping: instability remains

 Nonlinear, fugitive character
 Intuitive predictions impossible

Tworzydlo et al. [71] Experimental and analytical (1999)  Surface separations

 Friction couples bending modes
 Bifurcations
 Harmonics due to limit cycles
 Irregular bursts and surface damage
 Frict. seizure: extreme amplitudes

Chen et al. [34] Experimental  Squeal but no mode coupling

 Squeal: negative or positive friction-velocity curve

Chen and Zhou [36,50] Experimental (2007)  Time delay between normal and friction force responsible for
squeal
 Squeal due to elastic vibration of friction system
Chen et al. [44]

Duffour and Woodhouse [72,73] Analytical (2004)  Instability boundaries extremely sensitive to damping
 Negative gradient of mðvs Þ destabilizes

Duffour and Woodhouse [38] Experimental and analytical (2007)  High capriciousness
 Linear theory detected 70 percent of squealing modes (CEA)

Butlin and Woodhouse [74] Analytical (2009)  System extremely sensitive to changes in parameters
 Neglecting physical effects can give rise to faulty results
 High sensitivity even at low modal amplitudes

Table 3
Works on beam-on-disc systems.

Author Approach Findings

Tarter [75] Experimental (1983)  Slotted rotor eliminates squeal

 Pad material and geometry important

Suganami et al. [76] Experimental (1998)  Possible squeal due to in-plane vibrations

Akay et al. [77] Experimental (2000)  Coupling of parts (locking-in)

Tuchinda et al. [47] Experimental (2001)  Instability by locking pin and disc mode
 Fulguration point (locking-out)

Baillet et al. [30] Numerical (2005)  Rotational speed o 6 rev=min ð  0:7 km=hÞ: stick–slip
 6–35 rev/min: stick–slip-separation
 Over 35 rev=minð  4:2 km=hÞ: slip followed by separation

frequencies of the pad and the disc were close (see also [25]). They considered the size of the limit cycle more important
than its existence.
Evolution in experimental set-ups: From Tables 2–5, a change in experimental set-ups of predominantly simplified brake
systems can be observed. In the beginning, pin-on-disc test rigs were used, running at low velocity regimes: researchers
were interested in stick–slip phenomena. Over the years, the focus of research changed to mechanisms from stick–slip then
sprag-slip to binary flutter theory with mode coupling favoured by most researchers nowadays, necessitating the
development of new experimental set-ups. The set-up of a pin-on-disc system changed to a beam-on-disc, which allows
for elastic support and coupling with other components. In Tables 4 and 5, the laboratory brake (LB) and the tribobrake
958 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

Table 4
Works on laboratory brake (LB).

Author Approach Findings

Giannini and Sestieri [43] Experimental and numerical (2006)  Squeal developed between 5 and 30 rev/min ð  0:5723:42 km=hÞ
 Angle of attack: high-frequency squeal at 01

Giannini et al. [26] Experimental analytical (2006)  Onset of squeal: linear methods
 After limit cycle: nonlinear models
 Feedback: coupling pad and rotor-motion

Gianinni and Massi [78] Experimental and numerical (2008)  Pad size changes split-mode frequency
 Sine, cosine and rotating squeal
 Rotating squeal higher nonlinearity

Massi and Giannini [56] Experimental (2008)  High pad pressure sensitivity of squeal
 Squeal: high quasi-random low frequency content

Table 5
Works on tribo brake (TB).

Author Approach Findings

Massi [79,55] Experimental and numerical (2006)  Low freq. modes: excited at negative angle
Massi et al. [55]  Squeal: phase difference of in- and out-of-plane direction
 Squeal develops only in right conditions
 Instability by higher modal damping (large tune-in range)
 Tangential pad vibration couples with disc bending mode

Giannini et al. [54] Experimental and numerical (2007)  Disc doublet modes veer (travelling waves)
 Extreme exfoliations and cracks (lining)
 Coupling: pad-disc, support-disc squeal
 Damping increases squeal propensity
 Extreme exfoliations and cracks (lining)

Massi et al. [41] Experimental and numerical (2007)  Linear CEA over-predicts instabilities
 Nonlinear simulation match experiments

Massi et al. [80] Experimental and numerical (2008)  Fatigue excitations: cracks within layers
 Exfoliations: squeal and large amplitude vibrations

(TB) can be seen as a further development of the beam-on-disc set-up which allows a wider range of parameter variations
under laboratory conditions. Real lining material with reduced dimensions was used. The TB is the successor of the LB and
allows for investigating tribological/dynamic aspects due to contact conditions [26]. Both set-ups, the LB and the TB are so
far used mainly to study the mechanism of mode coupling.
Stick–slip as the dominant mechanism in disc brake squeal seems to be abandoned, due to its required low relative
velocity [27]. However, one has to be careful with conclusions concerning the validity of mechanisms, as squeal develops in
a wide range of velocity regimes and different (unknown) nonlinearities are involved, such as local sticking and
detachments of the contacting surfaces [28–30]. Another example is the negative friction-velocity characteristic, which is
also a known destabilising mechanism [31–33]. By using a sinusoidally driven pin-on-plate apparatus, Chen et al. [34]
found experimentally, that squeal can occur not only in regions of negative but also positive friction-velocity gradients.
Further, the occurrence of squeal could not be attributed to modal coupling. Beloiu and Ibrahim [35] studied friction-
induced noise both numerically and experimentally by analysing the friction force, normal force and vibrations in the
frequency-time domain using discrete wavelet transform for a pad-on-disc system. They found that due to the time
variation of the contact forces, the motion of disc and pad is intermittently modulated by low and high frequency
oscillations. By interpreting squeal occurrence based on forces and vibrations, they found that squeal noise was
accompanied by a larger mean value of the instantaneous friction coefficient. In 2007, Chen and Zhou [36] applied various
time-frequency analysis techniques to analyse friction-induced vibrations obtained in a pin-on-plate test rig. They found
that contrary to the expectations arising from mechanisms of the negative friction-velocity gradient and modal coupling,
all vibrations are limited in magnitude by nonlinearity.
Nonlinearity: With the advent of sophisticated computer technology, FE methods are more often used in recent years by
applying linear methods as the complex eigenvalue analysis (CEA) to full brake systems. However, it is well known that the
ability of the complex eigenvalue method to predict brake squeal propensity is limited and needs in practice to be
supplemented by extensive dynamometer tests [37]. Linear methods [38] like the CEA either underestimate [39,40] or
overestimate [23,33] the number of unstable vibration modes. Massi et al. [41] investigated linear and nonlinear numerical
methods in a laboratory brake. The CEA in their simulation was useful to detect the frequency where an instability might
occur but to determine the strength of this instability, a nonlinear time domain analysis was necessary to estimate the size
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 959

of the limit cycle. It is unclear how nonlinear the behaviour of brake components is and how this nonlinearity influences
the overall acoustic radiation properties (see also [42]). It is estimated that the onset of squeal happens in linear conditions
and that nonlinearity stabilises the vibration amplitudes [43,44]. It has also been found that the development of brake
squeal noise is related to Andronov-Hopf-bifurcations in the time domain which are sometimes detectable by CEA [39,45].
However, other bifurcations may be possible, such as for instance border collision bifurcations, which are already observed
in analytical discontinuous systems [46].
In general, it is accepted that nonlinearity plays an important role in the generation of brake squeal (see for instance
[47,34,30,43,26]) and nonlinear models are necessary to estimate the vibration amplitude after the onset of squeal. Only
recently, nonlinearity and steady-state solutions of vibration signals of a pad-on-disc system exhibiting multi-instabilities
have been numerically investigated by Coudeyras et al. [48,49]. For a reduced finite element model, they used a generalised
constrained harmonic balance method to compute nonlinear periodic or pseudo-periodic responses but no chaotic
solutions were presented.
As nonlinearity, specifically the lifting off of the pad is studied and found to have an amplitude limiting function
[36,50,44]. The lift-off is also discussed in Hoffmann and Gaul [51] who found that an analytical 2-dof beam-on-disc
(sprag-slip) model does not have a steady sliding state and hence always shows limit-cycle behaviour, however, strong the
applied viscous damping is. Chen and Zhou [36] attributed the limitation of the vibration magnitude found in their
experiments to friction-induced vibrations being bounded by limit cycles due to the system nonlinearity. They found
further by conducting experiments and numerical simulations [50,44] that the contact separation between the two sliding
surfaces is a main nonlinear factor causing the squealing vibration to be bounded. In addition, they experienced that the
time delay between the varying normal force and the ensuing varying friction strongly influences squeal occurrence and
proposed this as a new mechanism of brake squeal.
Important in the context of nonlinearity and modelling is also the function the friction coefficient takes. In Ouyang et al.
[52], an analytical six degrees of freedom elastic slider on disc friction system is studied and stated to be non-smooth,
hence nonlinear, as conditions responsible for behaviour of the slider depend on the motion of the system. Important is,
that this even holds true for a constant friction coefficient. Mathematically, this nonlinearity is simply expressed by the
directionality of the friction coefficient which is dependent on the velocity (see also [16]) and represents a switching
mechanism. With their nonlinear model, Ouyang et al. [52] observed complex periodic and quasi-periodic vibrations of the
pad and the disc. They found that in-plane stiffness of the slider is a critical factor in triggering instability, whereas
transverse stiffness of disc or slider is stabilising. As the pressure is increased, the motion develops from periodic to quasi-
periodic and more unstable.
Damping: Ouyang et al. [52] elaborated for an elastic slider on disc system, damping of disc or slider in the transverse
direction stabilises unstable vibrations; although damping the slider in-plane reduces the magnitude of the vibration, it
fails to stabilise unstable motions. They also found that an increase in rotating speed of the drive point can cause the
system to destabilise. Shin et al. [23,24] have found that contrary to popular beliefs, increasing damping of either the pad
or the disc alone can cause instability as also observed in [53,54]. Massi et al. [55] found that for their tribo brake
(essentially a beam-on-disc set-up), a large modal damping can reduce the response of the damped mode and hence
prevent its participation in squeal coupling. However, a highly damped mode has higher probability to couple with other
modes close to its natural frequencies because its tune-in range becomes larger, resulting in a higher squeal propensity. In
Massi and Giannini [56] non-uniform modal damping of two coalescing modes (CEA) has shown to be responsible for
increased squeal propensity in a beam-on-disc system. They also found that a homogeneous damping distribution
decreases the likelihood of a developing dynamic instability. Herve et al. [57] studied friction-induced instabilities in a
2-dof analytical model of an automotive clutch, by examining changes in instability boundaries due to interactions
between structural proportional damping and gyroscopic effects. If the damping is non-evenly distributed, the gyroscopic
effects interact with the damping of the system and cannot be neglected anymore. In [58] it was found, that an iso-
distribution of damping is desirable in order to avoid destabilisation (see also [23,24,53,54]), but an iso-distribution is also
the worst configuration of damping for reducing the amplitudes of self-excited vibrations. In their nonlinear analysis of the
2-dof analytical model, stable quasi-periodic and chaotic motions were observed, highlighting the complex effects which
may be present in a real brake system.
Control of brake squeal: Of interest is also the control of vibrations due to friction-induced instabilities. For a
harmonically driven analytical 1-dof dry friction oscillators, it has been found that quenching stick–slip vibrations can
effectively eliminate destabilisation due to a negative friction-velocity characteristic [32]. Feeny and Moon [59] showed
that chaos inherent to a stick–slip oscillator can be removed by applying high-frequency excitations. Active control by
means of delayed feedback mechanism in controlling either the normal force or quenching with high-frequency dither [60]
have been also applied successfully in both analytical and experimental models (see also [45]). However, active vibration
control of brake squeal and its squeal propensity is still in its infancy stage.
Overall, terms such as for instance irregular [71], fugitive [69,70], capricious [38] or quasi-random [56] are used in
Tables 2–5 which are not directly related to analytical models, to describe disc brake squeal’s high sensitivity to changes in
system parameters. These are characteristics which are also inherent properties of nonlinear and chaotic systems. Studies
on brake squeal noise listed in Tables 1–5 are based on analysis of vibrations in mainly simplified analytical models of
friction oscillators (Table 1) or simplified experimental systems such as pin-on-disc/plate, beam-on-disc and laboratory
brakes (Tables 2–5). With a few exceptions such as [22–24], these studies do not consider friction-induced vibrations from
960 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

the perspective of deterministic chaos. As shown in Table 1, like many other studies of friction-induced instabilities, Shin
et al. [23,24] only examined analytical 2-dof friction systems in the context of chaos. While Hinrichs et al. [22] used
nonlinear analysis tools (such as phase space plots and Lyapunov exponents) to analyse the results of simulation and
experiments, there are a number of limitations: (a) the model is only a pin-on-disc system; (b) only vibrations are
considered although the manifestation of squeal is acoustic; and (c) bifurcations are illustrated but it is not clear whether
other routes to chaos are possible for brake squeal especially for squeal in a full brake system.
To the authors’ knowledge no experiment is known where deterministic chaos and its route have been identified in the
squeal noise of a full brake system. The objective of this paper is, therefore, to close the gap between chaotic phenomena
found in simplified analytical models or simplified experimental systems both investigated by means of structural
vibrations and the true dynamics of squeal in terms of acoustics in a full commercial disc brake system. A second objective
is to explore a different approach to conventional FE analysis in predicting brake squeal propensity by using the theory of
nonlinear dynamics. Firstly, the measures used in nonlinear dynamics are introduced and applied to five analytical
functions in order to illustrate their meanings. Then, a sinusoidally driven 1-dof friction oscillator is examined using a
range of these measures and the different regimes of oscillations are classified based on the attractors revealed. With the
insights obtained, nonlinear recurrence quantification analysis is applied for the first time to squeal data (sound pressure)
of a full brake system obtained in an industrial brake dynamometer. The development of squeal is quantified using
nonlinear dynamics theory and possible routes to chaos and hence squeal are identified.

2. Basic measures

Many quantitative measures are used for analysing nonlinear dynamics [81,82]. Most of these are indicators and
provide an estimate of the predictability of the dynamic system based on the reconstructed system’s trajectories in phase
space. Some of the measures used in this study are listed in Table 6. For the analysis of time series experimental data,
conventional delay embedding techniques according to Takens embedding theorems are used [83]. The embedding delay is
estimated by the first minimum of the averaged auto-mutual information. By applying the embedding delay to the time
series and by using the false nearest neighbour algorithm [84], the minimum embedding dimension is calculated, so that
the solution trajectories in the phase space are unfolded from the attractor. Parameters such as the embedding delay and
embedding dimension used to embed the time series to reconstruct the solution trajectories in the phase space are called

Table 6
Brief explanation of dynamic invariants [81] and recurrence quantification analysis (RQA) measures [82,90].

Measure Description

Embedding delay t  Time delay applied to time series data for the reconstruction of the phase space of a dynamical system
 Estimated with the algorithm of the averaged auto-mutual information from its first minimum
Embedding dimension m  Gives the dimension of phase space
 Estimated by false nearest neighbour algorithm

Correlation dimension D2  Derived from slope of correlation sum

 Lower bound topological/fractal dimension

Correlation entropies h2  System complexity

 Affinity of generating information

K-entropies K2, K1  As lower bound related to sum of positive Lyapunov exponents

Maximal Lyapunov exponent Lmax  Exponential divergence/convergence of nearby points in phase space

Recurrence plot RP  Evaluation of recurrent incidents

 Non-stationary, periodic, chaotic states
 Recurrence quantification analysis (RQA) based on RP

Recurrence rate RR  Quantifies recurrent states

 Random behaviour has low value
 Corresponds to definition of correlation sum

L-entropy ENTR  Average amount: information of diagonals

 Decreasing: less diagonals

Average line length LL  Average diagonal line length

 Decreases: periods get in average shorter

Determinism DET  Measures predictability

 Relates to diagonal lines (non-tangential motion)

Laminarity LAM  States which remain fairly constant

 Relates to horizontal lines, thus tangential motion

Divergence DIV  Decreases: periodic structures vanish

 Related to a lower limit to the sum of positive Lyapunov exponents
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 961

Fig. 1. Schematic of a friction oscillator [19].

embedding parameters [85]. One of the classic invariants as used in a metric analysis [86] to characterise a system is the
correlation dimension. The correlation dimension is an estimate of the attractor’s dimension which indicates fractality for
non-integer values [87] and is calculated from a scaling region of the slope of the correlations sum [82]. Another classic
invariant measure, the Lyapunov exponents, represents the rate of separation of the solution trajectories [88]. A positive
Lyapunov exponent indicates exponential divergence of solution trajectories, hence sensitivity to initial conditions, which
results in the loss of long-term predictability and fractal structures for the attractor [86]. Closely related are the K-entropy
estimates, for which the sum of positive Lyapunov exponents defines an upper bound [89]. A positive entropy estimate,
therefore, indicates chaos [82]. In time series analysis of nonlinear systems, recurrence plots are used to visualise
the return of the solution trajectory in the phase space to a neighbourhood of a point of a previous state. The size of the
neighbourhood is determined by a prescribed value of the parameter e, hence this neighbourhood is referred to as the
e- neighbourhood. The recurrent states are expressed in a binary matrix with the value of 1 indicating the existence of a
recurrent state and the value of 0 indicating the state is not recurrent [91]. Measures used in recurrence quantification
analysis [85] listed in Table 6 are based on the statistics of recurrence plots and are used to quantify the dynamical
behaviour of the system. The recurrence rate RR measures the percentage of recurrent states in a recurrence plot (laminar
and deterministic states) [85] and corresponds to the correlation sum [82]. ENTR is the Shannon entropy of the probability
distribution of all diagonal lines formed by ones in the binary matrix, whenever the distance between two points is found
to be in the neighbourhood of a reference point. The shorter a diagonal line, the less information about the system’s
trajectory is included and hence the lower the value of entropy, indicating randomness. Determinism is the percentage
recurrence rate for the diagonal lines in the recurrence plot and laminarity is percentage recurrence rate for the horizontal
and/or vertical lines. The averaged line length LL is a measure to describe the evolution of the system states and is
calculated from all the diagonal lines in the recurrence plot. A high value of the averaged line length implies long and stable
periods while a low value of the averaged line length generally implies unstable or no periods. Divergence is the reciprocal
value of the longest diagonal line length, corresponding to the longest period in the recurrence plot, and is related to the
Kolmogorov–Sinai entropies (K1, K2), hence to the sum of the positive Lyapunov exponents [92]. A high value of divergence
generally implies loss of stable periodic orbits. Consequently, an increasing divergence together with a decreasing average
line length is an indication of a route to instability. Applications of these measures to analytical functions such as
sinusoidal or logistic functions and the effect of noise can be found in [93,94]. The nonlinear dynamics of a forced 1-dof
friction oscillator (Fig. 1) is chosen here as an example to illustrate the application of some of the measures (e.g., recurrence
plots) given in Table 6 and to serve as a benchmark for comparisons with the analysis of real brake data in terms of the
routes to chaos. The governing equations and detailed procedure in applying some of the measures in Table 6 are given
in A.1. Results in Appendix A reveal three regimes: a limit cycle (one frequency); a weakly chaotic regime with a toroidal
attractor (multiple distinct frequencies); and a broadband chaotic regime with the identification of the Ruelle–Takens–
Newhouse route to chaos.

3. Experimental set-up

Measures described in Section 2 are applied to sound pressure signals obtained from a full front braking system of a
midsize sedan with two-pistons and a floating fist calliper assembly. The brake system was tested in a computer-controlled
industrial noise dynamometer of shaft type (LINK) [95] as part of a wider study described in more details by Moore et al.
[96]. The calliper was made of cast aluminium, the bracket of spheroidal graphite cast iron, and the piston and the rotor of
grey cast iron. The pads used in the production version of the brake system included chamfers and slots in the lining as
well as a noise shim bonded to the back-plate of the pad. The pads used for the test did not have chamfers and any vertical
slot, but noise shims. Noise shims are typically a metal-rubber laminate bonded to the pad back-plate and are therefore
located between the piston or calliper housing and the pad’s back-plate. Shims are very effective in improving the noise
performance of a brake system [97]. Fig. 2(a) shows the computer-controlled chamber and the control unit. Fig. 2(b) shows
962 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

Fig. 2. Set-up of brake system on shaft dynamometer: (a) dynamometer chamber with control stand; (b) disc brake assembly fixed on hub and
cardan joint; (c) brake rotor and microphone 0.5 m apart; (d) spring-damper (shock absorber) assembly with steering knuckle of a car’s left
front corner.

the inside of the dynamometer chamber, the brake disc bolted to a wheel adapter, drive plate and cardan drive shaft with
the calliper housing flapped back for presentation purpose. Fig. 2(c) shows the microphone located at 0.5 m above the
brake disc [98]. Fig. 2(d) shows the full set-up of a front braking system of a car’s left corner, with steering control arms,
bushes, shock absorber and spring, to provide boundary conditions representative of those on a car, with a cooling vent at
the lower left corner. Pressure, velocity, initial humidity and temperature, were set to controlled values. After calibration
and bedding-in brake applications, the pre-conditioned brake system was tested. In a sequence of 1669 stops, including
forward/backward and warm/cold sections, squeal noise was recorded. The warm test matrix was conducted according to
specification SAE J2521 [98], while the cold section was adapted to NVH customer specifications. The microphone’s
sampling rate was set to 44.2 kHz. Brake stops were ranked according to their highest sound pressure peak level. The top
20 stops were examined by analysing their time series, distance plots and spectrograms. Squeal phenomena found were
quite different and time series with short squeaking events were discarded. In the end, eight events of 1669 stops with
relatively long and constant squeals were found to exhibit similar behaviour. All of these stops were forward drag-stops in
the cold section with a brake temperature near 0 1C and a rather constant velocity of around 40.5–41.0 rev/min
ð  4:8 km=hÞ. Fig. 3 displays the initial brake temperature and humidity for the warm and cold sections recorded for each
brake stop. It can be seen that the chamber temperature was kept rather constant by the computer control, but that,
according to test specifications, the initial brake temperature varied significantly with stops. The humidity was not
measured properly in the cold section due to a malfunctioning of the sensors at temperatures below 0 1C and was
henceforth removed. From the sound files recorded using the microphone, only two examples are presented, each for a
different case their position in the test matrix indicated in Fig. 3 by A and B. Case A is often observed in the eight brake
stops and depicts the formation from a fixed point, over a limit cycle with tonal squeal frequency, to an unstable torus-like
structure. This route to aperiodicity is typically observed and detectable by means of recurrence quantification measures.
Case B provides evidence that not only the route to instability is relevant to brake squeal, but that the formation of a
strange attractor with a more pronounced chaotic regime is possible. The route to this attractor is observed either over a
torus or immediately over a limit cycle. Normalised brake-line pressure, rotational speed, m and rotor temperature
recorded at a sampling rate of 50 Hz for the two sample brake stops are shown in Fig. 4. These parameters are considered to
be highly influential for brake squeal and may be treated as bifurcation parameters. The rotational release speed for the
stops was about 4.8 km/h ð  40:5 rev=minÞ. The mean brake-line pressure was around 8 bar, the initial rotor temperature
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 963

Fig. 3. Operating conditions recorded over the whole test matrix.

Fig. 4. Time traces of (a) brake-line pressure; (b) rotational speed; (c) friction coefficient; (d) rotor temperature for case A ð  Þ and case B (—).

was between 276 and 278 K and normalised m was on average 0.56 (0.378 not normalised) in case A and 0.60 in case B
(0.416 not normalised). In most sound files investigated the value of divergence started to peak with decreasing velocity,
hence for the friction model described in Section 2, the belt’s velocity was the chosen bifurcation parameter. RQA is applied
to the whole time series using a window of size 99 and a shift of one to allow the capture of fine structures. A larger
window would smear short and important signals making it difficult to detect either chaos or its onset. For the RQA,
embedding parameters t ¼ 1 and m= 5 were chosen [99]. The value of e- neighbourhood was calculated as the ratio of the
calibration signals standard deviation to the squeal’s standard deviation and has been found to be 0.067 (case A) and 0.056
(case B), which was smaller than 10 percent of the maximum phase space diameter (case A: 0.071, case B: 0.062)
[82,94,90].

4. Analysis of brake squeal time series data—case A

At first the recurrence quantification analysis (RQA) measures are applied to the time signal. Fig. 5(a) displays the time
series of the sound pressure signal for the whole time series together with three measures: determinism/recurrence rate
964 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

/
/
/
/

/
/

Fig. 5. Case A: (a) sound pressure signal, DET/RR, LAM/DET and DIV; (b) boxed area magnified; (c) spectrogram; (d) average line length (LL) and entropy
of diagonal lines (ENTR).

(DET/RR), laminarity/determinism (LAM/DET) and divergence (DIV). A section of the whole time series was extracted to
form case A (i.e., 0.63–1.31 s) displayed in Fig. 5(b). Fig. 5(c) displays the spectrogram obtained for the whole time series in
Fig. 5(a). Signals of additional measures (normalised average line length of diagonal lines (LL) and normalised entropy
(ENTR)) corresponding to Fig. 5(a) are shown in Fig. 5(d). In a second step classic dynamic invariants will be presented.

4.1. Recurrence quantification analysis

Zone I : Fig. 5(a) indicates high LAM compared with DET. Laminar states can be seen in the corresponding RP in Fig. 6(a)
and (b) in which more vertical and horizontal lines are present than isolated diagonal line structures before a transition to
a limit-cycle regime appear. In Fig. 6(a), the segment from 0.34 to 0.45 s shows an auto-regressive process [90] before
contact between the brake pad and rotor is fully established and squeal has not developed yet, which is confirmed by the
spectrogram in Fig. 5(c) and its power spectral density (PSD) in Fig. 7(a). This state corresponds to a stable fixed point
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 965

Fig. 6. RPs case A: (a) auto-regressive pre-squeal regime; (b) Zone I, transition to limit cycle; (c) Zone II, limit cycle/torus regime; (d) Zone III, unstable torus.

which is depicted in Fig. 7(a) in the first column. The system’s evolving trajectory is captured in a state of very little change
(laminar), rather than evolving (as expressed by DET). At the beginning of the brake stop no distinct frequency apart from a
broadband spectrum can be seen (Fig. 7(a)). Three zones are identified in the magnified interval (Fig. 5(b)) with their
corresponding RPs in Fig. 6. Fig. 6(b) depicts the RP of a transition from an unstable fixed point to a limit cycle, as shown in
the phase space plot of Fig. 7(b), and represents zone I. This system behaviour is due to supercritical Andronov-Hopf-
bifurcations [45,11]. The black region in the upper right corner of the RP in Fig. 6(b) corresponds to already limit-cycle
behaviour with a high value of DET. As soon as the amplitudes of the time series increase (Fig. 5(a) and (b)), the
spectrogram becomes richer and distinct frequencies appear (Fig. 5(c)). These frequencies indicate the tonal character of
squeal noise.
Zone II : As soon as zone II is approached, the fundamental frequency and its harmonic (8.475 and 16.95 kHz) appear in
the PSD (Fig. 7(b) and (c)). This indicates the birth of a limit cycle as depicted in Fig. 7(b) leading to (c). This transition to
zone II is accompanied by a drop in LAM/DET, whereas at the same time, the ratio of DET/RR increases slightly (Fig. 5(b)).
This is an effect of the shrinking value of the RR whereby the system’s behaviour becomes more complex. DET, as well as a
low LAM, indicates that a recurrent signal is present which gives unbroken, diagonal lines in the RP in Fig. 6(c). In the limit-
cycle regime, transitions between two periodic-like regimes occur and are marked by a maximum in LAM at 0.79 s
(Fig. 5(b)). These transitions are called period-period transitions and, finally marked by bifurcations, they initiate the
formation of a torus [11], as shown in Fig. 7(d).
Zone III : The divergence DIV, which is the reciprocal value of the maximum diagonal line length in the window of
interest, with a rather constant and low value in zone I and II, starts to increase (Fig. 5(b)). If the value of the maximum line
length approaches zero, periods are getting disrupted, thereby indicating chaos. However, as changes in the DIV simply
show that the diagonal line structures in the RP become more complicated, it is necessary to be careful about drawing
definite conclusions. If other measures do not show behaviour usually encountered in unstable becoming systems, the
torus will only get more complicated, more modes will be involved and the system will merely remain quasi-periodic; for
example, in zone III, at 10.5 kHz, the second harmonic becomes distinguishable (Fig. 7(d)). This corresponds to a so-called
secondary Hopf-bifurcation, which can follow after a limit cycle [11]. Further, it is observed that LAM in zone III increases
slightly and that fluctuations might indicate the instability of the attractor. Also, the value of the LL drops as well as the
ENTR (Fig. 5(d)). The transition to a state of higher energy is indicated by a maximum in LAM/DET after which the DIV
966 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

Fig. 7. Power spectral density (PSD) case A: (a) Zone I, transition to limit cycle; (b) Zone I, forming limit cycle (vortex); (c) Zone II, limit cycle;
(d) Zone III, (chaotic) torus attractors for case A.

increases. In the first part of zone III, a torus is formed (Fig. 7(d)). The steep increase in the measure DET/RR initiates the
upcoming chaotic regime in which the maximum length of the diagonal lines decreases significantly; the DIV becomes
almost 5 percent in comparison to approximately 0.1 percent previously (Fig. 5(b)). The RQA measures in Fig. 5(b) show
that the RR decreases but DET, which indicates the system’s predictability, does not decrease at the same rate. The
spectrogram shows an increase in the frequencies involved (Fig. 5(c)) while the PSD indicates a broadband characteristic
which is slightly lifted up (Fig. 7(d)). Although DET/RR drops abruptly as the system loses its stability, it increases again
when a chaotic regime is approached. LAM becomes lower in comparison with the system’s DET. An important measure,
which must always be considered, is LL (Fig. 5(d)). If only DIV becomes high, without a corresponding decrease of LL, then
chaos could not be detected in the experiments. The PSD in Fig. 7(d) displays more distinct frequencies and is more
broadband in nature but obviously not turbulent, yet. After the DIV returns to a lower value similar to that in zones I and II,
a new cycle is initiated except that, in the audible range, the squeal does not vanish. This is confirmed by one major squeal
frequency and its harmonic in the spectrogram (Fig. 5(c)). Clearly, the phase space dimension of the attractor has grown,
indicating that the torus underlying dynamics (Fig. 7(d)) radiate at a higher sound pressure level than does the limit cycle.

4.2. Dynamic invariants

Table 7 provides an overview of the attractor’s invariant estimates in the three zones. In order to interpret these
attractors, the dynamic invariants D2, K2, K1 and h2, as well as Lmax given in Table 6, have been calculated and are listed in
Table 7 using the TISEAN software package [100]. The invariant measures were calculated for the described zones of the
sound pressure signal firstly without and then with nonlinear filtering [101,102] to reduce the effects of noise in the data.
In case A, the development towards chaos from zones I to III is evident as the D2 estimate increases until it becomes a non-
integer and fractal. It was found that the D2 of zone I approaches zero as a fixed point with the dimension of zero in the
phase space. Reasons for this value being not exactly zero are: experimental data are always noisy despite this data set
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 967

Table 7
Parameters and dynamic invariants.

Case Zone Interval (s) t m D2 K2 K1 h2 Lmax

A I 0.67–0.72 2 2 0.1 0.014 0.0298 0.000  0.070

II 0.72–0.81 1 3 1.0 0.003 0.0099 0.012 0.006
III 0.94–1.02 15 4 2.6 0.478 0.3790 0.174 0.147
B III 7.53–7.62 50 5 2.2 0.42 0.3125 0.15 0.175

Time delay t, embedding dimension m, correlation dimension D2, K—entropies K1, K2, correlation entropy h2, maximal Lyapunov exponent Lmax.

being de-noised with nonlinear filtering; the sequences chosen are finite, here using the interval 0.63–0.72 s; and zone I
represents a transitional regime from a fixed point to a limit cycle. D2 for zone II has an obvious scaling behaviour for many
different embedding dimensions around the integer value of one, indicating a limit cycle and the absence of fractality. In
zone III, D2 is about 2.5 which indicates fractality. Since a non-integer dimension can be due to fractality, which is a
by-product of the stretching and folding process of the attracting sub-set, the attractor in Fig. 7(d) is very likely to lie
somewhat close to a torus dimension of D2 =2 in order to fit into the prescribed phase space dimension. The K-entropies are
non-zero, hence the sum of the positive Lyapunov exponents is greater than zero. Lmax was calculated and if positive, is a
typical sign of chaos since the trajectories no longer exhibit volume-preserving behaviour and diverge. For the point
attractor in zone I, the characteristic Lyapunov exponent is negative while, for the limit cycle in zone II, it is near zero and,
for the unstable torus-like structure in zone III, Lmax is positive.

5. Analysis of brake squeal time series data—case B

Case B illustrates the route to chaos and also the formation of a strange attractor apparently not related to a typical
quasi-periodic torus structure. Fig. 8(a) displays the time series of the sound pressure signal for the whole time series,
together with three different measures: DET/RR, LAM/DET and DIV. A section of the whole time series is extracted to form
case B (i.e., 7.31–7.62 s samples), as displayed in Fig. 8(b). Fig. 8(c) and (d) displays the corresponding spectrogram and
time series of normalised LL and ENTR. Similar to case A, three zones are identified in Fig. 8(b). Zones I and II in case B
behave mostly in a similar fashion to those in case A. Therefore, only zone III is investigated in detail in the following in
order to explore whether a strange attractor exists.

5.1. Recurrence quantification analysis

Zone I : After an epoch of high LAM this measure drops before a sequence of periodic-like behaviour follows (Fig. 8(b)).
Zones I and II are in this case rather short, the limit cycle changes directly into a chaotic regime. In comparison with the
brake stop in case A, the spectrogram in Fig. 8(c) shows many more distinct frequencies. LL and ENTR remain at high and
constant values (Fig. 8(d)).
Zone II : The LAM diminishes and DET/RR increases while the DIV does not show any particular change. Shortly before
zone III begins, LAM reveals a local maximum, indicating a transition from periodicity to chaos. At the same time, DET/RR
shows a minimum but rises afterwards to a slightly higher level. The LL and ENTR based on diagonal lines start oscillating
and then decrease in their overall level.
Zone III : High values of DIV can be observed while the LL decreases. DIV reaches values from 10 percent up to 53 percent
(Fig. 8(b)). The distinct peaks in LAM reveal that while the system is in the chaotic regime, chaos/chaos transitions occur
[99,103]. In Fig. 9(b) and (c) the attractor and the RP for zone III are given. The RP in Fig. 9(c) now shows broken and
irregular lines evidently due to deterministic chaos. A sudden drop of LL and ENTR can be observed (Fig. 8(d)).

5.2. Dynamic invariants

The calculated dynamic invariants are given in Table 7, clearly indicating the presence of chaos. The estimates for the
PSD in zone III is depicted in Fig. 9(a). In comparison to the PSD of case A in (Fig. 7(d)), evidently more frequencies are
involved and the spectrum is more broadband. The correlation dimension D2 is estimated to be 2.2. The attractor in
Fig. 9(b) is plotted in a folded state, whereas in Fig. 10 the unfolded attractor is presented. Fractality is due to the basic
operations stretching and rotation (linear) and folding (nonlinear).

6. Discussions

Real-life brake squeal data presented here indicate that when conditions for chaos are established, the brake system
squeals. Thus, a qualitative definition of the route to chaos in brake squeal can be provided here in terms of RQA, as
illustrated in Table 8. The first row of the matrix is set to be the initial state to which the second row is related to. The third
row expresses the system’s development in relation to the second row, etc. The signs ‘‘ + /  ’’ indicate increasing/decreasing
968 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

/
/
/
/

Fig. 8. Time series for case B: (a) microphone signal, DET/RR, LAM/DET and DIV; (b) magnified interval indicated by dashed square in (a); (c) spectrogram;
(d) entropy (ENTR) and average line length (LL).

values relative to the previous row. Double signs, such as ‘‘ = = ’’, indicate that the system’s behaviour is almost constant,
whereas ‘‘++’’ represents a strong increase. A ‘‘(c)’’ is the placeholder for a constant and ‘‘  (c)’’ means that the measure, as it
decreases, is then almost constant in this segment. The route to chaos described consists of a section with high LAM
followed by a section with high DET and a drop in LAM, and a third section with high DIV, low LL where more frequencies
become involved. However, no bifurcation diagram could be obtained because the control parameters could not be held
constant, even in a computer-controlled dynamometer test rig. Also, it is not sure if the attractors found are codimension-1
attractors or if a combination of several control parameters is necessary to induce bifurcating behaviour. It was observed in
the existing data, that usually a fixed point leads to a limit cycle which develops into an (unstable) torus leading to louder
sound pressure amplitudes. A torus regime could be found quite often although, when the DIV increases, very quickly
strange attractors were established. The formation of an attractor, whether strange or not, is indicated by the RQA with a
local minimum in DET/RR and a local maximum in LAM/DET. As some of these transitions correspond to bifurcations, more
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 969

Fig. 9. Chaotic regime case B: (a) PSD; (b) phase space plot; (c) recurrence plot.

frequencies in the PSD are observed. Further, the spectrogram shows different frequencies at different times and may be
used to locate bifurcations. Bursts of chaos can be seen in Fig. 8(a) and (b), as indicated by sudden increases in the DIV. It is
clear from Fig. 8 that chaos (typified by high DIV) does not always coincide with the highest sound pressure amplitude.
Nevertheless, case A is archetypical since, in most cases, very severe and high squealing amplitudes are observed in the
regimes of high DIV, for example, as depicted in Fig. 5(a) and (b). One type of chaos in brake squeal, zone III, can be defined
as the part of the time series of the sound pressure signal in which high DIV, a medium level of DET, a low RR, low LAM, low
ENTR exist, when initiated by a significant increase then constant LAM/DET, and a decrease in LL. The classical invariants
show that, where the torus structure is formed, the system already behaves in an unpredictable fashion (see K2-, K1-, Lmax-
estimates) even when the PSD is not lifted up significantly relative to the previous limit cycle regime and does not show
many distinct frequencies. The process of stretching, rotation and folding has already started. However, it is important to
note that chaos is not a pre-requisite for squeal. Brake squeal is able to be detected and predicted in some cases by means
970 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

Fig. 10. Attractors zone III: (a) case A; (b) case B.

Table 8
The table shows the trends of the brake system’s dynamics, in terms of RQA measures.

Transitions RR LAM DET LL DIV ENT Squeal # freq.

I (initially) high High High High Low High No Low

I-II    (c) + (c) + (c) = =  (c) Yes +
II-III  =+   ++  Yes ++
III-I ++ ++ + ++  + no 

The signs  ,= ,+ indicate the rate of change relative to a previous regime.

of linear methods, for instance, by the complex eigenvalue analysis. This is due to the centre manifold theorem which
analyses the dynamical system around its linearised equilibrium [104]. Brake squeal can start with an instability indicated
by an Andronov-Hopf-bifurcation [95,45], then it might proceeds in nonlinear conditions [41]. Evidently, chaos is due to
existing nonlinearity which dominates the system behaviour in these regimes and starts with an audible transition from a
fixed point to a limit cycle. However, researchers of disc brake squeal usually focus and discuss only the limit-cycle
phenomenon rather than the route towards instability which is investigated here. It is found that the actual limit-cycle
regime is never purely present for a long time but that transitional states, for example, between the limit-cycle regime and
the torus, are the attractors most often encountered. Furthermore, a regime may not look turbulent, however, the long time
predictability might already be lost (positive Lyanpunov exponent), hence this regime has to be called chaotic. Also,
broadband spectra similar to turbulence often cannot be observed while some frequencies remain quite distinct. This is
consistent with a system behaving in a weakly chaotic (non-turbulent) fashion [11]. Therefore, it seems necessary to
distinguish between weak chaos (no real broadband spectrum), chaos (broadband spectrum, where one Lyapunov
exponent becomes positive), and stronger chaos (hyper-chaos up to turbulence) where the trajectory is diverging in more
than one direction [105].
In case B, distinct frequencies are still visible although the PSD function in Fig. 9(a) clearly shows that the number of
modes involved has significantly increased. For the time interval chosen (7.24–7.81 s), the formation of a torus has not
been observed as the strange attractor is formed directly after the limit-cycle regime. Similarly, in Feeny and Moon [21],
the route to chaos through bifurcations was also not observed in a dry friction 1-dof oscillator. They assumed that the
system becomes unstable in such a very short time that these bifurcations are simply not observable. This is consistent
with the description of the Ruelle–Takens–Newhouse route to chaos [11] and is also confirmed in case B. However, in other
parts of the time same series the formation of the torus was observed. Another possibility is the formation of border
collision bifurcations from a limit cycle directly into chaos which were already evidenced in discontinuous dynamical
systems [46]. The attractor in case B was found and formed whenever the recurrence quantification measures indicated
instability but also vanished after some time. If present for a longer period, it rotated in phase space initiated by (a) varying
operation conditions which trigger a (b) change in embedding parameters. This indicates that the geometric structure
formed by chaos is in itself quite robust (attracting); however, to investigate longer time series of the attractor, a
continuous synchronisation technique of dynamic regimes is necessary [103,94]. Further, even in the chaotic regime,
squeal remains tonal and clearly audible. This indicates that deterministic chaos does not have to sound and act like white
noise. However, in the present brake squeal data, several control parameters (such as temperature, stiffness, friction
coefficient, brake-line pressure and rotor speed) change continuously with time. Even when the attractor in zone III in case
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 971

B is found in the whole section (7.24–7.81 s), it was not possible to enlarge the basis for one calculation by using the whole
section of sample points. Changes in the system’s characteristics make it almost impossible to capture a regime long
enough to enable the extraction of detailed features of the chaotic attractor. Also, the effects of noise need to be considered
in a constantly changing system with unsteady bifurcation parameters as they make capturing a long stationary state very
difficult. Only quasi-stationary regimes are chosen with the help of distance plots, RPs and their quantification measures.
Owing to changing parameters, chaotic attractors cannot be observed with more data points without phase space
synchronisation even if the same mechanism was at work as indicated by the same geometry of the strange attractor.

7. Conclusions

By using traditional invariant measures used for nonlinear dynamics analysis, a sinusoidally driven 1-dof friction
oscillator, with relative velocity as control parameter, has been shown to exemplify the transition from a limit cycle over a
torus to a chaotic attractor. Then, similar quantitative analysis and for the first time recurrence quantification analysis
were applied to brake squeal data obtained by a single microphone in a full brake system tested in an industrial
dynamometer. Despite the poor signal to noise ratio in the brake squeal microphone data and varying parameters, a
number of dynamic regimes of interest have been identified as transiting from a limit cycle to an unstable torus attractor
which is very similar to the behaviour of the single dof friction oscillator. Further, it is the first time that a qualitative
correlation between nonlinear dynamics theory (chaos) and real-life brake squeal data has been established. Preliminary
numerical simulation studies of a pad-on-disc system using sound pressure data show similar chaotic phenomenon to that
observed here [106,107]. Further, the route to chaos confirms the developing character of squeal [55,41] and more routes are
likely. Also, it should be noted that chaos is not the only mechanism for brake squeal generation. However, the results
presented demonstrate that nonlinear dynamics is a viable tool for analysing brake squeal data in order to provide
improved understanding of the mechanism of brake squeal generation. Knowledge of routes to chaos, attractors and their
underlying mechanisms could have potential application for the prediction of brake squeal propensity and its control.
More insight could be gained by exploring further analytical models with dry friction and numerical simulation of a full
brake system with nonlinear time series analysis of the sound pressure. Additional test data of a simplified brake system

Fig. A1. Phase space plots and power spectral densities (PSD) for the 1-dof friction oscillator: (a) limit cycle; (b) torus; (c) chaotic regime.
972 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

and a full brake system in an industrial dynamometer with a high sampling rate, several accelerometers, laser vibrometer
and microphones, and better-controlled parameters (constant pressure, specifically designed velocity regimes), would help
to reveal more routes to chaos and, hence, brake squeal generation.

Acknowledgments

The first author (Oberst) acknowledges the receipt of a University of New South Wales University College Postgraduate
Research Scholarship for the pursuit of this study. Also, the authors would like to thank PBR Pty. Ltd. and especially
Dr. Antti Papinniemi, for the provision with measurement data from an industrial noise dynamometer.

Appendix A

A.1. Example of sinusoidally driven one degree-of-freedom friction oscillator

In this section, the nonlinear dynamics of a single dof friction oscillator, as shown in Fig. 1, and is examined. The
governing equations are taken from [22].
€ þcxðtÞ
mxðtÞ _ þ kxðtÞ ¼ Ff ðvrel Þ þkAcosOt (A.1)

vrel ¼ v0 v, with Ff ¼ mðsgnðvrel ÞÞFn (A.2)

The oscillator is excited by force, F, which acts through a spring of stiffness, k. The rotor is simulated by a moving plane in
the form of a moving rigid surface. The mass, the spring’s stiffness, the damping coefficient, friction and normal force
and the friction coefficient are represented by m, k, c, Ff, Fn and m respectively. With m = 1 kg, c = 0.1 Ns/m and k = 1 N/m,
the frequency, f, and amplitude, A, of the exciting harmonic force, F, are set to f= 0.1 Hz and A= 3 N, respectively. Eq. (A.2) is
important in the sense, that it represents the slip velocity and allows for a slip velocity of zero, back and forth switching
between different systems of differential equations, a sticking and slipping state [108]. The belt’s velocity is chosen as the
bifurcation parameter and is varied from 15 to 10 and then 0.2 m/s to explore the ensuing motion. In the friction model
used, m decreases linearly with increasing relative velocity vrel. The threshold value of the sticking regime is e ¼ 0:01. If vrel
is less than e the friction coefficient assumes its static value ms ¼ 0:7; otherwise it is given by m ¼ ms avrel with mk ¼ 0:1 as a
lower bound. The kinetic friction coefficient becomes active when numerically m o mk . The negative slope of the friction

Fig. A2. Correlation dimension (D2) for attractors: (a) limit cycle; (b) torus; (c) chaotic regime and recurrence plots (RP): (d) limit cycle; (e) torus;
(f) chaotic attractor.
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 973

characteristic is given with a ¼ 0:1. As contact definition, a continuous representation is chosen rather than a locking
formulation: No real sticking with a block velocity of zero is enforced in the sticking regime corresponding to creep-slip/
microslip behaviour. When the relative velocity of pad and disc is less than the threshold value e ¼ 0:01, the velocity
decreases prior to the default static value ms .
The phase space representation calculated from the time series data are shown in Fig. A1 together with the
corresponding power spectrum. Different values v (15, 10 and 0.2 m/s) reveal three regimes: A limit cycle (one frequency);
a weakly chaotic regime with a toroidal attractor (multiple distinct frequencies); and a broadband chaotic regime. This
example shows the Ruelle–Takens–Newhouse route to chaos as also investigated by for instance Hinrichs et al. [22]. The
correlation dimension D2 of the three attractors, for the sake of illustration is presented as the average value of 20
dimensions in Fig. A2. The estimate of the fractal dimension is indicated by the horizontal part of the curves where less
oscillations are present. In Fig. A2(a) scaling around D2 =1 becomes obvious which is indicative of the topological
dimension of a limit cycle. In Fig. A2(b), a scaling around dimension two is visible and in Fig. A2(c) for small e around
6  10  2–4  10  1 a scaling above two is discernible. The high values of D2 at very small length scales ðeÞ and large length
scales are due to the discretization process and the finite length of the time series, respectively [81,82]. In Fig. A2, the RPs of
the three regimes are presented. The periodic state in Fig. A2(d) shows straight lines. The RP of the weak chaotic torus in
Fig. A2(e) shows generally long, straight and broken lines which indicate quasi-periodicity. However, the lines are broken
and indicate unpredictability on another time scale. The chaotic regime in Fig. A2(f) shows different, changing regimes and,
in general, shorter, broken lines which correspond to recurrent, though non-periodic and unpredictable, states.

References

[1] A. Akay, Acoustics of friction, Journal of the Acoustical Society of America 111 (4) (2002) 1525–1548.
[2] H. Abendroth, B. Wernitz, The integrated test concept: dyno-vehicle, performance-noise, SAE Technical Paper Series, 2000-01-2774, 2000.
[3] M. Yang, P. Blaschke, A.-H. Afaneh, A study of disc brake high frequency squeal and disc in-plane/out-of-plane modes. SEA Technical Papers,
2003-01-1621, 2003, pp. 1–8.
[4] N. Hinrichs, M. Oestreich, K. Popp, On the modelling of friction oscillators, Journal of Sound and Vibration 216 (3) (1998) 435–459.
[5] N.M. Kinkaid, O.M. O’Reilly, P. Papadopoulos, Automotive disc brake squeal, Journal of Sound and Vibration 267 (2003) 105–166.
[6] H. Ouyang, W. Nack, Y. Yuan, F. Chen, Numerical analysis of automotive disc brake squeal: a review, International Journal of Vehicle Noise and Vibration
1 (2005) 207–231.
[7] F. Chen, Automotive disk brake squeal: an overview, International Journal of Vehicle Design 51 (1/2) (2009) 39–72.
[8] S. Oberst, J.C.S. Lai, A critical review on brake squeal and its treatment in practice, Internoise 2008, Shanghai, China, 2008.
[9] N. Hoffmann, L. Gaul, Friction induced vibrations of brakes: research fields and activities, SAE Technical Paper Series, 2008-01-2579, 2008, pp. 1–8.
[10] F. Cramer, Chaos and Order: Complex Structure of Living Systems, Wiley VCH Verlagsgesellschaft, Weinheim, New York, 1993.
[11] H.G. Schuster, W. Just, Deterministic Chaos: An Introduction, Wiley VCH Verlagsgesellschaft, Weinheim-New York, 2005.
[12] S.J. Linz, J.C. Sprott, Elementary chaotic flow, Physics Letters A 259 (1999) 240–245.
[13] S.W. Shaw, On the dynamic response of a system with dry friction, Journal of Sound and Vibration 108 (2) (1986) 305–325.
[14] E. Ott, Chaos in Dynamical Systems, Cambridge University Press, GB, 1993.
[15] M. Kunze, T. Küpper, Qualitative bifurcation analysis of a non-smooth friction-oscillator model, Zeitschrift für Angewandte Mathematik und Physik 48
(1997) 87–101.
[16] M. Kunze, Non-smooth dynamical systems, Univeritát Kóln, Habilitationsschrift, 1998.
[17] J. Awrejcewicz, P. Olejnik, Analysis of dynamic systems with various friction laws, Applied Mechanics Reviews 58 (November) (2005) 389–411.
[18] B. Erickson, B. Birnir, D. Lavallée, A model for aperiodicity in earthquakes, Nonlinear Processes Geophysics 15 (2008) 1–12.
[19] K. Popp, P. Stelter, Stick–slip vibrations and chaos, Philosophical Transactions: Physical Sciences and Engineering 332 (1624) (1990) 89–105.
[20] B.F. Feeny, F.C. Moon, Bifurcation sequences of a coulomb friction oscillator, Nonlinear Dynamics 4 (1992) 25–37.
[21] B. Feeny, F.C. Moon, Chaos in a forced dry-friction oscillator: experiments and numerical modelling, Journal of Sound and Vibration 170 (February)
(1994) 303–323.
[22] N. Hinrichs, M. Oestreich, K. Popp, Dynamics of oscillators with impact and friction, Chaos, Solitons & Fractals 8 (4) (1997) 535–558.
[23] K. Shin, M.J. Brennan, J.-E. Oh, C.J. Harris, Analysis of disc brake noise using a two-degree-of-freedom model, Journal of Sound and Vibration 254 (5)
(2002) 837–848.
[24] K. Shin, J.-E. Oh, M.J. Brennan, Nonlinear analysis of friction induced vibrations of a two-degree of freedom model for disc brake squeal, JSME
International Journal Series C 45 (2002) 426–432.
[25] N. Hoffmann, M. Fischer, R. Allgaier, L. Gaul, A minimal model for studying properties of the mode-coupling type instability in friction induced
oscillations, Mechanics Research Communications 29 (2002) 197–205.
[26] O. Giannini, A. Akay, F. Massi, Experimental analysis of brake squeal noise on a laboratory brake setup, Journal of Sound and Vibration 292 (2006) 1–20.
[27] A. Buck, Simulation von Bremsenquietschen (Brake Squeal), PhD Thesis, Lehrstuhl für Baumechanik, Technische Universität München, 2008.
[28] G.G. Adams, Self-excited oscillations of two elastic half-spaces sliding with a constant coefficient of friction, ASME Journal of Applied Mechanics
62 (1995) 867–872.
[29] G.G. Adams, Steady sliding of two elastic half-spaces with friction reduction due to interface stick–slip, ASME Journal of Applied Mechanics 65 (1998)
470–475.
[30] L. Baillet, S. D’Errico, Y. Berthier, Influence of sliding contact local dynamics on macroscopic friction coefficient variation, Revue europeenne des
elements finis 10 (2005) 1–16.
[31] H.R. Mills, Brake squeak, Technical Report, The Institution of Automobile Engineers, Research Report, 9000 B and 9162 B, 1938.
[32] J.J. Thomson, Using fast vibrations to quench friction-induced oscillations, Journal of Sound and Vibration 228 (5) (1999) 1079–1102.
[33] A. Bajer, V. Belsky, S. Kung, The influence of friction-induced damping and nonlinear effects on brake squeal analysis, SAE Technical Paper,
2004-01-2794, 2004, pp. 1–9.
[34] G.X. Chen, Z.R. Zhou, P. Kapsa, L. Vincent, Experimental investigation into squeal under reciprocating sliding, Tribology International 36 (2003) 961–971.
[35] D.M. Beloiu, R.A. Ibrahim, Brake squeal as dynamic instability: an experimental investigation, Structural Control and Health Monitoring 13 (2006)
277–300.
[36] G.X. Chen, Z.R. Zhou, Time-frequency analysis of friction-induced vibration under reciprocating sliding conditions, Wear 262 (2007) 1–10.
[37] D. Stanef, A. Papinniemi, J. Zhao, From protoype to production—the practical nature of brake squeal noise, SAE Technical Paper Series, 2006-01-3217,
2006, pp. 1–10.
[38] P. Duffour, J. Woodhouse, Instability with frictional point contact. Part 3: experimental tests, Journal of Sound and Vibration 304 (2007) 186–200.
974 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975

[39] M.T. Bengisu, A. Akay, Stability of friction-induced vibrations in multi-degree-of-freedom systems, Journal of Sound and Vibration 171 (4) (1994)
557–570.
[40] R.A. AbuBaker, H. Ouyang, A prediction methodology of disk brake squeal using complex eigenvalue analysis, International Journal of Vehicle Design
46 (2008) 416–435.
[41] F. Massi, L. Baillet, O. Giannini, A. Sestieri, Brake squeal: linear and nonlinear numerical approaches, Mechanical Systems and Signal Processing 21
(2007) 2374–2393.
[42] J. Jerrelind, A. Stenson, Nonlinear dynamics of parts in engineering systems, Chaos, Solitons & Fractals 11 (2000) 2413–2428.
[43] O. Giannini, A. Sestrieri, Predictive model of squeal noise occurring on a laboratory brake, Journal of Sound and Vibration 296 (2006) 583–601.
[44] G.X. Chen, Q.Y. Liu, X.S. Jin, Z.R. Zhou, Stability analysis of a squealing vibration model with time delay, Journal of Sound and Vibration 311 (2008)
516–536.
[45] K. Popp, Modelling and control of friction-induced vibrations, Mathematical and Computer Modelling of Dynamical Systems 11 (3) (2005) 345–369.
[46] P. Kowalczyk, Robust chaos and border-collision bifurcations in non-invertible piecewise-linear maps, Nonlinearity 18 (2005) 485–504.
[47] A. Tuchinda, N.P. Hoffmann, D.J. Ewins, W. Keiper, Mode lock-in characteristics and instability study of the pin-on-disc system, Proceedings of the
International Modal Analysis Conference, IMAC XX, Kissimee, vol. 1, 2001, pp. 71–77.
[48] N. Coudeyras, J.-J. Sinou, Nacivet, A new treatment of predicting the self-excited vibrations of nonlinear systems with frictional interfaces: the
constrained harmonic balance method, with application to disc brake squeal, Journal of Sound and Vibration 319 (2009) 1175–1199.
[49] N. Coudeyras, S. Nacivet, J.-J. Sinou, Periodic and quasi-periodic solutions for multi-instabilities involved in brake squeal, Journal of Sound and
Vibration 328 (2009) 520–540.
[50] G.X. Chen, Z.R. Zhou, A self-excited vibration model based on special elastic vibration mode of friction systems and time delays between the normal
and friction forces: a new mechanism for squealing noise, Wear 262 (2007) 1123–1239.
[51] N. Hoffmann, L. Gaul, A sufficient criterion for the onset of sprag-slip oscillations, Archive of Applied Mechanics 73 (2004) 650–660 10.1007/S00419-
033-0315-4.
[52] H. Ouyang, J.E. Mottershead, M.P. Cartmell, D.J. Brookfield, Friction-induced vibraion of an elastic slider on a vibrating disc, International Journal of
Mechanical Science 41 (1999) 325–336.
[53] N. Hoffmann, L. Gaul, Effects of damping on mode-coupling instability in friction induced oscillations, Zeitschrift für Angewandte Mathematik und
Mechanik 83 (8) (2003) 524–534.
[54] O. Giannini, A. Sestieri, F. Massi, A. Akay, Experimental investigation and modeling of brake squeal using simplified test rigs, SAE Technical Paper
Series, 2007-01-3963, 2007, pp. 189–200.
[55] F. Massi, O. Giannini, L. Baillet, Brake squeal as dynamic instability: an experimental investigation, Journal of the Acoustical Society of America 120 (3)
(2006) 1388–1398.
[56] F. Massi, F. Giannini, Effect of damping on the propensity of squeal instability: an experimental investigation, Journal of Acoustical Society of America
123 (2008) 2017–2023.
[57] B. Hervé, J.-J. Sinou, H. Mahé, L. Jézéquell, Analysis of squeal noise and mode coupling instabilities including damping and gyroscopic effects,
European Journal of Mechanics A/Solids 27 (2008) 141–160.
[58] B. Hervé, J.J. Sinou, H. Mahé, L. Jézéquell, Extension of the destabilization paradox to limit cycle amplitudes for a nonlinear self-excited system
subject to gyroscopic and circulatory actions, Journal of Sound and Vibration 323 (2009) 944–973.
[59] B.F. Feeny, F.C. Moon, Quenching stick–slip chaos with dither, Journal of Sound and Vibration 237 (1) (2000) 173–180.
[60] F. Chen, A. Wang, C.A. Tan, Dynamic system instability suppression with dither technique, Proceedings of the 22nd SAE Brake Colloquium, Anaheim, CA,
USA, 2004.
[61] M. Paliwal, A. Mahajan, T. Don, J. Chu, P. Filip, Noise and vibration analysis of a disc-brake system using a stick–slip friction model involving coupling
stiffness, Journal of Sound and Vibration 282 (2002) 1273–1284.
[62] H. Hetzler, D. Schwarzer, W. Seemann, Analytical investigation of steady-state stability and hopf-bifurcations occurring in sliding friction oscillators
with application to low-frequency disc brake noise, Communications in Nonlinear Science and Numerical Simulations 12 (2007) 83–99.
[63] H. Ouyang, Assignment of complex eigenvalues to second-order damped asymmetric systems through state-feedback, NOVEM2009, 5–8 April,
Oxford, England, 2009.
[64] H. Ouyang, L. Baeza, S. Hu, A receptance-based method for predicting latent roots and critical points in friction-induced vibration problems of
asymmetric systems, Journal of Sound and Vibration 321 (2009) 1058–1068.
[65] F.P. Bowden, L. Leben, Nature of sliding and the analysis of friction, Nature 3572 (1938) 691–692.
[66] F.P. Bowden, D. Tabor, Mechanism of metallic friction, Nature 3798 (1942) 197–199.
[67] S.W.E. Earles, G.B. Soar, Squeal noise in disc brakes, in: Vibration and Noise in Motor Vehicles, Institution of Mechanical Engineers, C 101/71, 1971, pp. 61–69.
[68] S.W.E. Earles, A mechanism of disc-brake squeal, SEA Technical Papers, 770181, 1978, pp. 1–6.
[69] S.W.E. Earles, M.N.M. Badi, Oscillatory instabilities generated in a double-pin and disc undamped system: a mechanism of disc-brake squeal,
Proceedings of the Institution of Mechanical Engineers, Part C: Mechanical Engineering Science 198 (4) (1984) 43–49.
[70] S.W.E. Earles, P.W. Chamber, Disc brake squeal noise generation: predicting its dependency on system parameters including damping, International
Journal of Vehicle Design 8 (4/5/6) (1987) 538–552.
[71] W.W. Tworzydlo, O.N. Hamzeh, W. Zaton, T.J. Judek, Friction-induced oscillations of a pin-on-disk slider: analytical and experimental studies, Wear
236 (1999) 9–23.
[72] P. Duffour, J. Woodhouse, Instability of systems with a frictional point contact. Part 1: basic modelling, Journal of Sound and Vibration 271 (2004)
365–390.
[73] P. Duffour, J. Woodhouse, Instability of systems with a frictional point contact. Part 2: model extensions, Journal of Sound and Vibration 271 (2004)
301–410.
[74] T. Butlin, J. Woodhouse, Sensitivity of friction-induced vibration in idealized systems, Journal of Sound and Vibration 319 (2009) 182–198.
[75] J.H. Tarter, Disc brake squeal, SAE Technical Paper, 830530, 1983, pp. 1–8.
[76] M. Suganami, M. Nakai, M. Yokoi, T. Miyahara, H. Matsui, I. Yamazaki, Disk brake squeal by a model of a rotating disk and a beam, Proceedings of the
First International Symposium on Impact and Friction of Solids, Structures and Intelligent Machines, Congress Center, Ottawa, Canada, 1998.
[77] A. Akay, J.A. Wickert, Z. Xu, Investigation of mode lock-in and friction interface, Technical Report, Final Report, Department of mechanical
engineering, Carnegie Mellon University, 2000.
[78] O. Giannini, F. Massi, Characterization of the high-frequency squeal on a laboratory brake setup, Journal of Sound and Vibration 310 (2008) 394–408.
[79] F. Massi, Dynamic and Tribological Analysis of Brake Squeal, PhD Thesis, University of Rome ‘‘la Sapieza’’, Italy, 2006.
[80] F. Massi, Y. Berthier, L. Baillet, Contact surface topography and system dynamics of brake squeal, Wear 265 (2008) 1784–1792.
[81] J.C. Sprott, Chaos and Time-series Analysis, Oxford University Press, Oxford, New York, 2003.
[82] H. Kantz, T. Schreiber, Nonlinear Time Series Analysis, Cambridge University Press, Cambridge, 2004.
[83] J. Stark, D.S. Broomhead, M.E. Daview, J. Huke, Takens embedding theorems for forced and stochastic systems, Nonlinear Analysis, Theory, Methods &
Applications 30 (8) (1997) 5303–5314.
[84] J.M. Martinerie, A.M. Albano, A.I. Mees, P.E. Rapp, Mutual information strange attractors, and the optimal estimation of dimension, Physical Review A
45 (10) (1992) 7058–7064.
[85] N. Marwan, J. Kurths, P. Saparin, Generalised recurrence plot analysis for spatial data, Physics Letters A 360 (2007) 545–551.
[86] R. Gilmore, M. Lefranc, Topology Analysis of Chaos, Wiley VCH Verlagsgesellschaft, 2002.
[87] P. Grassberger, I. Procaccia, Characterization of strange attractors, Physical Review Letters 50 (5) (1983) 346–349.
 S. Oberst, J.C.S. Lai / Journal of Sound and Vibration 330 (2011) 955–975 975

[88] F.C. Moon, Chaotic and Fractal Dynamics, John Wiley & Sons, New York, Chichester, Brisbane, Toronto, Singapore, 1992.
[89] Ya.B. Pesin, Characteristic Lyapunov exponents and smooth ergodic theory, Russian Mathematical Surveys 32 (4) (1977) 55–114.
[90] N. Marwan, M. Carmen Romano, M. Thiel, J. Kurths, Recurrence plots for the analysis of complex systems, Physics Reports 438 (2007) 237–329.
[91] J.-P. Eckmann, D. Ruelle, Ergodic theory of chaos and strange attractors, The American Physical Society 57 (3) (1985) 617–656.
[92] M.C. Romano Blasco, Synchronization Analysis by Means of Recurrences in Phase Space, PhD Thesis, Mathematisch-Naturwissenschaftliche
Fakultät/Universität Potsdam, 2004.
[93] N. Marwan, M.H. Trauth, M. Vuille, J. Kurths, Comparing modern and pleistocene ENSO-like influences in NW Argentina using nonlinear time series
analysis methods, Climate Dynamics 21 (2003) 317–326.
[94] M. Thiel, Exploiting Naturally Occurring Analogues, PhD Thesis, Universität Potsdam, Mathematisch-Naturwissenschaftliche Fakultät, 2004.
[95] F. Chen, F. Tan, C. Chen, C.A. Tan, R.L. Quaglia, Disc brake squeal: mechanism, analysis, evaluation and reduction/prevention, SAE-Society of
Automotive Engineers, 2006.
[96] S. Moore, J.C.S. Lai, S. Oberst, A. Papinniemi, Z. Hamdi, D. Stanef, Design of experiments in brake squeal, Internoise 2008, Shanghai, China, 2008.
[97] J. Flint, J. Hald, Traveling waves in squealing disc brakes measured with acoustic holography. SEA Technical Papers, 2003-01-3319, 2003, pp. 1–8.
[98] Surface vehicle recommended practice, disc and drum brake dynamometer squeal noise matrix, Technical Report, SAE J2521, 2006.
[99] L.L. Trulla, A. Giuliani, J.P. Zbilut, C.L. Webber Jr., Recurrence quantification analysis of the logistic equation with transients, Physical Letters A 223
(4) (1996) 255–260.
[100] R. Hegger, H. Kantz, T. Schreiber, Practical implementation of nonlinear time series methods: the [small-caps TISEAN] package, Chaos: An
Interdisciplinary Journal of Nonlinear Science 9 (2) (1999) 413–435.
[101] H. Kantz, T. Schreiber, I. Hoffmann, T. Buzug, G. Pfister, L.G. Flepp, J. Simonet, R. Badii, E. Brun, Nonlinear noise reduction: a case study on
experimental data, Physical Review E 48 (2) (1993) 1529–1538.
[102] P. Grassberger, R. Hegger, H. Kantz, C. Schaffrath, T. Schreiber, On noise reduction methods for chaotic data, Chaos: An Interdisciplinary Journal of
Nonlinear Science 3 (2) (1993) 127–141.
[103] N. Marwan, N. Wessel, A. Meyerfeldt, A. Schirdewan, J. Kurths, Recurrence plot based measures of complexity and its application to heart rate
variability data, Physical Review E 2 (026702) (2002) 1–8.
[104] J. Guckenheimer, P. Holmes, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields, vol. 42, Springer-Verlag, Berlin, New York,
1997.
[105] H. Kantz, A robust method to estimate the maximal Lyapunov exponent of a time series, Physical Letters A 185 (1) (1994) 77–87.
[106] S. Oberst, J.C.S. Lai, Numerical study of friction-induced pad-mode instability in disc brake squeal, 20th International Congress on Acoustics
(ICA 2010), 23–27 August, Sydney, Australia, 2010.
[107] S. Oberst, J.C.S. Lai, Acoustic radiation of friction-induced pad-mode instability in disc brake squeal, 20th International Congress on Acoustics
(ICA 2010), 23–27 August, Sydney, Australia, 2010.
[108] R.I. Leine, D.H. Van Campen, B.L. Van De Vrande, Bifurcations in nonlinear discontinuous systems, Nonlinear Dynamics 23 (2000) 105–164.