Applied Acoustics 91 (2015) 12–24

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

Reduction strategy for a brake system with local frictional

non-linearities – Application for the prediction of unstable
vibration modes
O. Fazio a,b, S. Nacivet b, J-.J. Sinou a,⇑
a
Laboratoire de Tribologie et Dynamique des Systèmes, UMR CNRS 5513, Ecole Centrale de Lyon, 36 avenue Guy de Collongue, 69134 Ecully Cedex, France
b
PSA Peugeot Citroën, Centre technique de La Garenne Colombes, 18 rue des Fauvelles, 92250 La Garenne Colombes, France

a r t i c l e i n f o a b s t r a c t

Article history: Although the problem of friction-induced vibration has been the subject of many investigations over
Received 13 March 2014 recent decades, it is still responsible for a large number of nuisances in the field of automotive. This study
Received in revised form 5 November 2014 presents a numerical process based on modal reduction minimizing the size of the model thanks to a spe-
Accepted 14 November 2014
cific non-linear modelling at the frictional interface in order to predict the stability analysis of large finite
Available online 13 December 2014
element models that correspond to real automotive braking systems subjected to friction-induced vibra-
tions. In the present study, the effect of the number of contact nodes at the frictional interface will be
Keywords:
investigated for various operating conditions. It will be illustrated that an original contact reduction at
Vibration
Friction
the frictional interface coupled to Super-Element use can be developed in order to well represent squeal
Squeal instabilities (i.e. the frequencies and the associated unstable modes). It will be demonstrated that the
Reduction proposed strategy guarantees a high-quality estimation of the stability analysis for the real brake system
Super-Element subjected to friction-induced vibration.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction the origin of noises are nowadays a current challenge for the scien-
tific community and the automotive industrialists.
The problem of brake squeal in automotive was a subject of In general, there are two main steps toward friction-induced
great interest for many researchers. The problem of squeal model- problems: the stability analysis and the non-linear transient anal-
ling has been extensively undertaken in the past years [1–5] and a ysis. The stability analysis can also be divided into two parts. The
lot of numerical and experimental studies including uncertainties first part concerns the non-linear static problem: a sliding equilib-
[6,7], statistical analysis of brake squeal noise [8], non-linear for- rium point is calculated by solving the non-linear static equations
mulations and behaviours at the frictional interfaces [9–14], cha- for a given brake pressure. Then, the linearized equations of motion
otic phenomenon [15] or acoustic emissions [16–18] have been are estimated by introducing small perturbations about the sliding
formulated and developed in order to explain squeal phenomena equilibrium point into the non-linear system. Finally, the classical
and nuisances in the field of automotive. Some researchers have Complex Eigenvalue Analysis (CEA) is performed on the linearized
also highlighted the predominant role of the brake pads [19] or system in order to predict the squeal propensity of the brake in a
the significant impact of the contact surface topography and the given frequency range. This typical deterministic approach is of
frictional contact model [20–22] in squeal noise phenomena pre- common practice and well appreciated in industry. The main
diction. Other studies illustrated the fact that introducing uncer- advantage of the stability analysis is the computational efficiency
tainty and robustness concepts during simulations can also in order to estimate the occurrence of self-excited vibrations. How-
improve accuracy of squeal occurrences [7]. However, it is still dif- ever, the main disadvantage and limitation of the stability analysis
ficult to predict brake squeal early enough to reduce development concerns the underestimation or over-estimation of the unstable
costs in a manufacturing process of brake system. So the problem modes that can be observed in the non-linear simulation. In fact,
of modelling squeal noise and understanding the mechanisms at the stability analysis can only predict the onset of instability
around a given equilibrium point for a non-linear system and so
⇑ Corresponding author. a complete non-linear approach can be more informative [11,23].
E-mail addresses: [email protected] (O. Fazio), [email protected] Unfortunately, the calculation of the transient and non-linear
(S. Nacivet), [email protected] (J-.J. Sinou). vibrations are mostly performed by numerical integrations which

http://dx.doi.org/10.1016/j.apacoust.2014.11.005
0003-682X/Ó 2014 Elsevier Ltd. All rights reserved.
 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24 13

(a) Exploded view (b) Finite element model

Fig. 1. Brake system details.

Table 1 TGV brake system. In these studies, all the physical contact degrees
Contact modelling. of freedom are kept. Vermot des Roches [24] proposed parame-
Contact Friction coefficient Separation trized reduction called the Component Mode Tuning method
Disc/Pad 0.5 Yes
(CMT) that use the components free/free modes as explicit degrees
Piston/Pad 0.15 No of freedom in order to perform non-linear time simulations.
Bracket/Pad 0.15 Yes However, few efforts have been dedicated to discuss the possi-
Caliper/Pad 0.15 No bility to reduce the size of the frictional interface that is one of the
main drawback of the classical use of the Craig–Bampton approach
(see for example [12]). So the main contribution of the present
are rather expensive and requires considerable resources both in study is to propose several improvements in order to reduce the
terms of computation time and data storage [24,25]. For a large- number of contact nodes at the frictional interface. The generation
scale finite element model, numerical integration can become of the reduced non-linear interface model is performed by creating
computationally so expensive that it is impractical or irrealistic Super-Elements (SE) with reduced node to node contact at the disc/
in an industrial context. Based on those facts and according on pad interface. The global reduction will then combine the classical
the size of the problem, industrial studies focus only on the Com- Craig–Bampton approach (i.e. a modal reduction strategy) and an
plex Eigenvalue Analysis for linear equivalent brake systems or implicit reduction at the frictional interface by applying the inter-
perform a non-linear approach in order to assess the non-linear face Super-Element (SE) generation. This provides a reduction
self-induced vibrations. method that returns accurate results for a drastically reduced finite
Therefore, some researchers proposed to develop reduced mod- element system in a lesser time while keeping the non-linear
els when finite element models are considered in order to reduce behaviour at the frictional interface. The proposed method gives
the computational times. For brake system and the squeal predic- then a great advantage in terms of memory.
tion, the most widespread reduction method in industry is the The main objective of this study is to propose a strategy based
Craig–Bampton method [26]. One of the major limitations is asso- on numerical simulations for predicting the stability analysis of a
ciated with the size of interface matrices due to the explicit use of real brake system subjected to friction-induced vibration. This
the degrees of freedom at interface. Recently, performances of strategy consists in defining an efficient contact reduction at the
some others reduced bases built from the component modes or frictional interface in order to use a Super-Element for further
the real coupled modes have been tested by Brizard et al. [27] for study of the stability.
the stability analysis of a disc/pads system in sliding contact. They To achieve the previous objectives, the paper is organized as fol-
proposed some enriched bases to improve the precision on the cal- lows: firstly, the brake system under study is presented. Secondly,
culated complex modes and eigenvalues. Loyer et al. [25] devel- the proposed reduction method and the global strategy for the gen-
oped also spatial model reduction with different kinds of eration of the reduced non-linear interface are discussed. More
reduction bases to approximate the non-linear vibrations of a precisely the Super-Element creation and Super-Element assembly
are detailed. Finally, application for an industrial representative
finite element model is proposed. Results on different cases and
the influence of the number of nodes at the non-linear frictional
interface are presented and commented. The efficiency of the
reduction strategy is undertaken by performing a stability analysis
around a non-linear equilibrium point for various operating
conditions.

2. The brake model under study

For this study a finite element model of an industrial brake sys-

tem that has been developed in Abaqus is considered. It can be
decomposed in several parts as detailed in Fig. 1(a). This system
uses a floating caliper technology. The caliper that holds the two
Fig. 2. The reduced contact model. brake pads can move with respect to the disc, along a line parallel
14 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24

(a) 1st degree

Fig. 3. Global strategy.

(b) 2nd degree

Fig. 4. Reduced contact nodes layout. (c) 3rd degree

Fig. 5. Weight factor distribution inside a triangle of three reduced contact nodes
to the axis of rotation of the disc. During a brake operation, hydrau- for several interpolation degrees. An equal force is applied on each node.
lic pressure is applied on the piston, which pushes the inner brake
pad until it makes contact with the disc. Then, reaction force pulls
sentative models [24,27]. Alternatively, Villard et al. [28] used an
the caliper body with the outer brake pad against the other side of
assembly of Super-Elements (SE) with reduced contact interface.
the disc. Considering the finite element model, a volume mesh is
However, some questions remained after this study particularly
realized using 10-node quadratic tetrahedron (see Fig. 1(b)). It rep-
concerning the modelling of the simplified contact interface.
resents a total of 176,956 nodes for 100,710 elements.
This work was achieved by performing simulations with Abaqus
Contact and friction modelling are key parameters for squeal
and Matlab. This section will describe the different steps in detail,
simulation. Here,  CONTACT PAIR card is used for node to surface
but it is interesting for the reader comprehension to first sum up
contact modelling. The principal components of the brake system
the whole process. Firstly, Abaqus simulation is used to define an
in contact with the pads are the disc, the piston, the caliper and
original reduced contact interface at the disc/pad interface with
the bracket. For all these contacts, pads are always defined as slave
reduced number of nodes on both sides of pad and disc that will
parts.
be later used for condensation and node to node contact. Secondly,
A linear penalty method is used for the contact constraint
the static equilibrium is found on this Abaqus model with reduced
enforcement and the classic Coulomb’s law is used for all the fric-
contact interface. Then, deactivating the disc/pad contact while fix-
tional interfaces. Also depending on the contact considered, sepa-
ing the displacement of the nodes of this reduced contact interface
ration can be taken into account or not. All these parameters are
observed at the static equilibrium allows generating a Super-Ele-
summed up in Table 1.
ment which is a linearization of this equilibrium configuration.
Node to node contact is then used in Matlab at the disc/pad inter-
3. Model reduction method face when the SE element is used to compute stability analysis
with a new cubic contact law for the disc/pad interface.
Previous research works were carried out to define a model In the next paragraphs, all these previous steps are detailed.
reduction method to apply stability analysis on industrial repre- This section is organized as follows: first the simplified contact
 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24 15

Table 2 the part surface. Indeed, this reduction method is interesting only
Weight factor computation results for several interpolation degrees: optimal radius, if the reduced contact interface computation is relevant and does
mean value (x ) and standard deviation (r).
not generate error on the non-linear static equilibrium. In this
Interpolation degree Radius (mm) x
 r strategy it is interesting to exploit the  DISTRIBUTING and  COU-
1 0.9501 1 0.0574 PLING cards in Abaqus in order to determine the best set of param-
2 0.7425 0.9999 0.0835 eters [29] and decrease the error induced by this reduction and
3 0.9597 1 0.0234 thus improve the previous reduction method.
One of the contribution of this study on simplified contact mod-
elling is to create one Super-Element reducing only the disc/pad
modelling will be explained. Then, the Super-Element creation will interface. All the other interfaces are then linearized during the
be detail to then gives some information concerning the Super ele- Super-Element creation. This choice allowed us to reduce the error
ment assembly and more particularly the method used to define involved by contact reduction and better understand its effect.
contact force in the node to node contact used for the Super-Ele- Indeed increasing the number of reduced contact interfaces sums
ment assembly. Finally some new developments brought to up the error involved by reduction. The major drawback is that
improve the reduction method proposed by Villard et al. [28] will individual damping cannot be considered anymore for each com-
be presented. ponent during Super-Element assembly.

3.1. Global strategy and simplified contact modelling

3.2. Super-Element creation and Super-Element assembly

The idea put forward by Villard et al. [28] is to create an assem-

3.2.1. Super-Element creation
bly of Super-Elements at the contact zones defining a node to node
The first step concerns the Super-Element creation. Its genera-
contact. Due to the fact that industrial meshing process drives to
tion can be summarized as follows: first of all the reduced disc/
non-coincident nodes at the frictional interfaces, Villard suggested
pad contact interface (i.e. a contact with a reduced number of
to generate new contact interfaces for SE assembly using node to
nodes using  DISTIBUTING and  COUPLING cards) is created on
node contact with no need to re-mesh the model. Moreover this
the Abaqus model. Then, a non-linear static analysis is performed
process allows to reduce the number of degrees of freedom
on the Abaqus model with reduced contact interface and the dis-
(DOF) at the frictional interfaces of the initial brake system.
placements at the nodes retained for Super-Element creation are
The nodes involved in the new contact interface will be later
stored. We will see later why it is also interesting to store reduced
used as reduction nodes for the SE creation. The objective here is
contact variables (contact force and gap). In a second step, contact
to have the smallest number of contact nodes in order to reduce
at disc/pad interface is deactivated but the displacements mea-
the size of the Super-Element created in retaining this reduced
sured are still applied to the future reduction nodes with the
number of contact nodes. Of course reducing the number of nodes
‘‘ Boundary, FIXED’’ parameter. This allows to linearize the other
kept for computing the contact at the disc/pad interface must not
contacts of the model, that will not be taken in account in Matlab,
introduce error in the static equilibrium estimation.
at the static equilibrium while generating the Super-Element.
This reduced number of nodes involved in the several contacts
Finally a Craig and Bampton reduction [30] is computed and the
will be used to later define a node to node contact between the
Super-Element is generated. The reduced number of nodes used
Super-Element (whereas a node to surface contact is used in the
in the disc/pad contact interface are kept as reduction nodes.
reference model). In so far as the reduction of the number of nodes
involved in contact must not affect the static solution Villard et al.
[28] suggest to distribute the contact forces around these reduced 3.2.2. Super-Element assembly
contact nodes. The  DISTRIBUTING and  COUPLING options are The second step concerns the Super-Element assembly. It may
therefore used in Abaqus. Fig. 2 illustrates this new proposed con- be noticed that Abaqus in no more used and all the computations
tact modelling. A reference node is created in the plane of the ini- are done thanks to Matlab software, as illustrated in Fig. 3. This
tial contact surface. It allows to distribute contact force on the subsection details the method to compute the non-linear static
nodes of the initial meshed surface involved in the contact. This step of the SE assembly at the exact static equilibrium.
reference node cannot directly be used for the contact computation The non-linear static system solved by Abaqus is described as
due to an incompatibility. Due to the fact that SE reduction cannot follows:
be directly applied to this reference node, a new node must be cre-
ated for contact force computation and used as a reduction node Knl;Us ðUs ÞUs þ Fnl ðUs Þ ¼ Fext ð1Þ
when SE will be created. An additional spring is created with arbi-
trary stiffness (i.e. the value of this spring is assumed to be very where Us corresponds to the static equilibrium position. Fext repre-
superior to the local stiffness) to link these two nodes. For the sents the external force due to the external pressure applied to the
reader comprehension, it can be noticed that these nodes have piston and the caliper representing a brake operation. Knl is the
the same geometric position in the model but they are dissociated stiffness matrix due to the structural components and the three
on Fig. 2 for a better illustration of the reduced contact modelling. contact interfaces Piston/Pad, Bracket/Pad and Caliper/Pad. Fnl con-
Then, the contact forces (but not the moment) computed at the tains the non-linear forces at the frictional interface between the
reduced contact interface are distributed on the coupling nodes of disc and the pad.

Table 3
Reduced models characteristics.

Abaqus models Case 1 Case 2 Case 3 Case 4

Abaqus12 Abaqus44 Abaqus104 Abaqus212
Reduced contact nodes per pad 6 22 52 106
Total coincident nodes on disc volume mesh 12 44 104 212
Total references nodes for SE 26 90 210 526
Number of nodes to node contact elements 12 44 104 212
16 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24

(a) 12 nodes (b) 44 nodes

(c) 104 nodes (d) 212 nodes

Fig. 6. Reduced contact nodes layout over the pad surface for several reduced model.

As the Super-Element is created in a perturbation step, the sys- where kl and knl are respectively the linear and the non-linear stiff-
tem linearized around the static equilibrium can be written as nesses and Ui and Uj the displacements of the coincident nodes i
follows: and j from the master and slave sides of the contact respectively.
The subscript matlab indicates that the non-linear forces computation
Knl;Us Us þ Fnl ðUs Þ ¼ Fext ð2Þ
is performed on the Matlab environment.
where Knl;Us defines the linearized stiffness matrix at the vicinity of The main advantage of this choice is that this formulation is
the static equilibrium Us . All the contacts, except for the disc/pad easy to linearize for further stability analysis. As said earlier we
interface, are linearized as a constant value in the stiffness matrix. need to ensure the continuity in the contact force estimation
The local stability is studied introducing a perturbation DU between Abaqus and Matlab software. With two different contact
around the static equilibrium: formulations the only way to respect this constraint is to find an
equivalent contact penetration for each node to node contact
Knl;Us ðUs þ DUÞ þ Fnl ðUs þ DUÞ ¼ Fext ð3Þ
defined in Matlab in order to match the normal contact force value
Considering the Matlab implementation, the linearized system obtained with Abaqus software. Knowing the value of Abaqus nor-
at the static non-linear equilibrium point is described in the fol- mal contact force for each reduced node and the cubic stiffness
lowing equation: parameters kl and knl , it is easy to solve the following relation
Knl;Us Us þ Fnl;matlab ðUs Þ ¼ Fext ð4Þ Fmatlab ðd þ nÞ ¼ Fnl ðdÞ ð8Þ
where Fnl;matlab corresponds to the non-linear contact as the disc/pad
where Fnl are the contact forces computed earlier with Abaqus soft-
interface that will be described in the next section. Knl;Us is obtained
ware and d the contact penetration associated to the static equilib-
thanks to the Super-Element creation in Abaqus. The only difference
rium point that has been previously estimated using Abaqus
with Eq. (2) is the way to compute the non-linear force involved in
software. As said earlier this contact is computed using a linear pen-
disc/pad contact. According to the strategy defined above this con-
alty method in Abaqus software. Then the corrective term n in node
tact is now a node to node contact between condensation nodes. As
to node contact penetration can be easily found.
the calculated linearized equilibrium points for Abaqus software
Fig. 3 details all the steps explained earlier of our model reduc-
and Matlab software are the same, an equivalence of the contact
tion strategy.
force at the disc/pad interface for the non-linear equilibrium point
has to be performed:
3.3. Interface reduction study: radius of influence, weighting methods
Fmatlab ðUs Þ ¼ Fnl ðUs Þ ð5Þ and triangular distribution
The next paragraph will develop the solution proposed here to
respect the relation detailed in the previous equation. In this section the simplified contact modelling defined in 3.1
will be detailed. By default, using  DISTRIBUTING and  COUPLING
3.2.3. Generation of the contact force cards to couple a single node to a meshed surface transfer the
Considering experimental data, Coudeyras [12] chose a cubic forces computed at the reference node to all the nodes of the
stiffness for the disc/pad non-linear contact. For each contact node, meshed surface. An interesting option is to limit the distribution
its formulation is given in Eqs. (6) and (7). in a spherical region centred on the reference node by defining a
( radius of influence. In this case the forces computed at the refer-
kl d þ knl d3 if d > 0 ence node are distributed only to the nodes of the surface that fall
Fmatlab ¼ ð6Þ
0 otherwise inside this spherical region. Another interesting parameter is the
control over transmission through weight factors specified at each
d ¼ Ui  Uj ð7Þ
coupling nodes which allow the forces transferred to the coupling
 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24 17

Fig. 7. Contact state comparison for each reduced model. The left column is for the inner pad and the right one is for the outer pad. (a) and (b) Abaqus12, (c) and (d) Abaqus44,
(e) and (f) Abaqus104, and (g) and (h) Abaqus212.

nodes to vary inversely with the radial distance from the reference interface. Indeed it is not recommended to concentrate the forces
node. For the proposed contact reduction, the radius of influence at the vicinity of the new reduced contact nodes and thus to mod-
and weighting method are the main parameters to be fitted in ify the static equilibrium. Considering these three weighting meth-
order to have a smooth variation of contact force over the coupling ods it is interesting to find for each of them the most interesting
nodes. radius of influence in order to ensure an uniform contact force dis-
In Abaqus 6.10 the influence of the mesh size is also taken into tribution that matches the most with the initial non-reduced con-
account in the weighting factor computation but it was not consid-
ered in this study. To facilitate the understanding of the present
study, the classic polynomials defined below as weighting method
will be used
ri
wi ¼ 1  ð9Þ
R
r 2
i
wi ¼ 1  ð10Þ
R
r 2 r 3
i i
wi ¼ 1  3  þ2 ð11Þ
R R
where ri is the distance between the reference node and the ith cou- (a) Non-coincident mesh (b) Coincident mesh
pling node, R the radius of influence and wi the weight factor
Fig. 8. Contact force distribution according to the mesh coincidence. Triangles
computed. represent the master surface facets. The circle corresponds to the projection of a
As said before, reducing the number of contact nodes and dis- slave node on the master surface. Squares shows the master nodes where a contact
tributing force must not affect the contact force mapping on the force is computed.
18 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24

Fig. 9. Constraint field comparison between Abaqus reference model (e–h) and Abaqus212 (i–l) for disc/pad interface friction coefficient l ¼ 0:5 – (e) and (i) Outer pad (f) and
(j) Inner pad (g) and (k) Outer disc surface (h) and (l) Inner disc surface. Results are given in Mpa (a–d).

tact. As distribution is made in a circular region centred on the ref- card used for contact in Abaqus due to the fact that the methodol-
erence node, it is interesting to define the reduced contact nodes ogy involves node-based surface creation which cannot be defined
group as an isometric grid to reduce voids in distribution (see as the master surface of a contact pair in Abaqus. One solution
Fig. 4). Thus optimal radius will depend on the triangle size of could be to directly define node to node contact between reduced
the isometric grid. So the optimum was calculated for an academic node of both side directly in Abaqus with  GAP elements with fric-
example of a simple triangle of three nodes representing the ele- tion. However, this strategy does not allow us to generate unsym-
mentary pattern of our isometric grid. Weighting coefficient map- metrical terms in matrices which is the key point for stability
ping inside this triangle of reduced nodes is computed in Matlab analysis. So we decided to create reduced contact nodes only for
for a triangle length of 1 mm. As an equal force on each node of the pad: coincident nodes on the disc volume mesh are imposed
the triangle is applied, the optimal radius that offers an homoge- and a node to surface formulation with  CONTACT PAIR card is
neous force distribution has to be found for each weighting kept. The reduced contact nodes and their coincident nodes on disc
method, which means a computed weight factor centred on 1 mesh will then be used for condensation and node to node contact
inside this triangle. Results are given on Fig. 5 and Table 2. will be defined later in Matlab between them.
It seems interesting to use a cubic distribution that offers the To illustrate the effectiveness of the proposed strategy and dis-
smallest standard deviation. Whatever the number of reduced con- cuss the impact of the size of the proposed reduction, several Aba-
tact nodes and so the size of the triangle chosen later in our reduction qus models with several triangle size for disc/pad contact interface
it will be easy to find, proportionally to these previous results, the reduction will be investigated. Table 3 sums up the characteristics
radius of distribution that needs to be used. However the edge of con- of the four models under consideration. Fig. 6 gives an insight of
tact surface might be subjected to boundary effect of the distribution. the reduced contact nodes layout over the pad surface. As point
of comparison, the industrial Abaqus model contains 1370 nodes
4. Application for the reduced industrial model on the surface of each pad.
In the following of the study, the four Abaqus models with
This section presents the application of the reduction strategy reduced contact interface will be called Abaqus12, Abaqus44, Aba-
defined above to the industrial brake model. The main objective qus104 and Abaqus212 in reference to the number of reduced con-
of this section is to illustrate the efficiency of the proposed meth- tact elements they contain.
odology to build a reduced finite element model while keeping a
good prediction on the three main instabilities observed on the ref-
4.2. Abaqus reduced models: non-linear static analysis
erence Abaqus model.
As previously explained in [31], the complex eigenvalue analy-
4.1. Contact formulation and reduced models characteristics sis is based on the determination of eigenvalues for the linearized
system around the non-linear static equilibrium. The first step to
First of all, it must be noted that the proposed strategy and evaluate the performance of the reduced contact interface is to
reduction formulation is not compatible with the  CONTACT PAIR compare some static quantities between the Abaqus reference
 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24 19

reference model is superimposed for each reduced models: the

green and blue colours represent the closed and opened contact
state.
As illustrated in Fig. 7(a)–(b), the first reduced model Abaqus12
(with only 6 nodes by pad) is not sufficient to respect an appropri-
ate contact state at the static equilibrium point: all the reduced
contact nodes are closed, whereas the reference model indicates
both opened and closed contact area. Refining the number of
reduced contact nodes allows to represent the variation in the con-
tact state for the reference model. For example, it clearly appears
that the reduced model Abaqus212 is in perfect agreement with
the reference model, as indicated in Fig. 7(g)–(h).
Despite one or two nodes at the boundary between opened and
closed contact zone the numerical results with the other two
reduced models (i.e. Abaqus44 and Abqus104) are consistent with
those of the reference model (see Fig. 7(c)–(f), respectively). This
illustrates the fact that the number of nodes at the frictional con-
tact interface can be drastically reduced in order to estimate the
contact states by applying an appropriate strategy.

4.2.2. Constraints
The second verification to illustrate the efficiency of the pro-
posed reduction during consideration of the non-linear static anal-
ysis is to investigate the evolution of the constraint field at the
frictional interface.
Firstly, it may be noted that the general node to surface contact
is defined such as each slave node interacts with a projection point
on the master surface. If the projection of the slave nodes on the
master surface falls inside an element facet, the transmitted load

1
Fig. 10. Constraint field comparison at the inner pad between Abaqus reference
1500
Abaqus ref
model (c and d) and Abaqus212 (e and f) for disc/pad interface friction coefficient Abaqus12 0.9
l ¼ 0:2 – (c–e) and l ¼ 0:8 (d–f) – results are given in MPa (a and b). Abaqus44
0.8
Abaqus104
Abaqus212
0.7
and the four reduced models (Abaqus12, Abaqus44, Abaqus104 1000
and Abaqus212): the estimation of the contact state at the fric- 0.6
Real Part

tional interface and the constraints distribution will be

0.5
undertaken.
0.4
500
4.2.1. Contact state 0.3
The first proposed comparison between the four reduced mod-
0.2
els and the reference model corresponds to the contact state when
establishing the non-linear static equilibrium. So, a non-linear sta- 0.1
tic analysis for a friction coefficient value of l ¼ 0:5 (at the disc/
0 0
pad interface) is computed. Fig. 7(a)–(h) shows the results for the 1500 2000 2500 3000 3500 4000 4500 5000
four reduced models (i.e. Abaqus12, Abaqus44, Abaqus104 and Frequency
Abaqus212, respectively). Contact state at each reduced node is
represented by a circle (respectively by a plus sign) if the contact Fig. 12. Evolution of real part and frequency of the instabilities studied against the
is open (respectively closed). Moreover, the contact map of the friction coefficient for Abaqus12, Abaqus44, Abaqus104 and Abaqus212.

Fig. 11. Vom Mises stress comparison on outer disc surface for l ¼ 0:5 (a) Abaqus reference, (b) Abaqus12, (c) Abaqus44, and (d) Abaqus104. Results are given in MPa (see
Fig. 9(c) for reference).
20 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24

Fig. 13. Evolution of the frequencies (a–c) and the real parts (d–f) of the eigenvalues against the friction coefficient l for the (a and d) 1.9 kHz, (b and e) 3.8 kHz and (c and f)
4.7 kHz instabilities.

Table 4 contact interface is estimated. First of all, non-linear static analysis

Mean error on frequency for each pair of modes on the whole friction range between for a friction coefficient l ¼ 0:5 is computed. Results of the con-
Abaqus104/212 and the Abaqus reference model.
straint fields are presented for both the reference model and the
Instability number 1 (%) 2 (%) 3 (%) reduced model Abaqus212 on Fig. 9(e)–(l), respectively. The colour
Abaqus104 0.713 0.289 0.536 scales given in Fig. 9(a)–(d) are based on the reference model and
Abaqus212 0.563 0.247 0.415 according to the finite element model units, constraint value is
given in MPa. It can be observed that the constraint fields for the
reference model and the reduced model Abaqus212 are very simi-
lar. Even if the disc mesh on the reduced model differs from the
Table 5
Min/Max error on frequency for each pair of modes between Abaqus104/212 and the reference models due to the imposed coincident nodes a good glo-
Abaqus reference model. bal coincidence in the constraint field can be noticed for each inter-
face of the brake system. Then, numerical results at the inner pad
Instability number 1 (%) 2 (%) 3 (%)
interface for two different friction coefficients l ¼ 0:2 and l ¼ 0:8
Abaqus104 0.332/1.688 0.073/0.459 0.208/0.779
are also indicated in Fig. 10. Whatever the friction coefficient and
Abaqus212 0.152/1.821 0.149/0.411 0.064/1.293
the evolution of the associated non-linear static equilibrium, the
constraint fields that are calculated via the reduced model Aba-
qus212 are consistent with those of the reference. Results at the
is shared by the nodes of this facet. With the coincident mesh the outer pad, outer and inner disc surfaces (not presented in this
projected node falls at the exact position of a node of the master study for l ¼ 0:2 and l ¼ 0:8) corroborate these conclusions.
surface. In this last case, the transmitted load is shared by the coin- Finally, it should be interesting to observe the influence of the
cident master node and all of the master surface nodes that share number of reduced contact nodes on the constraint fields. Fig. 11
an adjacent surface facet with that node, as illustrated in Fig. 8. In gives the constraint fields for the outer surface of the disc with
one way we can say that the coincident node smooths the contact the four reduced models (i.e. Abaqus12, Abaqus44, Abaqus104
force distribution on the master surface. and Abaqus212) with a friction coefficient l ¼ 0:5.
In order to be able to compare the constraint fields at the con- If the number of reduced contact nodes is very small (i.e. 6 and
tact interface (i.e. Von Mises stress), the contact force calculated at 22 nodes per pad for Abaqus12 and Abaqus44), difference with the
the reduced contact node of the pad side is then distributed reference model is noticeable (see Fig. 11(b) and (c) for Abqus12
through the coupling element. Then, the constraint fields at the and Abaqus44 in comparison with Fig. 11(a) for the reference).
 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24 21

Fig. 14. Evolution of the frequencies (a–c) and the real parts (d–f) of the eigenvalues against the friction coefficient l for the (a and d) 1.9 kHz, (b and e) 3.8 kHz and (c and f)
4.7 kHz instabilities. The same Super-Element generated at l ¼ 0:5 is used for the whole friction coefficient range.

Constraint peaks are concentrated at the coincident nodes on the analysis is given in Fig. 12. Moreover, evolution of the frequencies
disc surface. For the reduced models Abaqus12 and Abaqus44, and real parts of the three unstable modes are drawn for the refer-
the distance between two reduced contact nodes is larger than ence model and the four reduced models on Fig. 13.
the mesh size of the disc, thus contact force are only computed It clearly appears that the reduced model Abaqus12 is not suffi-
on a really local zone around each slave node projection. Then, cient to predict instabilities beyond 4 kHz. Indeed, we observe that
increasing the number of reduced contact nodes (i.e. the reduced the unstable mode at 3.8 kHz disappears for l ¼ 0:4 (as indicated
models Abaqus104 or Abaqus212) allows a better distribution of in Fig. 13(e)) which suggests that a change in contact status can com-
the constraint field. So, the constraint field estimated by using pletely modify the characteristics of the brake system. This result is
these two reduced models are in perfect agreement with the refer- not surprising because of the inconclusive results of this reduced
ence model, as illustrated in Figs. 11(d) (for Abaqus104) and 9(k). model to correctly estimate the static states, as previously seen in
For the reader comprehension, results for the reference model the previous section. Concerning the second reduced model Aba-
are given in Fig. 9(g). Moreover, the other numerical results (not qus44, the three instabilities of interest (at 1900 Hz, 3800 Hz and
presented in this study) at the inner/outer pads and the inner/outer 4700 Hz) are predicted, even if a little delay for the first and second
disc surfaces for the four reduced models (i.e. Abaqus12, Abaqus44, unstable modes (at 1.9 Hz and 3.9 kHz) can be observed. Moreover,
Abaqus104 and Abaqus212) and for the three friction coefficients limitation of this reduced model Abaqus44 can be highlighted by
(i.e. l ¼ 0:2; l ¼ 0:5 and l ¼ 0:8) lead to the same conclusion. seeing evolution of the third unstable mode for high friction coeffi-
cients: indeed a decrease in frequency and a margin in real part eval-
4.3. Abaqus reduced models: stability analysis uation are clearly observed in Fig. 13(c) and (f), respectively.
Finally, the two last reduced model (i.e. Abaqus104 and Aba-
In the previous section, it was illustrated that it is possible to qus212) correctly approximate the three unstable modes on the
drastically reduce the number of contact nodes at the frictional frequency range of interest for both the value of real part and the
interface while being able to estimate the contact states and the value of the associated frequency, as illustrated in Figs. 12 and
constraint field at the static non-linear equilibrium. 13. Increasing the number of reduction nodes between Abaqus104
The results of stability analysis between the reference Abaqus and Abaqus212 does not lead to significant improvement. In order
model and the four reduced models Abaqus12, Abaqus44, Aba- to have a more detailed comparison of these two reduced models
qus104 and Abaqus212 are discussed. A classical complex eigen- Abaqus104 and Abaqus212, the mean error on the three instabili-
value analysis, with a friction coefficient varying from l ¼ 0:1 to ties of interest is estimated for all the values of the friction coeffi-
l ¼ 0:9 with a step size of 0.1, is performed and we focus on the cient in the range of interest (i.e. from l ¼ 0:1 to l ¼ 0:9 with a
instabilities in the 0–6 kHz range. a global overview of the stability step size of 0.1). The mean error is defined by
22 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24

6 kHz) in order to predict all the instabilities (in terms of frequency

and real part of unstables modes) while allowing to reduce signif-
icantly the number of contact nodes at the disc/pad interface.

4.4. Super-Element assembly: stability analysis

The previous results confirmed that reducing the disc/pad con-

tact interface with 104 contact nodes is sufficient to predict the
main instabilities on the 0–6 kHz range of interest. It is worth
remembering that the number of reduction nodes at the frictional
interface can vary with the system under study due to the mode
shape of unstable modes and the frequency range of interest.
Now we propose to illustrate the efficiency of the second part of
the global strategy based on the Super-Element assembly. As a
reminder, all these developments have been previously explained
in Section 3.
First of all, based on the retained reduced model Abaqus104, a
Super-Element (SE) for a friction coefficient of l ¼ 0:5 at the
disc/pad interface is generated. Real modes over 0–12 kHz are kept
for the Super-Element creation so that its validity can be assumed
over 0–6 kHz, our frequency range of interest. It represents a total
of 135 modes. Once the SE created, assembled in Matlab with con-
tact force matched and linearized around the non-linear static
equilibrium, a complex eigenvalue analysis is undertaken. The
results between the Super-Element assembly (SE assembly) and
the reduced model Abaqus104 are compared in Fig. 14. It can be
noticed that the stability analysis is performed on the whole fric-
tion range of interest (i.e. from l ¼ 0:1 to l ¼ 0:9) with only a
Super-Element created at l ¼ 0:5. It is observed that the global
reduction based on the Super-Element assembly is not able to
reproduce correctly the prediction of unstable vibration modes
(both in terms of frequency and real parts). For example it can be
seen that for high friction coefficient, the third instability (around
Fig. 15. Outer pad/Caliper contact status evolution on Abaqus reference model for 4.7 kHz) disappears with the reduced model via SE assembly. These
friction coefficient value of 0.2 (a), 0.5 (b) and 0.8 (c). Blue colour is for open contact, results shows that a Super-Element remains valid only at the vicin-
the others are for closed contact sticking (red) or slipping (green). (For interpre-
tation of the references to colour in this figure legend, the reader is referred to the
ity of the non-linear static equilibrium point. More particularly,
web version of this article.) varying the friction coefficient at the disc/pad interface changes
the static equilibrium on the Abaqus reference model as indicated
in Fig. 15. Such changes in the non-linear static equilibrium are not
red ref
X9 X 2
1 jf lj ;i  f lj ;i j taken in account if the same Super-Element created at l ¼ 0:5 is
etot ¼ ref
ð12Þ used for the whole stability analysis. This can explain the differ-
j¼1 i¼1
M f lj ;i ences observed between the reduced model with SE assembly
and the reduced model Abaqus104.
where lj corresponds to the value of the friction coefficient given by
Considering these first previous results based on the Super-Ele-
lj ¼ 0:1  j. Considering one instability, f ref red
lj ;i (respectively f lj ;i ) is the ment assembly, we propose to generate a Super-Element for each
frequency of the ith of the two modes that couple at the friction
step of the friction coefficient (i.e. from l ¼ 0:1 to l ¼ 0:9 with a
value l for the Abaqus reference model (respectively reduced
step size of 0.1). Then a complex eigenvalue analysis is performed
model). M stands for the number of friction coefficient for which
for each friction coefficient by using the associated Super-Element
a complex eigenvalue analysis is computed for the stability analysis.
assembly. As illustrated in Fig. 16, results for SE assembly of both
The minimum and maximum errors are also defined by
frequencies and real parts of the three unstable modes are in
0 red ref
1 agreement fit with those of the reduced model Abaqus104. Good
X lj ;i  f lj ;i j
2 jf
emin ¼ min @ A ð13Þ enough with the evolution of the results on the corresponding Aba-
lj ref
i¼1 f lj ;i qus104 model, even if a sudden rise in frequency of the third
0 red ref
1 unstable mode starting at l ¼ 0:6 can be noticed.
X lj ;i  f lj ;i j
2 jf
@ A Finally, the mean error (given in Eq. (12)), the minimum and
emax ¼ max ref
ð14Þ
lj
i¼1 f lj ;i maximum errors (given in Eqs. (13) and (14), respectively) are cal-
culated for the SE assembly. The two strategies defined for SE
All these results are summed up on Tables 4 and 5. Whatever assembly (by using only one Super-Element for the whole stability
the instability considered the error does not exceed two percent, analysis or by estimating one Super-Element for each friction coef-
the performance of the contact reduction is clearly highlighted ficient) are compared to the reduced model Abaqus104. We recall
here. Considering the mean error, the convergence of the reduction that this reduced model Abaqus104 was used for the SE generation.
strategy is observed and the remaining error can be explained by Results are given in Tables 6 and 7. It clearly shows the interest of
the boundary effect of the distribution. In conclusion, results on creating one SE for each friction coefficient value. With this strat-
stability analysis show that the reduced model Abaqus104 fits per- egy the error never exceeds five percent whatever the unstable
fectly on the whole frequency range of interest (i.e. between 0 and mode of interest. In conclusion, these results illustrate the
 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24 23

Fig. 16. Evolution of the frequencies (a–c) and the real parts (d–f) of the eigenvalues against the friction coefficient l for the (a and d) 1.9 kHz, (b and e) 3.8 kHz and (c and f)
4.7 kHz instabilities. A new Super-Element is generated at each l for complex eigenvalue analysis on SE assembly.

Table 6 This strategy allows to predict correctly both the non-linear sta-
Mean error on frequency for each pair of modes on the whole friction range between tic equilibrium (i.e. the contact states and the constraint field) and
SE assembly and the Abaqus104 model.
the stability analysis at this equilibrium. It is shown that the Super-
Instability number 1 (%) 2 (%) 3 (%) Element generation needs to be performed for each set of input
Matlab 104 single SE 4.829 0.543 1.872 parameters because of the possible influence of the static equilib-
Matlab 104 SE/l 0.461 0.713 2.276 rium position on the stability analysis.
Based on these results, future work on the calculation of self-
excited vibrations can be investigated by using reduced models
via Super-Element generation.
Table 7
Min/Max error on frequency for each pair of modes between SE assembly and the
Abaqus104 model. Acknowledgments
Instability number 1 (%) 2 (%) 3 (%)
This work was achieved within Stellab program. The authors
Matlab 104 single SE 0.477/12.818 0.223/0.873 0.081/5.131
would like to thank the PSA Peugeot Citroën staff for their
Matlab 104 SE/l 0.355/0.519 0.587/0.918 0.778/4.482
assistance.

efficiency of the global strategy based on both nodal interface References

reduction and Super-Element generation.
[1] Kinkaid NM, O’Reilly OM, Papadopoulos P. Automotive disc brake squeal. J
Sound Vib 2003;267:105–66.
5. Conclusion [2] Ibrahim RA. Friction-induced vibration, chatter, squeal, and chaos part 1:
mechanics of contact and friction. Am Soc Mech Eng Appl Mech Rev
1994;47(7):209–26.
The global reduction strategy that is based on a nodal reduction, [3] Ibrahim RA. Friction-induced vibration, chatter, squeal, and chaos part 2:
Super-Elements generation and a Craig and Bampton modal reduc- dynamics and modeling. Am Soc Mech Eng Appl Mech Rev 1994;47(7):227–63.
[4] Ouyang H, Mottershead JE, Brookfield DJ, James S, Cartmell MP. Methodology
tion at the frictional interface, is proposed for an industrial non-lin-
for the determination of dynamic instabilities in a car disc brake. Int J Veh Des
ear brake system model. One of the advantage of the proposed 2000;23(3):241–62.
strategy is based on an efficient contact reduction method for the [5] Ouyang H, Nack W, Yuan Y, Cocheteux F. Numerical analysis of automotive
disc/pad interface that allows a Craig–Bampton reduction to be disc brake squeal: a review. Int J Veh Noise Vib 2005;1(3):207–31.
[6] Sarrouy E, Dessombz O, Sinou J-J. Piecewise polynomial chaos expansion with
used for further stability analysis. Final number of variables on this an application to brake squeal of linear brake system. J Sound Vib
reduced model is about 1500. 2013;332:577–94.
24 O. Fazio et al. / Applied Acoustics 91 (2015) 12–24

[7] Heussaff A, Dubar L, Tison T, Watremez M, Nunes RF. A methodology for the [20] AbuBakar AR, Ouyang H. Complex eigenvalue analysis and dynamic transient
modelling of the variability of brake lining surfaces. Wear 2012;289:145–59. analysis in predicting disc brake squeal. Int J Veh Noise Vib 2006;2:143–55.
[8] Oberst S, Lai JCS. Statistical analysis of brake squeal noise. J Sound Vib [21] AbuBakar AR, Ouyang H. A prediction methodology of disk brake squeal using
2011;330(12):2978–94. complex eigenvalue analysis. Int J Veh Des 2008;46(4):416–35.
[9] Massi F, Baillet L, Giannini O, Sestieri A. Brake squeal: linear and nonlinear [22] Massi F, Berthier Y, Baillet L. Contact surface topography and system dynamics
numerical approaches. Mech Syst Signal Process 2007;21(6):2374–93. of brake squeal. Wear 2008;265(11-12):1784–92.
[10] Akay A, Giannini O, Massi F, Sestieri A. Disc brake squeal characterization [23] Sinou J-J, Loyer A, Chiello O, Mogenier G, Lorang X, Cocheteux F, et al. A global
through simplified test rigs. Mech Syst Signal Process 2009;23(8):2590–607. strategy based on experiments and simulations for squeal prediction on
[11] Sinou J-J. Transient non-linear dynamic analysis of automotive disc brake industrial railway brakes. J Sound Vib 2013;332(20):5068–85.
squeal – on the need to consider both stability and non-linear analysis. Mech [24] Vermot des Roches G. Frequency and time simulation of squeal instabilities –
Res Commun 2010;37(1):96–105. application to the design of industrial automotive brakes, PhD thesis. Ecole
[12] Coudeyras N, Nacivet S, Sinou J-J. Periodic and quasi-periodic solutions for Centrale Paris; 2011.
multi-instabilities involved in brake squeal. J Sound Vib 2009;328(4):520–40. [25] Loyer A, Sinou J-J, Chiello O, Lorang X. Study of nonlinear behaviors and modal
[13] Sinou J-J, Coudeyras N, Nacivet S. Study of the non-linear stationary dynamic reductions for friction destabilized systems. application to an elastic layer. J
of single and multi-instabilities for disk brake squeal. Int J Veh Des Sound Vib 2012;331(5):1011–41.
2009;51(1):207–22. [26] Craig RJ. A review of time-domain and frequency domain component mode
[14] Festjens H, Chevallier G, Renaud F, Dion J-L, Lemaire R. Effectiveness of synthesis methods. Int J Anal Exp Modal Anal 1987;2(2):59–72.
multilayer viscoelastic insulators to prevent occurrences of brake squeal: a [27] Brizard D, Chiello O, Sinou J-J, Lorang X. Performances of some reduced bases
numerical study. Appl Acoust 2012;73(11):1121–8. for the stability analysis of a disc/pads system in sliding contact. J Sound Vib
[15] Oberst S, Lai JCS. Chaos in brake squeal noise. J Sound Vib 2011;330(5): 2011;330(4):703–20.
955–75. [28] Villard P, Nacivet S, Sinou J-J. Superelement reduction of industrial finite
[16] Oberst S, Lai JCS, Marburg S. Guidelines for numerical vibration and acoustic element brake system for a constrained harmonic balance method. In: INTER-
analysis of disc brake squeal using simple models of brake systems. J Sound NOISE and NOISE-CON congress and conference proceedings; 2012. p. 3142–
Vib 2013;332(9):2284–99. 51.
[17] Soobbarayen K, Besset S, Sinou J-J. Noise and vibration for a self-excited [29] Abaqus Analysis User’s Manual. [chapter 31 Constraints].
mechanical system with friction. Appl Acoust 2013;74(10):1191–204. [30] Coudeyras N, Sinou J-J, Nacivet S. A new treatment for predicting the self-
[18] Soobbarayen K, Besset S, Sinou J-J. A simplified approach for the calculation of excited vibrations of nonlinear systems with frictional interfaces: the
acoustic emission in the case of friction-induced noise and vibration. Mech constrained harmonic balance method, with application to disc brake squeal.
Syst Signal Process 2015;50:732–56. J Sound Vib 2009;319(3):1175–99.
[19] Renaud F, Chevallier G, Dion J-L, Taudiere G. Motion capture of a pad measured [31] Sinou J-J, Thouverez F, Jezequel L. Methods to reduce non-linear mechanical
with accelerometers during squeal noise in a real brake system. Mech Syst systems for instability computation. Arch Comput Methods Eng: State Art Rev
Signal Process 2012;33:155–66. 2004;11(3):257–344.