Journal of Sound and Vibration 333 (2014) 1888–1898

Contents lists available at ScienceDirect

Journal of Sound and Vibration

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

Structural optimization of an asymmetric automotive

brake disc with cooling channels to avoid squeal
Andreas Wagner n, Gottfried Spelsberg-Korspeter, Peter Hagedorn
fnb, Dynamics and Vibrations Group, Graduate School CE, TU Darmstadt, Dolivostr. 15, 64293 Darmstadt, Germany

a r t i c l e in f o abstract

Article history: Brake squeal is still a major issue in the automotive industry due to comfort complaints of
Received 6 March 2013 passengers and resulting high warranty costs. Many measures to avoid squeal have been
Received in revised form discussed in the engineering community reaching from purely passive measures like the
19 November 2013
increase of damping, e.g. by the application of shims, to the active or semiactive suppression of
Accepted 23 November 2013
squeal. While active measures can be effective but are elaborate and therefore more expensive,
Handling Editor: H. Ouyang
Available online 17 December 2013 passive measure are less complex in most cases. This leads to the necessity to develop passive,
economic and robust measures to avoid squeal. Asymmetry of the brake rotor has been
proposed to achieve this goal and the resulting split of all double eigenfrequencies of the brake
rotor has lately been shown to stabilize the system.
Thus, a structural optimization of an automotive brake disc with cooling channels is
presented in this paper with the objective to split all eigenfrequencies of the brake rotor in a
certain frequency range by introducing asymmetry to the cooling channels. Constraints of the
optimization are balance constraints, to guarantee a balanced operation for all rotor speeds, and
minimal and maximal distance constraints of the cooling ribs, due to cooling and material
strength requirements. First, a modeling approach of the brake disc with cooling channels is
shortly presented which helps to avoid remeshing during the structural optimization. The
introduced optimization problem is known to be highly nonlinear, nonconvex and with many
local optima to be expected. Therefore, two approaches for the solution of the problem are
chosen. The first, a deterministic one, is a Sequential Quadratic Programming (SQP) approach
efficiently targeting local optima. In order to increase the possibility to find the global optimum,
a set of randomly distributed starting configurations is chosen, leading to satisfying results. The
other, a heuristic approach, uses a Genetic Algorithm (GA) directly aiming for the global
optimum. The GA also delivers very satisfying results, nevertheless, the best solution has been
found with the SQP approach. In order to validate the basic idea that a defined separation of
eigenfrequencies helps to avoid squeal, modal analysis and squeal tests have been performed
with a simplified disc with radial holes. The conducted experiments strongly support the
theoretical findings and demonstrate the superior squeal behavior of the optimized disc.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction

For many decades, design engineers all over the world have been confronted with the objective to find measures against
brake squeal. Brake squeal is not only a major issue in the automotive industry where it leads to high warranty costs due to

n
Corresponding author. Tel.: + 49 6151 16 2879; fax: + 49 6151 16 4479.
E-mail addresses: wagner@dyn.tu-darmstadt.de (A. Wagner), speko@dyn.tu-darmstadt.de (G. Spelsberg-Korspeter),
hagedorn@dyn.tu-darmstadt.de (P. Hagedorn).

0022-460X/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jsv.2013.11.035
 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898 1889

passenger's comfort problems, which can easily reach more than 100 million Dollar per year [1]. But it can also lead to
serious safety problems in bicycles or motorcycles with lightweight rims [2,3]. The engineering community agrees that the
root cause for brake squeal are self-excited vibrations originating from the frictional contact between brake pad and brake
disc. An overview about brake squeal in general can e.g. be found in [4], a more general overview over friction induced
vibrations in [5]. Currently, in the automotive industry, mainly two analysis approaches are taken to assess the susceptibility
of a brake system to squeal: The so-called complex eigenvalue analysis, reducing the analysis of the stability problem of the
rotating rotor in frictional contact to an eigenvalue problem, and the transient analysis based on the direct time integration
of the underlying equations of motion [6]. Nevertheless, brake squeal and the affinity of brakes to squeal are affected by
many different influences, reaching from wear and operating conditions to complicated tribological behavior and
temperature dependence of the brake pads, making the simulation of squeal difficult and prone to errors. The influence
of the special tribology between brake disc and brake pad is discussed e.g. in [7]. However, there can also be found many
minimal models for brake squeal describing the onset of squeal as an instability resulting from nonconservative friction
forces with a constant coefficient of friction, leading to circulatory terms in the stiffness matrix [8,9]. In order to be able to
predict the amplitude of the limit cycle oscillations during squeal, nonlinear effects have to be taken into account [10,11].
Partially resulting from simulations, partially from experiments, many countermeasures are proposed in the engineering
community to avoid squealing. They reach from the introduction of damping into the brake system, by choosing pad or rotor
materials or the application of shims, to the active or semiactive suppression of squeal [12,13]. On the one hand, the active
squeal countermeasures can be very effective against squeal but are elaborate to tune and expensive. On the other hand, the
passive measures are not generally able to avoid squeal completely but are less complex and expensive. Thus, it is highly
desirable to find passive, robust and economic countermeasures. Nishiwaki et al. [14] and Fieldhouse et al.[15]proposed the
introduction of asymmetries to the brake rotor to achieve this goal. There can also be found patents concerning the subject
[16,17]. While these approaches mainly originate from experimental work and partially aim for the splitting of two (or very
few) eigenfrequencies in direct vicinity to each other and the squeal frequency, the splitting of all eigenfrequencies of the
rotor in a certain frequency range has lately been shown mathematically to stabilize the brake system and avoid squeal [18].
This insight motivates a systematic structural optimization of the brake rotor to split all its double eigenfrequencies [19].
A structural optimization of a realistic automotive brake disc with a variation of the cooling channels to split all double
eigenfrequencies in a certain frequency range has not yet been discussed in the literature (to the best of the authors
knowledge) and will be the main focus of this paper. The frequency range in which the splitting is necessary will also be
estimated in this paper. This optimization problem has been shown to be nonlinear, nonconvex and with many local optima
to be expected [20] with nonlinear constraints originating from the requirement to keep the brake disc statically and
dynamically balanced. Furthermore, there are minimal and maximal distance constraints to be taken into account due to
cooling and material strength requirements.
An overview over the topic of structural optimization can be found in [21,22]. Since most structural optimization
problems are in fact nonlinear, the authors refer to [23] for a more general overview of nonlinear optimization. Frequently,
structural optimization is divided into sizing, shape and topology optimization. While sizing only allows the variation of
some predefined dimensions of a product, shape optimization allows more freedom in the design of the geometry by a more
flexible (and changeable) parametrization of the boundaries [24]. Finally, topology optimization allows the maximum
amount of freedom in the design space by a possible change even in the topology of a structure [21]. Some topology
optimization algorithms are discussed in [25,22]. However, frequently it is desired to allow large changes in the geometry
with a detailed description of the boundaries, which cannot directly be achieved with topology optimization. Therefore,
level set methods for this kind of structural optimization have been proposed [26,27]. In this paper, an alternative approach
to parameterize the geometry will be used, which also allows large variations in the geometry in detail and avoids time-
consuming remeshing prone to errors [28]. This method has been introduced in [29]. Often, the underlying nonlinear
optimization problems during structural optimization are solved with gradient-based algorithms [22], e.g. with Sequential
Quadratic Programming (SQP) methods [30]. These methods can be efficient for problems which do not exhibit too many
local optima (since they only converge locally for most nonlinear optimization problems [23]), especially if it is possible to
derive design sensitivities analytically or at least semi-analytically [28,31]. However, for noncontinuous problems [32] or
problems with many local optima, their performance might be poor. Therefore, also heuristic or metaheuristic approaches
are proposed in the context of structural optimization, which have the potential to approach the global optimum. They reach
from rather well-known genetic algorithms [33,34] over particle swarm [35] and firefly [36] algorithms to the so-called
harmony search algorithm [37], just to name a few. Also they have the advantage that they do not require design
sensitivities or gradients, which will be beneficial for the optimization strategy proposed in this paper.
The paper is organized as follows. First, a modeling strategy for the automotive brake disc with cooling channels is
proposed, which allows large changes in the geometry of the disc without the necessity to remesh the structure in every
iteration step. Then the minimal necessary distance between the eigenfrequencies of the (asymmetric) brake disc is
calculated together with the according frequency range. Next, the nonlinear, nonconvex optimization problem to split all
eigenfrequencies of the disc is introduced based on the insights gained before with the arising balance and minimal and
maximal distance constraints. Two approaches are presented to solve the proposed optimization problem. A deterministic
SQP approach with randomly distributed starting points leading to very satisfying results and a heuristic genetic algorithm
(GA) also leading to good results. A comparison between the two approaches follows before an experimental validation of
the basic idea behind the optimization follows. A short summary then concludes the paper.
1890 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898

Fig. 1. Part (a): Cross section of the brake disc. Part (b): Top view on the brake disc, two cooling ribs are shown exemplarily.

2. Modeling of the brake disc with cooling channels

In this section, a modeling technique for the automotive brake disc with cooling ribs will be introduced which allows an
efficient optimization and large changes in the geometry. A schematic drawing of the brake disc to be modeled is shown in
Fig. 1.
In order to allow an efficient structural optimization of the rotor with cooling channels, the hat of the brake disc is not
modeled explicitly, but the rotor is assumed to be simply supported at the inner lower ring (displacement degrees of
freedom (DOF) are set to zero, rotational DOF are not limited), as can be seen in Fig. 1(a). Fig. 1(a) shows the cross section of
the disc with an inner radius ri, an outer radius ra and a thickness h. Each friction ring has a thickness of hf, a modulus of
elasticity E, Poisson's ratio ν and a density ρ as well as the cooling ribs, which have a thickness of hc, the length lc and the
width bc. The cooling ribs are all assumed to be of the same shape. In the actual brake disc, the cooling ribs connect upper
and lower friction ring, if they were not present, both rings would not be connected. Since large changes in the geometry are
to be considered in the optimization presented in this paper, it may be necessary to frequently remesh the brake disc with
cooling ribs during the optimization process, if the disc is modeled conventionally with the FEM. In order to avoid this time-
consuming remeshing process, a method denoted as Energy-Modification-Method (EMM) has been proposed in [29]. The
basic idea is to separately discretize a basis structure, e.g. the brake's friction rings, and modifications, e.g. the cooling ribs,
and then to add (or subtract) the kinetic and potential energy of the modifications to (from) the kinetic and potential energy
of the basis structure. The mass and stiffness matrices of the assembled structure can then be gained by expressing the
coordinates of the modification by the coordinates of the basis structure and an extraction from the energy expressions.
They can then be used e.g. to calculate the eigenvalues and eigenvectors of the structure. It should be noted that in contrast
to the results shown in [29], here, the results presented in section four are obtained by an addition of energy expression with
positive material constants since substructures are added to the basis structure. The volume between the two friction rings,
where the ribs are placed, is modeled as material with very low density ρ~ and elastic modulus E~ in order to avoid
unbounded movement of the upper friction ring. The FEM discretization of the friction rings with the volume in between
leads to the mass matrix M and stiffness matrix K as the discretization of one cooling rib leads to its mass matrix M and
stiffness matrix K. The mass matrix Mc of the assembled system with N cooling ribs is then given as
N
Mc ¼ M þ ∑ STn MSn (1)
n¼1

and the stiffness matrix Kc as

N
Kc ¼ K þ ∑ STn KSn ; (2)
n¼1

resulting from the energy expressions. The coordinate coupling matrix Sn expresses the coordinates qn of each of the N
cooling ribs in terms of the coordinates of the basis structure q as

qn ¼ Sn q; (3)

and can be assembled using the local shape functions of the basis structure (here the friction rings). The circular
eigenfrequencies ω and eigenvectors v of the brake disc with cooling ribs can then be calculated from the eigenvalue
problem

ðKc  Mc ω2 Þv ¼ 0; (4)
 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898 1891

which can be solved in each optimization step using standard software. The coordinate coupling matrices Sn have to be
calculated anew after each change in the geometry during the optimization, however this is an efficient process [29] and
also allows large changes in the geometry. Nevertheless, it is not possible to calculate the gradients of mass and stiffness
matrices with respect to changes of geometry parameters analytically with this modeling technique, since due to the FEM
discretization a conversion between local and global coordinates is necessary to obtain the matrices Sn . This makes gradient
based structural optimization algorithms like SQP techniques less effective in combination with this modeling approach.
However, a SQP algorithm is satisfactorily used with this method as well as a gradient free GA as will be shown in Section 4
of this paper.

3. Frequency band and minimal necessary split of eigenfrequencies to avoid squeal

As has been demonstrated in [18], the splitting of all double eigenfrequencies of the stationary brake rotor helps to
stabilize the rotating brake system and thus avoids squeal. A larger distance between the eigenfrequencies leads to a more
stable system, however, for a comprehensive structural optimization of the brake disc with cooling channels this split has to
be quantified together with the desired frequency range. The optimization could concentrate on the whole audible
frequency range up to 20 kHz, but this would require to consider too many eigenfrequencies making the optimization
inefficient. A reduction in the necessary frequency range helps us to reduce the optimization effort largely since the solution
of the eigenvalue problem gets more effective. The method to estimate the frequency range in which a split of
eigenfrequencies is necessary and to estimate the minimal necessary distance between the eigenfrequencies to stabilize
the system has been proposed in [38]. The fundamental idea is to model the (asymmetric) brake rotor in frictional contact
with the brake pads as a linear conservative system under parametric periodic excitation and to approximate the stability
boundary of the system analytically. For the following estimate, the brake disc is simplified as a circular beam and the pads
as massless friction pins, as can be seen in Fig. 2.
These assumptions allow us to cover the major influences of the structure exhibiting self-excited vibrations while still
making it possible to achieve analytical expressions for the stability boundary. Furthermore, local and thermal effects are
neglected as well as the hat influence. Under the assumption of mass- and stiffness-proportional damping this
approximations then result in a conservative estimate of the minimal necessary distance Δf between the eigenfrequencies
of the brake rotor given by
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 2 2
1 u h k μ2
Δf Z pffiffiffi t
4 2 2
2
 4π 2 ð4π 2 αS f þαM Þ2 f ; (5)
2π r p ðmmin
2
G Þ

where f is the frequency, k is the overall stiffness of the brake pad, μ is the (constant) coefficient of friction between brake
pad and rotor, rp is the radius to the midpoint of the brake pad, mmin
G is the minimal generalized mass of all eigenmodes of
the brake rotor in the audible frequency range (which can be calculated from the FE model), while αS and αM give the mass-
and stiffness-proportional damping respectively. They can be calculated from a fit to experimental data. With increasing
frequency, the necessary minimal distance between the eigenfrequencies to achieve a stable system decreases such that
above a certain limit frequency the system is stable without any separation of eigenfrequencies due to material damping
alone. This limit frequency f lim can be calculated by setting Δf ¼ 0 leading to
2 2
h k μ2 2 2
2
 4π 2 ð4π 2 αS f lim þαM Þ2 f lim ¼ 0; (6)
r 2p ðmmin
G Þ

which can be solved for f lim . Using the parameters of the brake disc presented in Table 1, which are based on a medium
sized brake disc of a major car manufacturer, the minimal necessary distance can be estimated to be approx. 50 Hz and the
limit frequency above which the system is stable is 9 kHz. This information will be helpful in the next section, where the
structural optimization of the brake disc with cooling ribs to split its eigenfrequencies is discussed.

Fig. 2. Circular beam in frictional contact to massless pins rotating at a constant rotational frequency Ω. Sketch adopted in modified form from [19].
1892 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898

Table 1
Parameters of the automotive brake disc with cooling channels.

E ¼ 140 000 MPa ra ¼ 0:138 m hc ¼ 0:009 m k ¼ 1:1  106 N=m

ν ¼ 0:25 h ¼ 0:026 m bc ¼ 0:0065 m mGmin ¼ 1:81 kg
ρ ¼ 7200 kg=m3 hf ¼ 0:0085 m r p ¼ 0:125 m αM ¼ 0:184 s
ri ¼ 0:079 m lc ¼ 0:043 m μ ¼ 0:4 αS ¼ 2  10  10 1=s

4. Optimization of the eigenfrequency distribution of the brake disc with cooling ribs

The layout of the cooling channels of the considered automotive brake disc is to be optimized. Therefore, the radial
position rn and the angle φn of each of the N cooling ribs are varied, while their shape (lc and bc) is kept constant. In Fig. 1(b),
two cooling ribs are shown exemplarily, one at an angle φ1 and a radius r1 and the other with an angle φ2 and a radius r2.
The parameters of the brake disc are presented in Table 1. The optimization problem to split all eigenfrequencies fk of the
rotor up to a limit frequency of f lim (here 9 kHz) can be written as
max minjf k þ 1  f k j
r n ;φn k
s:t:
r min r r n r r max
0 rφn r 2π
φ1 o φ2 o⋯ oφN
φi φi þ 1 r  φmin ðmin: distance constraintsÞ
φi þ 1  φi r φmax ðmax: distance constraintsÞ
cb ðr; φÞ ¼ 0 ð2 balance constraintsÞ; (7)
where r min and r max are the minimal and maximal radius of each cooling rib to keep a minimal distance between cooling rib
and inner and outer radius of the brake disc. φmin and φmax are the minimal and maximal angle between two cooling ribs
and thus give the minimal and maximal distances between them, which have to be kept due to cooling and material
strength requirements. The brake disc has to be statically and dynamically balanced leading to the nonlinear equality
constraints
N
cb1 ðr; φÞ ¼ ∑ r n cos φn ;
n¼1
N
cb2 ðr; φÞ ¼ ∑ r n sin φn ; (8)
n¼1

which can easily be derived, since the brake disc's friction rings and each of the cooling ribs are assumed to be homogeneous
bodies of constant geometry rotating at a constant angular velocity. These balance constraints represent the projection of
the eccentricity of the brake disc induced by the placement of the cooling ribs onto the x-direction (cb1 ) and the y-direction
(cb2 ), respectively.
The presented optimization problem is a nonconvex, nonlinear optimization problem with linear and nonlinear
constraints with many local optima to be expected [20]. The brake disc to be optimized is modelled using the method
proposed in Section 2. The underlying FE meshes of the friction rings and each cooling rib have been assembled using
Abaqus with 3D solid isoparametric elements of quadratic shape function order (C3D20) leading to 65 250 DOF in the
friction ring mesh and 4860 DOF for each cooling rib fulfilling convergence requirements. The parameters of the filling
material between the friction rings have been chosen to be E~ ¼ 1:4 MPa and ρ~ ¼ 1 kg=m3 . Since the limit frequency f lim is
estimated to be 9 kHz, the first 21 eigenfrequencies of the brake disc with N ¼39 cooling ribs are considered in the
calculation. Solutions are treated as infeasible if one of the constraints cb1 and cb2 is violated by an absolute value of 0.005 m.
The parameters for the upper and lower bounds and the linear constraints can be found in Table 2.
First, a deterministic SQP approach is chosen to solve the optimization problem. Since SQP procedures are known to be
efficiently targeting local optima but unable to guarantee the finding of the global optimum for the problem class
considered here [23], 20 randomly chosen starting points were generated to increase the probability of finding satisfying
solutions. In the context addressed here, satisfying results mean results that exhibit a minimal separation of eigenfre-
quencies of at least 50 Hz in the frequency range up to 9 kHz as has been derived in Section 3. The Matlab Optimization
Toolbox provides the SQP implementation used here, an overview over the results is given in Table 3.
In order to avoid over-long runtimes, the optimization was stopped in the iteration following the reach of 10 000
function evaluations if the algorithm did not find a local optimum before or ended due to infeasibility. This lead to a stop of
the algorithm in 6 of 20 cases, all of which leading to feasible solutions and mostly very good results. Furthermore, 7 of 20
optimization runs lead to infeasible configurations. A solution was treated as a local optimum, if the change in the objective
function between two optimization iterations was below 10  12 , the absolute value of the step size below 10  8 and the
constraints were satisfied (to the limit given above).
 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898 1893

Table 2
Parameters for the optimization.

rmin ¼ 0:08 m φmin ¼ 4 deg:

rmax ¼ 0:094 m φmax ¼ 12 deg:

Table 3
Number of iterations, function evaluations, max. violation of constraints, minimal difference between the eigenfrequencies and infeasibility of the solutions
of the 20 SQP optimization approaches.

No. No. of iterations No. of function Constraint violation Minimal difference Infeasible
evaluations (m) (Hz)

1 10 963 4.1  10  3 45.3

2 58 5369 2.6  10  2 58.8 
3 111 10 026 2.5  10  6 64.5
4 44 4247 1.9  10  2 67.2 
5 16 1633 1.1  10  1 67.2 
6 13 1283 7.7  10  3 43.4 
7 70 6242 3.6  10  5 60.6
8 115 10 061 5.1  10  8 75.8
9 7 649 3.8  10  3 42.5
10 33 3589 3.2  10  2 57.1 
11 112 10 061 4.7  10  8 68.3
12 114 10 076 2.4  10  8 71.1
13 107 10 064 3.7  10  7 49.7
14 7 650 3.8  10  3 35.8
15 13 1304 4.1  10  3 45.3
16 57 5231 1.1  10  2 67.9 
17 20 1779 3.9  10  3 58.0
18 117 10 042 2.9  10  9 73.3
19 14 1391 4.7  10  2 43.0 
20 7 650 3.8  10  3 35.8

Fig. 3. Part (a): Maximum minimal distance between the eigenfrequencies of the 13 feasible SQP optimization runs. Part (b): Distances between the
21 eigenfrequencies of the best solution (solid, blue line, circular markers) and its starting configuration (dashed, orange line, square markers).
(For interpretation of the references to color in this figure caption, the reader is referred to the web version of this paper.)

Fig. 3(a) shows a comparison between the maximum minimal distances between the eigenfrequencies which could be
realized in the 13 feasible optimization approaches.
A dashed (orange) line shows the limit of 50 Hz, above which the solution can be assessed as satisfactory fulfilling the
requirement to avoid squeal. It can be seen that only 7 of the SQP approaches lead to good results, while 13 fail because of
either constraint violation or a solution exhibiting a too small separation of eigenfrequencies. In the best case, a minimal
difference between the eigenfrequencies of 75.8 Hz was realized, a comparison between the starting configuration's
distances between the eigenfrequencies and the optimized configuration's distances is shown in Fig. 3(b). Clearly, the
minimal difference between the eigenfrequencies has been increased largely during the optimization. Fig. 4(a) shows a top
view on the brake disc with the cooling ribs in the best configuration found with the SQP algorithm. Many ribs in the
optimized configuration have a radius rn being on either the lower or the upper bound. This can be seen in more detail in
Fig. 4(b), where the radius rn of each rib is plotted together with the angle φn . Also, the upper and lower bounds for the
1894 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898

Fig. 4. Part (a): Rib configuration of the best result optimized with the SQP algorithm. Part (b): Plot of the radius rn and the angle φn of each cooling rib for
the best configuration (blue  ) and the starting configuration (orange þ ). (For interpretation of the references to color in this figure caption, the reader is
referred to the web version of this paper.)

Fig. 5. Part (a): Evolution of the minimal distance between the eigenfrequencies. Part (b): Evolution of the maximum absolute constraint violation.

radius are given in the figure as dashed, black lines. Since the lower and upper bounds of the radius are induced by physical
necessities, it is not possible to relax these bounds to achieve even better results. The SQP approach leads to very satisfying
results, as has been shown, nevertheless it exhibits two drawbacks. The first is that it is based on the calculation of gradients
of the objective function and the nonlinear constraints which is less effective with the EMM used to model the brake disc.
The second is the fact that only 35 percent of all optimization approaches led to satisfying results, which is a rather low
percentage even if the nonlinearity and nonconvexity of the problem are considered.
Therefore, as a second approach to solve the problem, the use of a heuristic GA is proposed, provided the Matlab Global
Optimization Toolbox. For the initial population, 200 random distributions of cooling ribs have been generated. The
algorithm's stopping criterion has been chosen to be a minimal change in the objective function of 10  14 . The further
settings of the algorithms have been chosen such that the linear constraints and upper and lower bounds are always fulfilled
during the optimization. Also, the fraction of individuals performing crossover (generation of children from the parent
generation without mutation) has been set to be 60 percent. The algorithm took 6 generations and approx. 28 000 function
evaluations to reach an optimum solution with a minimal split between the eigenfrequencies of 60.2 Hz, which is satisfying.
Fig. 5(a) shows the evolution of the minimal distance between the 21 considered eigenfrequencies during the optimization
with the GA.
It can be seen that in each generation the objective function increases, which could be considered as unusual for genetic
algorithms. The development of the maximum absolute violation of the constraints is documented in Fig. 5(b). First, the
violation increases, then stays constant and then abruptly diminishes to a feasible value, which gives hint to the heuristic
characteristic of the approach. The rib configuration of the solution is shown in Fig. 6(a).
While many ribs of the SQP solution are at the upper and lower bounds, this is not the case for the solution generated
with the GA. The distance between the eigenfrequencies of the GA solution is presented in Fig. 6(b) exhibiting qualitatively a
 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898 1895

similar distribution as in the best case found with the SQP approach. It is obvious that the best solution of all approaches has
been found with one of the 20 SQP runs deterministically targeting a local optimum and not with the GA which is conceived
to target the global optimum due to its heuristic nature. This leads to the conclusion that for the presented nonlinear,
nonconvex optimization problem, a deterministic approach is a good choice for achieving satisfying results, even under the
expectation of many local optima [20]. Still, the GA has important advantages: It required approx. thrice as much function
evaluations to reach its solution as one SQP run, however, it directly led to a satisfying result, which the SQP approach did
only in 35 percent of all approaches. Furthermore, the GA approach profits more from parallelization than the SQP approach
(of course depending on the implementation) and is a gradient free method, which is beneficial when used in conjunction
with the EM method of modeling the structure to be optimized.

5. Experimental validation

In order to validate the result that a defined separation between the eigenfrequencies of the brake disc is a measure
against brake squeal, squeal experiments were conducted at a brake test rig at TU Darmstadt. The test bench with an
attached automotive disc is shown in a front and a top view in Fig. 7.
For this validation, a prototype has been optimized and tested that can readily be manufactured in a machine shop and
which does not require a casting process for manufacturing. The idea is to drill radial holes into a full steel disc as an
academic approximation of an automotive brake disc with cooling channels. In the optimization process, the angular
position and drilling depth of the inserted holes are varied to achieve the desired split between the eigenfrequencies.
Despite the fact that the prototype's geometry differs from the geometry of the discs optimized in Section 4, the goal of
introducing a split of all eigenfrequencies of 50 Hz in a frequency range up to 10 kHz is approximately the same as before.

Fig. 6. Part (a): Rib configuration of the result optimized with the GA. Part (b): Distances between the 21 eigenfrequencies of the solution of the GA.

Fig. 7. The brake test bench. Part (a): Front view. Part (b): Top view.
1896 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898

Table 4
Parameters of the automotive brake disc with radial holes.

E ¼ 210 000 MPa ρ ¼ 7850 kg=m3 r a ¼ 0:162 m r h ¼ 0:008 m

ν ¼ 0:3 r i ¼ 0:076 m h ¼ 0:026 m Nh ¼ 18

Fig. 8. Top view on the best optimized result of the prototype brake disc with radial holes.

Table 5
Measured eigenmodes of the full disc (FD) and the optimized one (OD).

Eigenmode fi (FD) fi (OD) Δf i (OD)

2/0 1486 1476 53

1529

3/0 3131 3107 47

3154

4/0 5194 5000 186

5186

5/0 7600 7132 212

7344

The parameters of the manufactured prototype are given in Table 4, where rh is the hole radius and Nh is the number
of holes.
Both parameters are kept constant during the optimization process. The optimization has been conducted with a
deterministic interior-point algorithm in direct analogy to the optimization with the SQP algorithm presented above. Twelve
randomly generated initial configurations were optimized and the best result was selected for manufacturing. It is shown in
a top view in Fig. 8.
The optimized disc and the full disc without holes as a reference were generated using S235JR steel for testing. Two types
of tests were performed with these prototypes: First a modal analysis to determine the eigenfrequencies of the tested discs
and the distances between them. And then squeal tests to determine the reduction in squeal affinity of the optimized brake
rotor. In both cases, the discs are mounted to an original car suspension, as can be seen in Fig. 7. While during the modal
analysis the disc is not rotating and the caliper is detached, the disc is driven by an electrical motor with adjustable
rotational speed for squeal tests. In the latter case, the caliper is attached for the application of variable brake pressure.
The surface velocity of points on the rotor is measured by a POLYTEC laser vibrometer with a PSV-400 scanning head in the
modal as well as in the squeal analysis test case. The disc is excited by hammer impulses at a point at the backside of the disc
in the modal analysis case and the transfer function between the velocity signal of the laser vibrometer and the force
transducer of the modal hammer is calculated. A Hanning window is used in the evaluation of the vibrometer signals to
avoid leakage effects. A regularly spaced grid with 150 measurement points on the surface of the disc is used to enable the
recording of disc eigenmodes up to higher orders. Ten measurements were taken at each point followed by an averaging of
the complex signals, to increase the reliability of the measurements. The sampling frequency was set to 32 kHz and the
frequency response spectra were evaluated up to 8 kHz using POLYTEC's analysis unit. For the modal analysis case, the
resolution of the frequency response spectrum was 1.95 Hz, while it was 3.9 Hz in the squeal test case offering a reasonable
accuracy. In addition to the generation of the transfer functions, it is possible to visualize the operating deflection shapes of
the brake disc using the POLYTEC software accompanying the laser vibrometer. This allows the identification of the
eigenmodes of the brake rotor realistically attached to the test rig. The denomination of the bending-dominated modes
 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898 1897

Fig. 9. Spectrum of the laser vibrometer's velocity signal during squeal of the full, symmetric disc.

of the disc is given by α=β, where α represents the number of nodal lines in circumferential direction and β the number of
nodal lines in radial direction. The clearly identifiable eigenmodes of the full disc and the optimized one are presented in
Table 5 in conjunction with the corresponding measured eigenfrequencies fi and the distances Δf i between the
eigenfrequencies of the optimized brake rotor.
Since all eigenmodes of the symmetric disc are double ones, only one measured frequency is given in the table. The
measurement results clearly show that the goal to split the eigenfrequencies of the brake disc with radial holes at least by
50 Hz could successfully be reached by the conducted optimization, if the missing 3 Hz in the distance of the 3/0 modes is
attributed to measurement inaccuracy.
Following this modal analysis, squeal tests were performed to assess the reduction in squeal affinity of the disc with
radial holes in comparison to the full disc. For these tests, the brake pressure was varied between 1 bar and 15 bar and the
rotational frequency between 1 rad/s and 20 rad/s. The surface temperature of the brake rotor was measured before and
after each test, and it was kept below 100 1C. Using the laser vibrometer, the velocity amplitude of one point on the surface
of the disc was measured, from which the spectrum during squeal could be determined. As a result, the spectrum for squeal
of the full disc is presented in Fig. 9.
The disc did squeal in a broadband of rotational frequencies from 5 rad/s to 15 rad/s but in a more narrow pressure band
of 8–10 bar. The squeal was always monofrequent with a frequency of 7.6 kHz, which exactly corresponds to the 5/0
eigenmode of the disc without brake pads. The pads obviously change the eigenfrequencies of the disc only marginally.
The same squeal test procedure was applied to the optimized brake disc with 18 radial holes. The measured surface
velocity spectra for the same range of rotational speed and brake pressure were analyzed. Still, no significant amplitude in
the spectra could be identified, the optimized brake disc did not squeal under any applied braking condition.
The three following points should be kept in mind concerning the interpretation of these test results. First, the existing
test rig at TU Darmstadt does not allow us to change the environmental conditions like air humidity or temperature during
the test, as is possible with more sophisticated test rigs used in the automotive industry. Second, the tests were performed
with constant rotational speed, realistic stops with decreasing speed could not be performed. And third, wear occurring over
the lifetime of a brake disc and especially of the pad could not be simulated. Despite these three facts, the test results
strongly indicate that the desired split of 50 Hz between the eigenfrequencies of the brake disc indeed avoids brake squeal
completely. This can be considered as a validation of the basic idea behind the proposed optimization. In future work the
authors intend to manufacture the realistic brake discs with cooling channels presented in Section 4 and test them as well.
The expectation is that these discs will show the same superior squeal behavior as the disc with radial holes.

6. Conclusions

In this paper, a structural optimization of an automotive brake disc to efficiently split its eigenfrequencies is presented,
which has been identified as a proper measure against brake squeal. The brake disc has been modeled with the EMM
approach, where friction rings and cooling ribs are separately discretized and then combined in the energy expressions. This
modelling allows large changes in the geometry and an efficient computation of the brake rotor's eigenfrequencies. Then,
the minimal necessary separation between the eigenfrequencies has been calculated to avoid squeal as well as the frequency
band, in which this separation is necessary. The splitting of the eigenfrequencies has been introduced as a highly nonlinear,
nonconvex optimization problem with linear and nonlinear constraints. They originate from the necessity to keep the brake
disc balanced and from geometrical and material strength requirements. Two approaches have successfully been used to
solve the optimization problem. The first being a deterministic SQP approach targeting local optima from 20 randomly
generated starting configurations, the second being a heuristic GA. The best solution has been determined with the SQP
method, however, only 35 percent of all starting points led to satisfying results. Since the GA has also advantages due to the
fact that it is a gradient free method and directly gave a reliable and satisfying result, both approaches can be recommended
for the solution of the elaborate optimization problem presented here. Both lead to a successful splitting of the
eigenfrequencies with a resulting brake disc design expected to help to avoid brake squeal. In order to verify this
1898 A. Wagner et al. / Journal of Sound and Vibration 333 (2014) 1888–1898

expectation, experiments with a simplified brake disc with radial holes have been conducted. This disc serves as a first,
rather academic approximation of a brake disc with cooling channels, which can easily be manufactured in a machine shop.
A modal analysis as well as squeal tests has been performed and it is shown that the asymmetric brake disc with radial holes
does not squeal under any applied brake pressure or rotational speed. This result strongly encourages to use systematically
designed asymmetric brake discs as a passive, low-cost measure for brake squeal avoidance.

Acknowledgments

The authors gratefully acknowledge the support of the DFG, SP 1198/3 and the Adam Opel AG.

References

[1] J.D. Fieldhouse, Professional overview on squealing phenomena and models, Eurobrake 2012, Dresden, Germany, 16th–18th April 2012.
[2] D. Hochlenert, G. Spelsberg-Korspeter, P. Hagedorn, A note on safety-relevant vibrations induced by brake squeal, Journal of Sound and Vibration 329
(2010) 3867–3872.
[3] T. Nakae, T. Ryu, A. Sueoka, Y. Nakano, T. Inoue, Squeal and chatter phenomena generated in a mountain bike disc brake, Journal of Sound and Vibration
330 (2011) 2138–2149.
[4] N.M. Kinkaid, O.M. O'Reilly, P. Papadopoulos, Automotive disc brake squeal, Journal of Sound and Vibration 267 (2003) 105–166.
[5] R. Ibrahim, Friction-induced vibration, chatter, squeal, and chaos, Part I: mechanics of contact and friction, Applied Mechanics Reviews 47 (1994)
209–226.
[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] M. Graf, G.P. Ostermeyer, Instabilities in the sliding of continua with surface inertias: an initiation mechanism for brake noise, Journal of Sound and
Vibration 330 (2011) 5269–5279.
[8] U. von Wagner, D. Hochlenert, P. Hagedorn, Minimal models for disk brake squeal, Journal of Sound and Vibration 302 (2007) 527–539.
[9] G. Spelsberg-Korspeter, D. Hochlenert, O. Kirillov, P. Hagedorn, In- and out-of-plane vibrations of a rotating plate with frictional contact: investigations
on squeal phenomena, Transactions of the ASME: Journal of Applied Mechanics 76 (2009). 041006-1–041006-15.
[10] D. Hochlenert, Nonlinear stability analysis of a disk brake model, Nonlinear Dynamics 58 (2009) 63–73.
[11] J.J. Sinou, Transient non-linear dynamic analysis of automotive disc brake squeal—on the need to consider both stability and non-linear analysis,
Mechanics Research Communications 37 (2010) 96–105.
[12] U. von Wagner, D. Hochlenert, T. Jearsiripongkul, P. Hagedorn, Active control of brake squeal via ‘smart pads’, SAE Technical Paper 2004-01-2773, 2004.
[13] M. Neubauer, J. Wallaschek, Vibration damping with shunted piezoceramics: fundamentals and technical applications, Mechanical Systems and Signal
Processing 36 (2013) 36–52.
[14] M. Nishiwaki, H. Harada, H. Okamura, T. Ikeuchi, Study on disc brake squeal, SAE Technical Paper 890864, 1989.
[15] J.D. Fieldhouse, W.P. Steel, J.C. Talbot, M.A. Siddiqui, Rotor asymmetry used to reduce disc brake noise, SAE Technical Paper 2004-01-2797, 2004.
[16] J. Day, Disc brake rotor, U.S. Patent No. 3,298,476, 1967.
[17] T. Suga, M. Katagiri, Disk brake rotor exhibiting different modes of vibration on opposite sides during braking, U.S. Patent No. 5,735,366, 1998.
[18] G. Spelsberg-Korspeter, Breaking of symmetries for stabilization of rotating continua in frictional contact, Journal of Sound and Vibration 322 (2009)
798–807.
[19] G. Spelsberg-Korspeter, Structural optimization for the avoidance of self-excited vibrations based on analytical models, Journal of Sound and Vibration
329 (2010) 4829–4840.
[20] G. Spelsberg-Korspeter, Eigenvalue optimization against brake squeal: symmetry, mathematical background and experiments, Journal of Sound and
Vibration 331 (2012) 4259–4268.
[21] K. Saitou, K. Izui, S. Nishiwaki, P. Papalambros, A survey of structural optimization in mechanical product development, Transactions of the ASME:
Journal of Computing and Information Science in Engineering 5 (2005) 214–226.
[22] L. Harzheim, Strukturoptimierung: Grundlagen und Anwendungen (Structural Optimization: Basics and applications), Wissenschaftlicher Verlag Harri
Deutsch, Frankfurt am Main, 2008.
[23] W. Alt, Nichtlineare Optimierung: Eine Einführung in Theorie, Verfahren und Anwendungen (Nonlinear Optimization: An Introduction to Theory, Algorithms
and Applications), Viewegþ Teubner, 2nd edition, Springer Fachmedien, Wiesbaden, 2011.
[24] Y.L. Hsu, A review of structural shape optimization, Computers in Industry 26 (1994) 3–13.
[25] S. Bulman, J. Sienz, E. Hinton, Comparisons between algorithms for structural topology optimization using a series of benchmark studies, Computers
and Structures 79 (2001) 1203–1218.
[26] Z. Luo, L. Tong, Z. Kang, A level set method for structural shape and topology optimization using radial basis functions, Computers and Structures 87
(2009) 425–434.
[27] P. Wei, M.Y. Wang, X. Xing, A study on X-FEM in continuum structural optimization using a level set model, Computer-Aided Design 42 (2010) 708–719.
[28] F. van Keulen, R.T. Haftka, N.H. Kim, Review of options for structural design sensitivity analysis. Part 1: linear systems, Computer Methods in Applied
Mechanics and Engineering 194 (2005) 3213–3243.
[29] A. Wagner, G. Spelsberg-Korspeter, An efficient approach for the assembly of mass and stiffness matrices of structures with modifications, Journal of
Sound and Vibration 332 (2013) 4296–4307.
[30] R. Sedaghati, Benchmark case studies in structural design optimization using the force method, International Journal of Solids and Structures 42 (2005)
5848–5871.
[31] K.U. Bletzinger, M. Firl, F. Daoud, Approximation of derivatives in semi-analytical structural optimization, Computers and Structures 86 (2008)
1404–1416.
[32] P.B. Thanedar, G.N. Vanderplaats, Survey of discrete variable optimization for structural design, Journal of Structural Engineering 121 (1995) 301–306.
[33] Y.D. Kwon, S.B. Kwon, S.B. Jin, J.Y. Kim, Convergence enhanced genetic algorithm with successive zooming method for solving continuous optimization
problems, Computers and Structures 81 (2003) 1715–1725.
[34] N. Ali, K. Behdinan, Z. Fawaz, Applicability and viability of a GA based finite element analysis architecture for structural design optimization, Computers
and Structures 81 (2003) 2259–2271.
[35] P.W. Jansen, R.E. Perez, Constrained structural design optimization via a parallel augmented Lagrangian particle swarm optimization approach,
Computers and Structures 89 (2011) 1352–1366.
[36] A.H. Gandomi, X.S. Yang, A.H. Alavi, Mixed variable structural optimization using Firefly algorithm, Computers and Structures 89 (2011) 2325–2336.
[37] K.S. Lee, Z.W. Geem, A new structural optimization method based on the harmony search algorithm, Computers and Structures 82 (2004) 781–798.
[38] A. Wagner, M. Schönecker, G. Spelsberg-Korspeter, P. Hagedorn, On criteria for the robust design of squeal free brakes, SAE Technical Paper 2012-01-
1816, 2012.