Mechanical Systems and Signal Processing 28 (2012) 574–584

Contents lists available at SciVerse ScienceDirect

Mechanical Systems and Signal Processing

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

Measured and estimated friction interface parameters

in a nonlinear dynamic analysis
C.W. Schwingshackl n, E.P. Petrov, D.J. Ewins
Imperial College London, Mechanical Engineering, Exhibition Road, SW7 2AZ London, UK

a r t i c l e in f o abstract

Article history: The modelling of nonlinear friction contact interfaces in structural dynamics currently
Received 29 May 2011 attracts large research interest since the accurate characteristics of the friction interfaces
Received in revised form are required in the nonlinear dynamic analysis of assembled structures. For an accurate
6 October 2011
representation of friction contact interfaces it is important to ensure that a sufficient
Accepted 6 October 2011
number of parameters are provided to characterise all aspects of the friction contacts that
Available online 26 October 2011
the measured values are accurate and reliable, and that the contact parameters are
Keywords: interpreted and used correctly in the numerical modelling of the contact interfaces.
Friction input parameters This investigation focused on three major parameters used in the nonlinear friction
Measurement
contact analysis: (i) the friction coefficient, (ii) the tangential contact stiffness, and
Repeatability
(iii) the normal load distribution in the contact. The accuracy of the measured friction
Test rig modelling
input parameters is evaluated and possibilities to obtain reliable normal load distribu-
tions for a successful prediction of the measured frequency response function are
investigated. Average friction input parameters, based on a large set of experimental
data, are introduced, and their effect on the nonlinear resonance response is discussed.
& 2011 Elsevier Ltd. All rights reserved.

1. Introduction

The simulation of the nonlinear dynamics of structures has become a major point of interest, since although the linear
response predictions today have reached a very high level of accuracy, increasingly it is evident that real engine structures are
significantly influenced by nonlinear characteristics. The friction contact interfaces are a major source of the nonlinearity and to
model these, two major topics must be addressed: (i) the development of an accurate numerical method to simulate the
behaviour between two contact surfaces, and (ii) an accurate measurement of the friction input parameters.
The ongoing research effort in the development of nonlinear structural dynamic models has resulted in a range of
proposed analysis methods to solve the problem. The modelling approaches either address specific problems, such as
underplatform dampers [1,2] blade roots [3,4], or shroud contacts [5,6] or allow a more general calculation of the
dynamics for structures with friction interfaces [7–11]. The methods allow a detailed numerical description of the contact
interfaces and lead to accurate nonlinear response predictions of the system, but they are only as accurate as their input
parameters.
Several test rigs have been developed to extract the dynamic friction contact properties of material pairs that can be
used for the nonlinear friction parameter models. Depending on the original intention of the rig, a range of different
contact conditions have been investigated. Some rigs provide high sliding velocities and small displacements with an

n
Corresponding author. Tel.: þ 44 207 594 1920; fax: þ44 207 594 5709.
E-mail addresses: [email protected] (C.W. Schwingshackl), [email protected] (E.P. Petrov), [email protected] (D.J. Ewins).

0888-3270/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ymssp.2011.10.005
 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584 575

Nomenclature N0 normal load

Fexi excitation force
m friction coefficient Ffri friction force
s, smax standard deviation and maximum value a contact width
kt tangential contact stiffness x,y contact coordinates
kn normal contact stiffness xrel,zrel relative displacement in x, y, and z

added possibility of high temperature testing [12–15]; some have been originally designed for fretting measurements, and
are now used to extract dynamic friction coefficients at high loads, large displacements, and relatively low velocities
[16,17]; others rely on the friction interface of real joint configurations (e.g. bolted joints) to extract the friction properties
[18–20]. In order to obtain accurate and reliable predictions for engineering applications, it must be ensured that the
correct parameters are measured in the rig to describe the problem that their values are identified accurately and that they
are interpreted appropriately in the analysis. This becomes of particular importance if no accurate measured values are
available, and the analysis must rely on estimated friction input parameters.
This work addresses the abovementioned issues by identifying the parameters required by the nonlinear analysis, evaluating
their measurement reliability, and assessing the agreement between simulated and experimental results of a nonlinear frequency
response. A set of averaged friction contact parameters, based on the data obtained in past measurements, are presented, to
provide reliable estimates if no accurately measured data is available.

2. The nonlinear friction interface model

The in-house nonlinear structural dynamics analysis code used in this investigation, FORSE, is based on a multi-
harmonic representation for steady-state response and large-scale realistic friction interface modelling of bladed discs.
Major features of the methodology are described in Refs. [9–11] and in this paper only a short overview of the analysis
method is provided. The flowchart of the calculations performed with the code is presented in Fig. 1(a). It requires the
dynamic properties of the linear components of the structure, which are obtained from a finite element analysis as input,
together with an accurate contact interface description. The contact interface elements developed in Ref. [10] are used to
model the nonlinear interactions at the contact interfaces and an advanced multiharmonic balance solver is applied to
calculate the amplitudes of the multiharmonic displacement over the frequency range of interest. The parameters required
in the nonlinear analysis can be separated into three main groups: (i) the friction interface parameters that describe the
material properties of the contact area, (ii) the modelling parameters that define the nonlinear model and its excitation,
and (iii) the analysis parameters, which control the accuracy and speed of the calculation. The measured interface
parameters will be the main focus of this paper whereas more information on additional nonlinear analysis parameters
and their influence on the nonlinear response can be found in Ref. [21].
The generic friction models (Refs. [9–11]) used in the analysis are illustrated in Fig. 1(b) for the two dimensional case.
They require four input parameters to represent the contact condition in the model: (i) the friction coefficient, m, limiting
the maximum force that can be transmitted in an element before it starts to slip; (ii) the tangential contact stiffness, kt,
describing the elastic flexibility of the contact area in the sliding direction; (iii) the normal contact stiffness, kn, describing

Contact interface
FE models of linear
description and
components
parameters
N0 Ffri
x rel
FORSE µN 0
zrel
High accuracy Contact interface kt
model reduction elements

x rel
kn 1
Multiharmonic balance
solver with continuation kt µ

Ffri
Amplitudes of multiharmonic
displacement over frequency
range of interest

Fig. 1. (a) Scheme of the forced response analysis and (b) the generic friction model.
576 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584

the flexibility of the contact layer in the out of plane direction; and (iv) the normal load, N0, defining together with the
friction coefficient, m, the limiting value of the friction force.
Values for the friction coefficient, m, and the tangential contact stiffness, kt, are available from measurements and can be
used directly in the analysis. The normal load distribution in the contact, N0, can either be obtained experimentally, or it
can be based on a finite element solution. The normal contact stiffness, kn, can be measured with an ultrasonic setup [22].
There are research attempts under way to predict the friction interface properties analytically [23,24], but at the current
time the experimental data are still the basis for an analysis.

3. The friction rig

The friction test rig at Imperial College London has been designed to extract the friction coefficient, m, and the
tangential contact stiffness, kt, for the nonlinear analysis. A short discussion of the rig will be provided here and more
details can be found in the literature (see Refs. [12,13]). The underlying principle of the measurement in the friction rig is
to provide a one-dimensional (1D) relative sliding motion, xrel, between two material samples, with a known normal load,
N0, applied across the contact. A simultaneous measurement of the resulting relative displacement and the transmitted
friction force, Ffri, allows the extraction of the friction interface parameters.
Fig. 2 shows the setup of the friction rig, which uses two cylindrical specimens with wedge-shaped ends (see the inset).
The ends have normally 1 mm wide flat surfaces at their apex and the orthogonal mounting of the two specimens opposite
each other provides a nominal contact area of 1 mm2 during tests (other width and contact areas have also been used). One
specimen is stationary and is connected via the ‘static arm’ to a heavy reaction mass, which is considered to be very stiff,
and the other is connected via the ‘moving arm’ to a drive mass on a flexible support. The drive mass is excited with a
shaker at 100 Hz to reproduce dynamic conditions typically experienced in aircraft engine components. The amplitude of
the relative motion ranges from 20 to 40 mm. The specimens are pressed together with a loading system, which can
provide normal loads, N0, between 5 and 75 N. By changing the contact areas, contact pressures up to 350 MPa have been
investigated. A furnace allows measurements up to 1050 1C. The relative displacement, xrel, between the sliding contacts is
measured with a Scanning Laser Doppler Vibrometer (SLDV) at points very close to the contact areas, and the transmitted
friction force, Ffri, is acquired by two force transducers placed between the static arm and the reaction block.
Plotting the friction force, Ffri, against the relative displacement, xrel, as shown in Fig. 3 leads to a hysteresis loop
describing the friction interface behaviour. The hysteresis loop shows three distinct stages of the stick–slip motion of the
friction contact interface: (i) the ‘stick’ stage occurring immediately after the reversal of the direction of motion, where
linear displacement–force dependency due to the elastic deformations of the rough contact surface is occurring, (ii) the
‘microslip’ stage, which represents the transition from stick to slip where some parts of the contact area are slipping while
others are still stuck, and (iii) the ‘macroslip’ stage during which slip occurs over the entire contact interface and the
transmitted friction force remains approximately constant. The stick stage can be used to extract the tangential contact
stiffness, kt, by fitting a straight line to its lower section, where the displacement originates fully from the elastic
deformation in the contact. The distance between the horizontal lines of the hysteresis loop during macroslip provides the
friction coefficient value, m ¼Ffri/2N0. An additional feature that can be sometimes observed in measured hysteresis loops
is an increase in the friction force during macroslip. The physical origin of this behaviour is not yet fully understood,
but recent investigations indicate that at least part of it may be caused by the wear footprint developing in the contact.

Moving N0
specimen Alternating
Force transducer SLDV
Xrel

Static
Normal load arm specimen Ffri
n
N0
Drive mass
N0 on leaf springs
Static Moving Shaker
Reaction mass Arm Arm Excitation

Static
support N0
N0 2
Normal load hanger system
2

Fig. 2. The friction rig.

 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584 577

Measured hysteresis loop

50

40

30

20

10

Force [N]
kt= Δffri Ffri
0 Δx μ=
2N 0
-10 ΔFfri

-20 Δx

-30

-40

-50
-0.015 -0.01 -0.005 0 0.005 0.01 0.015
Displacement [mm]

Fig. 3. Typical hysteresis loop and associated parameters.

Specimen A -reassembly:
Friction coefficient, µ Friction coeffecient, µ, of three
0.7 nominally idencial materials
First assembly 1.0
Second assembly
0.65 Third assembly 0.8

0.6 0.6
µ

0.4
0.55
0.2
0.5 0.0
0 5 10 15 20 Specimen B Specimen C Specimen D
acquisition time [min]
Specimen A -reassembly:
Tang contact stiffness, kt, of three
Tangential contact stiffness, kt
nominally idencial materials
4E+07 3.5E+07
4.E+07 3.0E+07
kt [N/m/mm ]
2
kt [N/m/mm2]

3.E+07 2.5E+07
3.E+07 2.0E+07
1.5E+07
2.E+07 First assembly 1.0E+07
2.E+07 Second assembly
Third assembly
5.0E+06
1.E+07 0.0E+00
0 5 10 15 20 Specimen B Specimen C Specimen D
acquisition time [min]

Fig. 4. Friction interface parameters: (a) single specimen pair: repeatability of measured friction input parameters after reassembly and (b) parameters
for three nominally identical pairs.

An average of ten successive hysteresis loops is used (a single loop can be seen in Fig. 3) to determine the friction interface
parameters.
The extracted parameters are strongly dependent on the shape of the measured hysteresis loops and in order to assess the
accuracy of the measured input parameters, a study has been carried out to demonstrate the reliability of the values provided.
For this purpose, the same specimen pair (A) was fitted, tested, and removed several times to check the repeatability;
nominally identical material pairs (B–D) were tested to investigate parameter variations; and misalignments were introduced
intentionally to the rig setup to evaluate their influence on the measurement accuracy (specimen E). A single measurement was
then carried out to provide a set of values for the friction coefficient, m, and the tangential contact stiffness, kt. The normal
contact stiffness, kn, which allows for a normal load variation in the model caused by the vibrations, is of less importance for the
test rig simulation, since the normal load of the rig is kept constant throughout the experiment.
Fig. 4(a) shows the results of a repeatability study for the test rig, where the same specimen pair (A) was removed and
refitted three times to the holder to measure values of the friction coefficient, m, and the tangential contact stiffness, kt, for
each reassembly. It can be seen that a certain running-in period (in this case,  15 min) is required, during which time
the surface undergoes wear and oxidation, before reaching a state when the contact surface conditions stabilise and
the extracted friction parameters converge towards constant values. After each refit of the specimen pair, the contact
578 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584

Specimen E - out of plane rotation Specimen E - out of plane rotation

Friciton coefficient, µ Tangential contact stiffness, kt
0.7 4.0E+07
0 deg 3.5E+07
0.65

kt [N/m/mm ]
2 deg

2
3.0E+07
4 deg
0.6 2.5E+07
µ
2.0E+07 0 deg
0.55 2 deg
1.5E+07
4 deg
0.5 1.0E+07
0 5 10 15 0 5 10 15
acquisition time [min] acquisition time [min]

Fig. 5. Dependence of friction coefficient, m, and tangential contact stiffness, kt, on the out of plane rotation angle of the contact areas.

condition was reset, as the realignment of the contact area is not perfectly repeatable, and the parameters require another
running-in period to converge once more towards the final stable values, which are nearly identical for all three repeats.
These converged results highlight the good repeatability of the measurement procedure.
Three nominally identical specimen pairs, B–D, of a different material than the previously used specimen A (Fig. 4(a)),
were tested under the same load and slip conditions (results shown in Fig. 4(b)), in order to establish the dependence of
the friction parameters on the specimen variability, possibly caused by slight variations in the bulk material or the contact
surface finish. The three converged friction interface parameters of specimens B–D show a good agreement for the friction
coefficients, m. Slightly more scatter is found for the tangential contact stiffness, kt, but it remains within the known overall
variations of the test rig and can therefore be considered a good and repeatable result. It should be noted here once more
that the results in Fig. 4(a) (specimen A) and (b) (specimens B–D) were obtained for two different high temperature
materials under varying test conditions and, for that reason, their absolute values are not the same.
To improve the confidence in the measured data further, an additional investigation into possible alignment errors for the
contacting planes of the specimens was carried out. Several parameters for in and out of plane alignment of the contact areas
were investigated, with Fig. 5 showing the results for an out of plane rotation of the contact areas for a new specimen E.
In general, it could be shown that even large misalignments of 4–51 lead to small variations of only  4% for the friction
coefficient, m, and  12% for the tangential contact stiffness, kt. These deliberately introduced alignment errors were
unrealistically large, and would be detected and eliminated by a standard alignment check before each measurement,
making the rig robust against alignment issues.

4. Averaged friction parameters

In case that no measured friction input parameters are available for the nonlinear analysis, estimated values are
required. To improve the quality of the estimated friction input parameters, a large selection of past measurements from
Imperial College London has been collected to identify overall trends and provide more reliable average values. It must be
noted however that not all material combinations behave in the same way under different test conditions, and only
dedicated measurements will reveal the accurate friction interface behaviour.
Experimental values for 15 unlubricated high temperature materials are available, at 17 different contact pressures
(5–350 MPa), 11 temperatures (20–1050 1C), and four contact areas (0.46–1 mm2). Fig. 6 shows the distribution of the data
sets over the temperature. A larger number of measurements are available for the friction coefficient, m, especially at room
temperature and 600 1C, whereas slightly fewer values have been extracted in the past for the tangential contact stiffness, kt.
The results cover a large range of test conditions, and therefore a combination of the values can give a good indication of the
overall behaviour of friction input parameters for high temperature materials, to improve the suggested average values for the
nonlinear analysis.
Fig. 7 introduces the friction parameters for 15 different unlubricated high temperature materials (Mat1-15) tested over
a wide temperature and pressure range to improve the quality of the estimates. Fig. 7(a) shows the dependence of the
friction coefficient, m, on temperature. It has been previously observed [15] that the friction coefficient decreases with
a rise in temperature, but Fig. 7(a) highlights how consistent the change is over a large number of different materials.
The scatter at room temperature is quite large, mainly due to the fact that the largest amount of data points, measured
at different pressure settings, is available there, with its mean value around m ¼0.66 and a standard deviation of 0.08.
At 200 1C the mean friction coefficient has dropped to nearly m ¼0.3, and it remains between 0.2 and 0.3 for all further
temperature increases with a standard deviation not larger than 0.05 (see Table 1 for the average values over the
temperature). It must be mentioned here that the quoted standard deviation values must be considered with some
caution, since they are based on a relatively small amount of data (see Fig. 6).
Fig. 7(b) is based on the same data set as Fig. 7(a), but it shows the dependence of the friction coefficient, m, on the
contact pressure at 20 and 600 1C. Some of the contact pairs have a dependence on the contact pressure [26] but the
overall trend of all the included materials is relatively constant over the pressure range for both temperature cases.
 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584 579

Available material pairs for each temperature

14

Material pairs
12 Friction coefficient
10
8 Tangential contact stiffness
6
4
2
0
20 50 200 350 400 600 750 800 950 1000 1050
Temperature [C]

Fig. 6. Available material pairs for each temperature.

Friction coefficient, µ, over temperature Friction coefficient, µ, over contact

15-350MPa values included pressure for 20 and 600°C
1.0 1.0
0.9 0.9
0.8 0.8

Friction coefficient, µ
Friction coefficient, µ

0.7 0.7
0.6 0.6
0.5 0.5
20°C
0.4 0.4
600°C
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
0 200 400 600 800 1000 1200 0 100 200 300 400
Temperature [C] Contact pressure [Mpa]

Fig. 7. Friction coefficients, m, for 15 different high temperature materials in varying test combinations as function of (a) temperature range and
(b) contact pressure for 20 and 600 1C.

Table 1
Average friction coefficient, m, over temperature.

Temp. (C) 20 50 200 350 400 600 750 800 950 1000 1050
Mean, m 0.66 0.46 0.35 0.33 0.33 0.24 0.30 0.24 0.27 0.27 0.31
Std. dev., s 0.08 0.09 0.05 0.02 0.05 0.04 0.05 0.02 0.02 0.03 0.05

Tangential contat stiffness over temperature Tangential contat stiffness over contact
for all contact pressures pressure for 20 and 600°C
6.0E+07 6.0E+07
Tangential contact stiffness kt [N/m]

Tangential contact stiffness kt [N/m]

5.0E+07 5.0E+07

4.0E+07 4.0E+07

3 0E+07
3. 3.0E+07

2.0E+07 2.0E+07

1.0E+07 1.0E+07

0.0E+00 0.0E+00
0 200 400 600 800 1000 1200 0 100 200 300 400
Temperature [C] Contact pressure [Mpa]

Fig. 8. Tangential contact stiffness, kt, for different high temperature materials in varying test combinations as function of over (a) temperature range and
(b) contact pressure for 20 and 600 1C.

Based on the evaluation of the available data set for the measured friction coefficients, m, of unlubricated high
temperature materials, a pressure independent average value of m ¼0.66 at room temperature and m ¼0.25 above 200 1C is
suggested for a nonlinear analysis, if no accurately measured input parameters can be provided. It should be mentioned
here that only an accurate measurements will show the correct behaviour of a certain material pairing, and the suggested
values should therefore be considered as guidelines only.
11 different material sets are available for the dependence of the tangential contact stiffness, kt, on the specimen
temperature in Fig. 8(a), with most data points measured at 20 and 600 1C. A strong scatter of the tangential contact
580 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584

stiffness, kt, between 1 and 5e7 N/m (smax ¼ 1.6e7 N/m) is present, which does not seem to be affected by a change in the
test temperatures.
The variation of the tangential contact stiffness, kt, over the contact pressure can be seen in Fig. 8(b). Similarly to the
friction coefficient, some materials show a dependence of kt on the contact pressure, while others appear to be
independent of it, but all materials remain within the 1  5e7 N/m range (smax ¼1e7 N/m). Regardless of the scatter
experienced, a pressure-independent mean tangential contact stiffness of 3e7 N/m with a standard deviation of 9e6 N/m is
suggested as an estimated input for an analysis, if no experimentally measured data are available. This value represents a
much more realistic input than the regularly used approach to tune the analysis with an artificial tangential contact
stiffness to fit a measured response.
A comparison of the two suggested averaged friction input parameters to some values from the literature can be found
in Table 2. All these values were obtained with different experimental setups and cover a wider range of material and test
conditions. Nevertheless a good agreement between the suggested averaged parameters and the values from the literature
can be found, which gives further confidence in the proposed parameters.

5. The linear and nonlinear models of the rig

A nonlinear structural dynamic analysis of the test rig has been carried out to validate the use of the measured data and
the suggested averaged friction input parameters in the calculations.
Modelling the experimental rig requires great care, since ideally all its dynamic features must be included with
considerable accuracy in order to focus on the nonlinear mechanics in the contact itself. A detailed linear finite element
model of the rig was therefore created with the commercial finite element code Ansys. This model consisted of three main
sub-assemblies (see Fig. 9): (i) the moving sub-assembly including the excitation mass on its leaf springs and the moving
arm with the moving specimen at its tip; (ii) the static sub-assembly comprising the static specimen and the static arm;
and (iii) the normal load sub-assembly with its arm and weight hangers. The three sub-assemblies were created with a
strong emphasis on an accurate modelling of their geometry, so that the mass and stiffness distribution of the rig could be
captured accurately. The bodies in the models were meshed with solid tetrahedral elements (solid95) with three degrees
of freedom (DOF) at each node, and the leaf springs, which provide the required flexibility to the rig, were modelled with
6DOF shell elements (shell63). Attached weight and stiffness less shell elements on the volume surfaces ensured that all
DOFs of the shells, used for the leaf springs, were connected fully to the solid structure. Experimental damping values were
extracted from linear sub-assembly measurements and added to the models to provide a realistic linear damping
behaviour. The contact interface of the specimen was very finely meshed to allow a good resolution for the modelling of
the friction contact interaction in the nonlinear analysis. The force transducers between the static arm and the reaction
block were included as two springs with a stiffness value taken from the transducer specification sheets. The transmitted

Table 2
Comparison of average friction input parameters to some values from the literature.

Reference Mean value [15] [29] [26] [27] [14] [28]

m 0.66–0.2 0.8–0.25 0.75–0.6 0.3–0.65 0.3–0.54

kt (N/m) 3e7 3.16e7 1.5–3.5e7

Fig. 9. The linear finite element model of the rig.

 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584 581

force at each transducer was calculated from the resulting displacement using the transducer stiffness. Each sub-assembly
model was updated and refined several times during validation against measured data, as the initial agreement did not
meet the demanding criteria set for this study. In particular, the modelling of the specimen arms proved to be challenging,
since four long threaded rods that hold the arms together were identified during the model updating as a main source of
flexibility. The final models incorporated these rods as spring elements, which led to an improved agreement with the
measured data. A modal analysis of the linear model with no contact between the moving and the static side was carried
out initially to provide the input model for the nonlinear analysis (see Fig. 1(a)).
The three-dimensional nonlinear model was based on the measured friction input parameters from the test rig and the
calculated modal response of the finite element models. Initially, only one nonlinear element was applied to the contact
interface connecting the central node of the linear FE mesh of the moving arm to its matching node on the linear static arm
FE model with a single nonlinear link (see Fig. 1(b)). In order to understand the influence of the normal load distribution,
N0, on the results, a finer contact mesh with 25 and 100 evenly spaced contact elements was also investigated. In the latter
case significantly more FE nodes in the contact plane between the moving and the static arm were connected by the
nonlinear elements allowing some areas of the contact to move while others could still be stuck. The mesh refinement was
thereby based on previous experience to achieve a good balance between mesh accuracy and computational cost, and the
need to capture the different load distributions over the contact area correctly. A finer nonlinear mesh allowed the
recreation of the non uniform loads in the contact, since each element could be assigned its own static load. An attempt
was made to directly measure the load distribution using a pressure sensitive paper (Fuji Prescale Film), but the results of
the small contact area of 1 mm2 were not conclusive. Accordingly, several different theoretical distributions, each with a
constant overall transmitted normal load of 44.5 N, were generated (see Fig. 10). These include the following: (i) a uniform
distribution with the same pressure in all elements, (ii) a random distribution, (iii) a pressure distribution from an FE
analysis, (iv) a ’radial distribution’ with increasing pressure zones towards the corners, and (v) an analytic distribution
from Ref. [25] based on a slightly modified equation to allow for two dimensions:
1
pðx,yÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1Þ
p2 ða2 x2 Þ ða2 y2 Þ
Pressure values for each of these distributions are shown in Fig. 10. Originally, the analytical and finite element results
lead to very sharp peak loads in the corners of the contact, which are significantly reduced by the nonlinear meshing
procedure, since the localised high corner pressures are smeared over the nonlinear element area. This produces much
smaller gradients in the corners, making the effect of these distributions similar to the uniform distribution of 44.5 MPa.
Only the radial distribution in Fig. 10 allows a direct influence of the load in each element and this, therefore, achieves
corner pressure values, which are closer to the analytical and finite element peak values.
Accurately measured and averaged friction parameter values were used in the nonlinear analysis as friction input
parameters. The selection of additional input parameters, mainly governing the accuracy and convergence of the nonlinear
multiharmonic analysis, was based on a previous investigation, which focused on the sensitivity of these parameters on
the nonlinear response [21]. Due to the unidirectional motion in the contact, only the first three uneven harmonics were
required to allow the multiharmonic balancing solver to calculate the results. Up to 500 modes were included in the
nonlinear analysis to capture the in and out of plane motion of the rig arms and a possible local deformation at its contact
surface, and reproduce the contact conditions accurately. With these input parameters the nonlinear frequency response
function (FRF) was computed and compared to the measured response curve. The FRF analysis considered a frequency
range from 50 to 300 Hz, including the standard 100 Hz test frequency. The models were excited with the measured
excitation force, Fexi, and the resulting displacement was extracted for a point close to the laser measurement location on
the moving arm.

Random Analytic Finite element Radial MPa

120
25 elements

100

80

60
100 elements

40

20

Fig. 10. Applied load distribution for 25 and 100 elements.

582 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584

Frequency response functions of each sub-assembly were measured and simulated, to provide an overall understanding
of the motion in the rig and to provide the required information to validate the FE models. Special attention was thereby
given to the moving sub-assembly, since it represented the major dynamic feature of the rig due to resonance of the
moving block on the leaf springs. Fig. 11 shows one measured and simulated FRF for the moving sub-assembly. It can be
seen that the simulation and the measured results agree well, and that the linear model can be considered accurate enough
for its use in a nonlinear analysis.
An accurate friction coefficient of m ¼0.74 and a tangential contact stiffness of kt ¼3.6e7 N/m were extracted from the
measured loops for a normal load of N0 ¼44.5 N. These parameters, together with the measured excitation force, Fexi, were
then supplied as input data to the numerical analysis to calculate the forced response of the test rig. The nonlinear
frequency responses of the model with 1, 25, and 100 nonlinear elements were calculated for the different normal load
distributions given in Fig. 10. The model parameters were updated several times by comparing the predicted response to
the measured data until reasonable agreement was achieved. The measured data shown in Fig. 12 shows a heavily damped
and slightly nonlinear behaviour at the resonance, caused by the friction interface of the rig. The nonlinear analysis was
able to reproduce the dynamic behaviour of the test rig well, leading to a good prediction of the resonance frequency,
and an acceptable predication of the response amplitude. The radial normal load distribution from Fig. 10 predicted the
resonance response frequency most accurately, whereas all the other pressure distributions, including the uniform one,
led to a slightly higher resonance response frequency. As previously mentioned, only the radial distribution had any
significant concentrated load peaks at the contact corners, and relatively low pressures in the centre, which caused a
reduction in the tangential contact stiffness, kt, since fewer elements were stuck during resonance, and resulted in a lower
resonance response frequency. The similar results for all the other distributions were attributed to the occurring averaging
during mesh generation, which led to a more or less uniform load distribution over the contact area. All models led to a
similar amplitudes response in Fig. 12. The remaining differences between analysis and measured values were mainly
attributed to the shortcomings of the linear rig model, since regardless of the large effort to include all the features of the
rig, the model remained a simplification of the real test setup.
No significant differences in the frequency response were observed for the 25 and 100 element models, since the nearly
unidirectional excitation of the test rig introduced a very similar amount of damping in all cases.

FRF validation of moving subassembly

1.E-02
Simulation
1.E-03 Measurement
Receptance [m/N]

1.E-04

1.E-05

1.E-06

1.E-07
0 50 100 150 200
Frequency [Hz]

Fig. 11. Measured and simulated Frequency response function of sub-assembly.

1D rig - FRF - Low Fexi - 25 elements 1D rig - FRF - Low Fexi - 100 elements
Absolute displacement Absolute displacement
1.0E-05 1.0E-05
Displ acement [m]

acement [m]

1.0E-06 1.0E-06
Displa

Measurement 01ele_uni Measurement 1ele_uni

25ele_uni 25ele_rand 100ele_uni 100ele_rand
25ele_ansys 25ele_anal 100ele_ansys 100ele_anal
25ele_rad 100ele_rad
1.0E-07 1.0E-07
150 170 190 210 230 250 150 170 190 210 230 250
Frequency [Hz] Frequency [Hz]

Fig. 12. Response of friction test rig for 25 and 100 elements, and different normal load distributions.
 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584 583

1D rig FRF -Absolute displacement -Low Fexi

Exact and average friction parameters
1.E-05

Displacement [m]
1.E-06

D Measurement
Exact friction parameters
Average friction parameters
1.E-07
150 170 190 210 230 250
Frequency [Hz]

Fig. 13. Measured and simulated frequency response function. Exact and averaged friction parameters.

The 100-element radial distribution model was recalculated with the suggested averaged friction input parameters
from the previous section (m ¼0.66, kt ¼3e7 N/m), in order to evaluate their impact on the response of the rig. The average
friction coefficient, m, was 12% smaller than the accurately measured value, and the tangential contact stiffness, kt, was
16% lower. The resulting frequency response curves for the measured data and the predictions based on the accurately
measured and averaged contact interface parameters can be seen in Fig. 13. The averaged friction input parameters reduce the
frequency accuracy slightly ( 5%), but do not affect the response amplitude significantly. The use of the suggested averaged
values in the simulation of the friction rig seems therefore an acceptable approach for the considered configuration in the
absence of measured input parameters.

6. Discussion

An investigation of the reliability of measured friction input parameters has been carried out. Test rig reassembly,
measurement repeatability, and test rig alignment studies have identified a good robustness and reliability of the
experimental friction parameters, provided that an adequate running-in time of approximately 15 min is allowed. The
investigation shows that exact and reliable values for the friction contact interfaces can be measured with the test setup.
Previous test results of the friction input parameters were collected and combined in an effort to highlight overall
trends and provide estimated input values for a material combination where no accurately measured data are available.
The collected measured friction input parameters of several different unlubricated high temperature materials showed a
strong reduction of the friction coefficient, m, from room temperatures to 200 1C and a more or less constant behaviour
thereafter. The average values, resulting from many different materials, seemed relatively independent of the contact load,
leading to the final suggestion of m20 ¼0.66 for room temperature calculations and m200 þ ¼0.25 above 200 1C (for a more
accurate distribution see Table 1). The scatter of the tangential contact stiffness, kt, was relatively large, but neither a
significant pressure nor temperature dependence could be detected on average, and an average tangential contact stiffness,
kt ¼ 3e7 N/m, is therefore suggested for all cases. It should be kept in mind that each individual material combination will
behave slightly differently, and only an actual measurement can provide accurate and exact parameters for a friction interface
analysis. If a measurement cannot be conducted, than the averaged values are suggested for a nonlinear analysis of high
temperature materials.
The overall good agreement of the measured and the simulated results of the rig simulation indicate that the chosen
approach to calculate the dynamic response of the nonlinear friction rig is valid. The two measured friction interface
parameters m and kt are sufficient to model the friction interface with a high accuracy and the analysis is able to predict the
dynamic behaviour correctly. The influence of the normal load distribution on the results was highlighted, where the most
realistic distribution with high load values in the corners and relatively low values in the centre of the contact led to the
best agreement with the measured results. No significant difference between 25 and 100 elements meshes was detected.
This was probably due to the relatively simple geometry, which minimises the importance of the number of elements, as
long as the load distribution of the contact patch can be represented accurately. Accurate friction input parameters should
be used when available, but in their absence the use of the suggested experienced based average values is an acceptable
approximation approach.

7. Conclusion

To improve the confidence in the modelling of friction contact interfaces, an investigation into the measurement of
dynamic friction contact parameters and their use in a nonlinear analysis has been carried out.
A set of accurately and averaged measured friction parameters was used in a numerical nonlinear model of the test rig
to validate the entire measurement and analysis procedure. It could be shown that the measured friction coefficient, m, and
584 C.W. Schwingshackl et al. / Mechanical Systems and Signal Processing 28 (2012) 574–584

the tangential contact stiffness, kt, can be extracted with good reliability from the test rig, and that these parameters are
sufficient and accurate enough to reproduce the dynamic behaviour of the test rig. The need of an accurate normal load
distribution over the contact is highlighted, and its influence on the response has been demonstrated. In the absence of
accurately measured friction input values the introduced average values with a temperature dependent friction coefficient, m,
and a constant tangential contact stiffness, kt, can be used for all high temperature materials. This represent an acceptable, alas
not ideal, approach to calculate the frequency response of a nonlinear system.

Acknowledgements

The authors are grateful to Rolls-Royce Plc. for providing the financial support for this project and for giving permission
to publish this work.

References

[1] C.M. Firrone, S. Zucca, M. Gola, Effect of static/dynamic coupling on the forced response of turbine bladed disks with underplatform dampers,
in: Proceedings of the ASME Turbo Expo, Orlando, Florida, USA, GT2009-59905, June 8–12, 2009.
[2] E. Cigeroglu, N. An, C.H. Menq, Wedge damper modelling and forced response prediction of frictionally constrained blades, in: Proceedings of the
ASME Turbo Expo, Montreal, Canada GT2007-27963, May 14–17, 2007.
[3] D. Charleux, C. Gibert, F. Thouverez, J. Dupeux, Numerical and experimental study of friction damping in blade attachments of rotating bladed disks,
Int. J. Rotating Mach. (2006) 1–13 Article ID 71302.
[4] E.P. Petrov, D.J. Ewins, Effects of damping and varying contact area at blade–disc joints in forced response analysis of bladed disc assemblies,
J. Turbomach.—Trans. ASME 128 (2006) 403–410.
[5] J.J. Chen, C.H. Menq, Prediction of periodic response of blades having 3D nonlinear shroud constraints, J. Eng. Gas Turbines Power—Trans. ASME 123
(2001) 901–909.
[6] E.P. Petrov, A method for use of cyclic symmetry properties in analysis of nonlinear multiharmonic vibrations of bladed discs, J. Turbomach.—Trans.
ASME 126 (2004) 175–183.
[7] O. Poudou, C. Pierre, Hybrid frequency–time domain methods for the analysis of complex structural systems with friction damping, AIAA/ASME/
ASCE/AHS/ASC Struct. Dyn. Mater. Conf. Norfolk, USA 1 (2003) 111–124.
[8] C. Siewert, L. Panning, C. Gerber, P. Masserey, Numerical and experimental damping prediction of a nonlinearly coupled low pressure steam turbine
blading, in: Proceedings of the ASME Turbo Expo 2008, Berlin, Germany, GT2008-51073, 2008.
[9] E.P. Petrov, D.J. Ewins, Generic friction models for time-domain vibration analysis of bladed disks, J. Turbomach.—Trans. ASME 126 (2004) 184–192.
[10] E.P. Petrov, D.J. Ewins, Analytical formulation of friction interface elements for analysis of nonlinear multiharmonic vibrations of bladed discs,
J. Turbomachin.—Trans. ASME 125 (2002) 364–371.
[11] E.P. Petrov, D.J. Ewins, State-of-the-art dynamic analysis for nonlinear gas turbine structures, J. Aerosp. Eng. Proc. IMechE Part G 218 (2004) 199–211.
[12] K.Y. Sanliturk, A.B. Stanbridge, D.J. Ewins, Friction dampers: measurement, modelling and application to blade vibration control, in: Proceedings of
the 1995 ASME Design Engineering Technical Conference, Boston, MA, pp. 1377–1382.
[13] A.B. Stanbridge, D.J. Ewins, K.Y. Sanliturk, J.V. Ferreira, Experimental investigation of dry friction damping and cubic stiffness non-linearity,
in: Proceedings of the 2001 ASME Design Engineering Technical Conference, Pittsburgh, PA, pp. 2141–2148.
[14] K.H. Koh, J.H. Griffin, S. Filippi, A. Akay, Characterization of turbine blade friction dampers, J. Eng. Gas Turbines Power—Trans. ASME 127 (2005)
856–862.
[15] S. Filippi, E.B. Rodrigues, M. Gola, Experimental characterization of contact hysteresis at high temperatures, in: Proceedings of the ASME Turbo Expo
2006, Barcelona, Spain, pp. 893–902.
[16] V. Lampaert, F. Al-Bender, J. Swevers, Experimental characterization of dry friction at low velocities on a developed tribometer setup for
macroscopic measurements, Tribol. Lett. 16 (1–2) (2004) 95–115.
[17] A.M. Kosunsky, K. Kim, Dissipated energy and friction coefficient evolution during fretting wear of solid lubricant coatings, Tribol. Int. 43 (2010) 861–867.
[18] L. Gaul, J. Lenz, Nonlinear dynamics of structures assembled by bolted joints, Acta Mech. 125 (1–4) (1997) 169–181.
[19] J. Szwdowicz, M. Kissel, B. Ravindra, R. Kellerer, Estimation of contact stiffness and its role in the design of a friction damper, in: Proceedings of the
Asme Turbo Expo 2001, 2001-GT-0290, New Orleans, USA, 2001.
[20] L. Gaul, R. Nitsche, Dynamics of structures with joint connections, in: D.J. Ewins, D.J. Inman (Eds.), Structural Dynamics 2000: Current Status and
Futures Directions, Research Studies Press Ltd., Baldock, Hertfortshire, England, 2001, pp. 29–48.
[21] C.W. Schwingshackl, E.P. Petrov, D.J. Ewins, Effects of contact interface parameters on vibration of turbine bladed disks with underplatform dampers,
in: Proceedings of the ASME Turbo Expo 2011: Power for Land, Sea and Air, Vancouver, Canada, GT2011-45699, June 6–10, 2011.
[22] B.W. Drinkwater, R.S. Dwyer-Joyce, P. Cawley, A study of the interaction between ultrasound and a partially contacting solid–solid interface, in:
Proceedings of the Royal Society of London, Series A (Mathematical, Physical and Engineering Sciences), 1996, vol. 452, pp. 2613–2628.
[23] S. Medina, D. Dini, A.V. Olver, D.A. Hills, Fast computation of frictional energy dissipation in rough contacts under partial slip, in: Proceedings of the
STLE/ASME International Joint Tribology Conference, Miami, USA, 2008.
[24] S. Medina, D. Dini, A.V. Olver, Adhesive contact between atomistic surfaces using a continuum analysis, in: Proceedings of the ASME/STLE
International Joint Tribology Conference, Memphis, USA, 2009.
[25] K.L. Johnson, Contact Mechanics, 1st ed., Cambridge University Press, 1985.
[26] C.H. Hager, J.H. Sanders, S. Sharma, Effect of high temperature on the characterization of fretting wear regimes at Ti6Al4V interfaces, Wear 260 (4–5)
(2006) 493–508.
[27] I. Sherrington, P. Hayhurst, Simultaneous observation of the evolution of debris density and friction coefficient in dry sliding steel contacts, Wear
249 (3–4) (2001) 182–187.
[28] G. Pintaude, D.K. Tanaka, A. Sinatora, The effects of abrasive particle size on the sliding friction coefficient of steel using a spiral pin-on-disk
apparatus, Wear 255 (1–6) (2003) 55–59.
[29] S. Filippi, A. Akay, M. Gola, Measurement of tangential contact hysteresis during microslip, J. Tribol. 126 (2004) 482–4893 126 (2004) 482–489.