Escolar Documentos
Profissional Documentos
Cultura Documentos
Gavin McGee
Research Assistant
e-mail: gavinmcgee@hotmail.com
Muhammad Haroon
Research Assistant
e-mail: mharoon@purdue.edu
Douglas E. Adams1
Assistant Professor,
e-mail: deadams@purdue.edu
Purdue University,
School of Mechanical Engineering,
Ray W. Herrick Laboratories,
140 S. Intramural Drive,
West Lafayette, IN 47907-2031
1 Introduction
The dynamic characteristics of tire-suspension systems have
many implications on the overall performance of passenger vehicles from safety to comfort. Much work has been done in the
area of vehicle system identification to support ride and handling
studies. Kim and Ro 1 demonstrated the validity of the two mass
quarter car model. Sharp and Hassan 2 used a linear quarter car
model to study passenger discomfort, suspension working space,
and tire loading characteristics for different combinations of
spring stiffness and damping levels, again showing the utility of
the quarter car model. Lin and Kortum 3 presented a timedomain direct identification method for estimating vehicle system
mass, damping, and stiffness matrices, in addition to coefficients
for nonlinear restoring forces that are linear in the parameters
Roberts et al. 4 presented a spectral identification method for
estimating the parameters associated with a nonlinear oscillator
that requires only a stochastic model of the excitation. Yi and
Hedrick 5 studied a technique for identifying nonlinear system
parameters based on a sliding observer and a least-squares
method. Nazaruddin and Yamakita 6 presented an approach
for identifying vehicle suspension models using a neurofuzzy
algorithm.
Most of this prior research recognizes that nonlinearities must
be identified and then included in subsequent analyses for the
results to be applicable for ride studies over a range of operating
conditions. Popular characterization techniques include frequency
deconvolution Siebert 7, Hilbert transforms Bracewell 8 and
Thrane 9, and wavelet transforms Chui 10. In addition, Leontaritis and Billings 11 have used correlation functions in time
to characterize nonlinear systems. Storer and Tomlinson 12 used
higher-order frequency response functions to characterize nonlinear structural dynamic systems. Cafferty et al. 13 used the re1
Author to whom correspondence should be addressed.
Contributed by the Technical Committee on Vibration and Sound for publication
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received January 16,
2003; final revision, December 23, 2003. Review conducted by D. Quinn.
2 Framework
2.1 The Model. A two degree of freedom DOF sprung/
unsprung mass quarter car model is used to introduce the technique proposed in this paper Fig. 1. This model is popular in
vehicle ride analysis because it is of low order two DOFs, fourth
M1
M2
C 1 C 2
x 1
x 2
C 2
K 1 K 2
K 2
K 2
K 2 K 3
C 2
C2
x 1
x 2
x1
x2
3 x 32 3 3 x 21 x 2 3 3 x 1 x 22 3 x 31
C 1 x b K 1 x b
.
0
3 x 31 3 3 x 1 x 22 3 3 x 21 x 2 3 x 32
(3)
Note that for this case, the nonlinearities are combinations of the
wheel unsprung and body sprung responses, x 1 and x 2 , and
not x b . To reiterate, this case occurs when the nonlinearity in the
quarter car model is in the suspension and not the tire.
By applying the Laplace transform, Eq. 3 may be subsequently manipulated more easily in the complex frequency
domain:
M 1 s 2 C 1 C 2 s K 1 K 2
C 2 sK 2
C 2 sK 2
M 2 s 2 C 2 sK 2 K 3
(1)
M 2 x 2 K 2 x 1 x 2 K 3 x 2 C 2 x 1 x 2 3 x 1 x 2 3
4 sgn x 1 x 2
(2)
To simplify the discussion here, assume that only the cubic stiffness nonlinearity in the suspension ( 2 3 4 0, 3 0) is
present. In this special case, Eqs. 1 and 2 in matrix-vector form
reduce to:
Quadratic stiffness
Cubic stiffness
Coulomb friction
Tire
Suspension
F 2 (x b x 1 ) 2
F 3 (x b x 1 ) 3
N/A
N/A
F 3 (x 2 x 1 ) 3
F 4 f susp1 sgn(x2x1)
X 1 s
C 1 sK 1
X b s Fnl s
X 2 s
0
(4)
B s
M 1 s 2 C 1 C 2 s K 1 K 2
C 2 sK 2
C 2 sK 2
M 2 s C 2 sK 2 K 3
2
(5)
D s
C 1 sK 1
0
(6)
M 2 s 2 C 2 sK 2 K 3
C 2 sK 2
C 2 sK 2
M 1 s 2 C 1 C 2 s K 1 K 2
det B
X 1 s
X 2 s
M 2 s 2 C 2 sK 2 K 3
C 2 sK 2
C 1 sK 1
X b s
0
M 1 s 2 C 1 C 2 s K 1 K 2
Fnl s ,
det B
C 2 sK 2
(7)
where Fnl(s) F nl1 (s) F nl2 (s) T . Equation 7 shows that the nonlinearities act as additional internal inputs to the two DOF system.
If the input is known or can be measured, then the transfer functions may be calculated from above as follows:
M 2 s 2 C 2 sK 2 K 3
C 2 sK 2
F nl2 s
C 1 sK 1 X b s F nl1 s
X 1 s
det B
det B
X b s
X b s
X b s
F nl1 s
M 2 s 2 C 2 sK 2 K 3
C 2 sK 2 F nl2 s
C 1 sK 1
det B
X b s
det B
X b s
F nl1 s
X 2 s C 2 sK 2
M 1 s 2 C 1 C 2 s K 1 K 2 F nl2 s
C 1 sK 1
X b s
det B
X b s
det B
X b s
(8)
(9)
However, if the excitation is unknown, as is often the case in vehicle studies involving road test data, then only the transmissibility
between X 2 (s) and X 1 (s) may be determined:
2
X 2 s C 2 sK 2 C 1 sK 1 X b s F nl 1 s M 1 s C 1 C 2 s K 1 K 2 F nl2 s
X 1 s
M 2 s 2 C 2 sK 2 K 3 C 1 sK 1 X b s F nl 1 s C 2 sK 2 F nl2 s
(10)
For Fnl(s) 0 0 T , the system is fully linear and the transmissibility function reduces to:
X 2 s H 21 s
C 2 sK 2
X 1 s H 11 s M 2 s 2 C 2 sK 2 K 3
(11)
The result in Eq. 10 changes significantly when the only nonlinearity is in the tire. Consider the equations of motion for the quarter
car model when the tire contains a cubic stiffness ( 3 0) as well as a linear stiffness ( 3 0, 4 0, 2 0):
M1
M2
C 1 C 2
x 1
x 2 C 2
C 2
C2
K 1 K 2
x 1
x 2 K 2
K 2
K 2 K 3
C 1 x b K 1 x b
3 x 3b 3 3 x b x 21 3 3 x 2b x 1 3 x 31
0
0
x1
x2
(12)
When the Laplace transform of this nonlinear forcing vector is taken, F nl2 (s) is equal to zero. Cancellation in Eq. 10 occurs in this
case and the transmissibility function is identical to that for the fully linear case, Eq. 11; consequently, if the nonlinearity is in the tire,
it will not be detectable using only the transmissibility computation between the sprung and unsprung mass. This result is intuitively
obvious and will be demonstrated later using nonlinear simulations. The implication of this result is that, in order to characterize
nonlinearities in tire, a test must be conducted where the input is known or can be measured.
The frequency response function FRF formulation for the transfer function relationship presented in Eq. 10 is given below by
substituting s j , where is the Fourier frequency, into the complex frequency relationships:
X2 j H 21 j b j C 1 K 1 Xb j F nl1 j c H 22F nl2 j
j C 2 K 2 j C 1 K 1 Xb j F nl1 j M 1 2 j C 1 C 2 K 1 K 2 F nl2 j
M 2 2 j C 2 K 2 K 3 j C 1 K 1 Xb j F nl1 j j C 2 K 2 F nl2 j
(13)
Unsprung mass
Sprung mass
Mass
Damping
Stiffness
M 1 40 kg
M 2 240 kg
C 1 49 kN s/m
C 2 7 kN s/m
K 1 140 kN/m
K 2 20 kN/m
(14)
cos 2 1 t cos 2 2 t
cos 1 2 t
2
2
cos 1 2 t
(15)
Due to the quadratic nonlinearity, additional response components will be generated at 0 rad/s, 2 1 , 2 2 , 1 2 , and
1 2 when the x i2 nonlinear input term is filtered by the nominally linear FRF characteristics in Eq. 10. If the frequency response components at 1 and 2 were assumed to be the responses at the modes of the sprung and unsprung mass,
Fig. 2 Frequency response and transmissibility functions for linear quarter car model. Hx2xb ,
, H x 1 xb , - - -, T x 2 x 1 , , Peaks occur in the frequency response functions at 1.5 and 10 Hz.
Table 3 Double frequency permutations for polynomial nonlinearities when x t cos 1 t cos 2 t Adams and Allemang 19
Polynomial
x
x2
x3
x4
Frequency permutations
1 , 2
0, 2 1 , 2 2 , 1 2 , 1 2
1 , 2 , 3 1 , 3 2 , 1 2 2 , 2 2 1 , 1 2 2 , 2
2 1
0, 2 1 , 2 2 , 4 1 , 4 2 , 1 2 , 2 1 , 2 1 2 2 ,
2 1 2 2 ,
3 1 2 , 3 2 1 , 3 1 2 , 3 2 1
Table 4 Nonlinear model and signal processing parameters in the Simulink quarter car
model simulations
Nonlinearity
in simulation
Quadratic
tire stiffness
Cubic
suspension
stiffness
Coulomb
friction in
suspension
Input m
x b 0.06* randn(Nt,1)
x b 0.06* randn(Nt,1)
x b * randn(Nt,1)
25%: 0.6
50%: 0.2
75%: 0.07
100%: 0.01
Nonlinear
parameters
Fs
Hz
N avg
BS
Nt
2 5.5104
N/m2
3 1.05104
N/m3
128
64
2048
2 16
128
64
2048
2 16
128
64
2048
2 16
4 1
f susp0.1M 2 g
N
Fig. 3 Frequency response function and coherence function between sprung mass response, x 2 t , and base input, x b t . Quarter car model with quadratic tire stiffness, input
known.
Fig. 4 Transmissibility function and coherence function between sprung mass response,
x 2 t , and unsprung mass response, x 1 t . Quarter car model with quadratic tire stiffness,
input unknown.
Fig. 5 Frequency response function and coherence function between sprung mass response, x 2 t , and base input, x b t . Quarter car model with cubic suspension stiffness, input
known.
Fig. 6 Transmissibility function and coherence function between sprung mass response,
x 2 t , and unsprung mass response, x 1 t . Quarter car model with cubic suspension stiffness, input unknown.
Fig. 7 Transmissibility function and coherence function between sprung mass response,
x 2 t , and unsprung mass response, x 1 t . Quarter car model with Coulomb friction in suspension, input unknown, 100% nonlinearity.
nonlinearity on the response are much more pronounced. The coherence function also exhibits signs of nonlinearity. A broad dip in
the coherence function can be seen near 30 Hz. Because the restoring force curve for the frictional force can be approximated
with an odd function i.e., x 3 , x 5 , etc., there is an additional
response component at 3 f 2 . The drop in the coherence function at
30 Hz is the result of this kind of permutation. The coherence
functions have degraded downward on both sides of the unsprung
mass resonance at 10 Hz. The trend downward from a coherence
of one to near zero in the region of 30 Hz appears to be almost
linear. This behavior of the coherence function can be used to
determine the unsprung mass resonant frequency when the input
is unknown. The coherence reaches one near the unsprung mass
resonance because, at that frequency, the unsprung mass is moving with relatively large amplitude, which results in a relatively
large velocity thereby making the Coulomb damping force less
noticeable in relation to the linear damping force. The rise in
coherence occurs at slightly less than 10 Hz because the large
frictional force causes the damped natural frequency ( 1 2 n )
to decrease, where n is the undamped natural frequency of the
system. The large frictional force causes an increase in the damping ratio thus causing a decrease in the undamped frequency.
M T P s 2 C 1 sK 1
C 1 sK 1
C 1 sK 1
M 1 s C 1 C 2 sK 1 K 2
C 2 sK 2
C 2 sK 2
M 2 s 2 C 2 sK 2 K 3
X b s
F b s
X 1 s
0
Fnl s
X 2 s
0
(18)
X b s
F b s
0
Fnl s
B s X 1 s
0
X 2 s
Fig. 8 Analytically generated frequency response functions FRFs and transmissibility function for three degree of freedom quarter car model with input force excitation, f b t , for M TP
0.05 M 1 " H x 1 xb ,
, H x 2 xb , - - -, T x 2 x 1 , , H xbxb , " - " -. Peaks occur in the FRFs at 0.66,
3.78, and 43.1 Hz.
Note that no damping or stiffness has been included to represent the tire patch itself; C 1 and K 1 correspond to the damping
and stiffness in the tire, not the tire patch. When the vector-matrix
equation above is solved, the three DOF transfer function matrix,
H(s) , is found by inverting the system impedance matrix,
B(s) . The functions of interest in this model include the FRF
between the tire patch excitation force, f b (t), and the motions of
the unsprung and sprung masses, as well as the transmissibility
functions between the unsprung and sprung mass motions. The
expressions for the purely linear case are:
X 2 s C 1 sK 1 C 2 sK 2
F b s
s
(19)
X 1 s M 2 s 2 C 2 sK 2 K 3 C 1 sK 1
F b s
s
(20)
X 2 s
C 2 sK 2
X 1 s M 2 s 2 C 2 sK 2 K 3
(21)
X 2 s
C 1 sK 1 C 2 sK 2
X b s M 1 s 2 C 1 C 2 sK 1 K 2 M 2 s 2 C 2 sK 2 K 3 C 2 sK 2 2
(22)
mass, were used to analytically generate the FRFs and transmissibility functions for the three DOF model. The FRF magnitudes
and phases and the transmissibility function magnitude generated
by inverting the impedance matrix in Eq. 18 are plotted in Fig.
8.
The three peaks in the FRFs occur at 0.66, 3.78, and 43.1 Hz. It
is clear from the phase plots that at the first mode 0.66 Hz, all
three masses are moving in phase. At the second mode 3.78 Hz,
the tire-patch mass and sprung mass are moving in phase and the
unsprung mass is out of phase with them. At the third mode 43.1
Hz, the tire-patch and unsprung mass are moving in phase and
FEBRUARY 2005, Vol. 127 69
Fig. 9 Experimental setup for electrohydraulic shaker test with acceleration measurements at the tire patch, spindle, and the top of the strut at the body
the sprung mass is out of phase with them. The forced response of
the three DOF model indicates that the sprung and unsprung
masses are oscillating out of phase at the lowest-frequency synchronous mode at 3.78 Hz. In the two DOF model, the two masses
oscillate in phase at the lowest-frequency mode at 1.36 Hz. Despite the differences between the responses of the 2 DOF and 3
DOF models, the peak at 1.45 Hz in the magnitude curve of the
transmissibility between the sprung and unsprung masses,
X 2 ( j )/X 1 ( j ), is identical for both models.
5 Experimental Verification
The characterization procedure presented in the previous section is applied to laboratory shaker and vehicle road data in this
section.
5.1 Laboratory Shaker Data. The laboratory data was taken
on the left corner of a midsize vehicle using a 1000 lbf. electrohydraulic shaker apparatus. An illustration of the experimental
setup is given in Fig. 9. Normally distributed broad band forces at
three different levels 5.5, 11.7, and 20.9 lbf root-mean square
were used and the corresponding excitation time history, f b (t), in
the positive vertical direction at the patch was measured using a
load cell with a 1000 lbf/mV sensitivity. The spindle acceleration,
x 1 (t), and the upper strut body connection point acceleration,
x 2 (t), in the positive vertical direction were also measured using
accelerometers with sensitivities of 102.5 and 97.3 mV/g, respectively. The other three tires of the vehicle were supported on uncoupled rigid platforms and it is assumed in this work that the
motions of the tires and suspension elements at these other corners
are negligible.
5.1.2 Frequency Response of Vehicle Corner. The measured
time histories from the test were filtered and then sampled at 5000
Hz to produce a total of 410,000 time points. The FRFs and transmissibility functions between f b (t), x 1 (t), and x 2 (t) were calculated using the signal processing parameters in Table 5.
Table 5 Signal processing parameters used in frequency response function FRF estimation of vehicle corner
t
s
f
Hz
BS
Navg
Overlap
Window
type
FRF estimator
1/5000
0.1356
36,864
20
47%
Hanning
H1
Fig. 10 Experimentally obtained frequency response function FRF data for input force,
f b t , at the tire patch and output motion, x 2 t , at the body connection point. Three different
FRFs are shown for three excitation levels.
Fig. 11 Experimentally obtained transmissibility function data between response acceleration, x 2 t , of the body and acceleration, x 1 t , at spindle in the vertical direction. Three
different functions are shown for three excitation levels.
Table 6 Signal processing parameters used in frequency response function FRF estimation of vehicle corner
t
s
f
Hz
BS
Navg
Overlap
Window
type
FRF estimator
1/1024
0.1111
9216
40
0%
Hanning
H1
evident in Fig. 11, i.e., the suspension nonlinearity is being excited at this frequency whereas tire nonlinearities are not.
Next, consider the drop in coherence near 5.4 Hz. This drop can
be attributed to a cubic stiffness nonlinearity in the suspension
system that produces a mapping from 1.8 Hz to 1.8 and 5.4 Hz
cubic maps into and 3. Note that the coherence in both
plots is low at 1.8 and 5.4 Hz. Thus, it can be concluded that the
suspension has both a quadratic nonlinearity and a softening cubic
stiffness nonlinearity, softening because the lowest peak in Fig. 11
decreases in frequency as the amplitude of the excitation level
increases.
The consistent drops in coherence in the neighborhood of 10
and 30 Hz, and the drop near 5 Hz for that matter, can also be
attributed to nonlinear cubic stiffness in the suspension by considering the relative motion across the strut. In general, whenever the
relative motion between the body and the spindle is large, the
suspension nonlinearities will be excited and the coherence will
tend to drop at those frequencies in addition to three times those
frequencies. More specifically, Fig. 11 shows that the spindle is
responding with much larger relative deflections than the body at
5, 10, 20, and 38 Hz, so it is expected that the cubic nonlinearity
within the suspension will create drops in coherence near 5, 10,
and 20 Hz and 15, 30, and 60 Hz. All of these drops in coherence
can be easily seen in Fig. 10 except for the 20 Hz case, which is
subtle but present nonetheless, and the 60 Hz case, which is beyond the range of the excitation spectrum 0.5 to 45 Hz.
The presence of Coulomb friction in the suspension is confirmed by the tendencies of the coherence functions in Figs. 10
and 11. Note the coherence function is worst for the lowest level
input but improves with increasing excitation level. Also, the coherence function dips slightly before 30 Hz. In the simulation
portion of the research, the friction characteristic first appeared in
Fig. 12 Transmissibility and coherence function between sprung mass response, x 2 t , and
unsprung mass response, x 1 t for a vehicle speed of 40 mph on rough road
Fig. 13 Coherence function between sprung mass response, x 2 t , unsprung mass response, x 1 t ,
, and normalized power spectrum of relative motion between the unsprung
mass and sprung mass, , for a vehicle speed of 40 mph on rough road. Two frequency
ranges; i 015 Hz and ii 1530 Hz.
Description
Nonlinearity
14.4
Frequency permutation,
f 1 f 2
Frequency permutation,
f 2 2 f 1
Frequency permutation,
2f2
Frequency permutation,
2 f 2 f 1
Worsening trend in
coherence
Quadratic stiffness in
suspension
Cubic stiffness in
suspension
Quadratic stiffness in
suspension
Cubic stiffness in
suspension
Coulomb friction in
suspension
16.5
25.5
27.5
24 50
Description
Nonlinearity
14.2
Frequency permutation,
f 1 f 2
Frequency permutation,
f 2 2 f 1
Frequency permutation,
2 f 2 f 1
Frequency permutation,
2 f 2 f 1
Consistently worsening
trend in coherence
Quadratic stiffness in
suspension
Cubic stiffness in
suspension
Cubic stiffness in
suspension
Cubic stiffness in
suspension
Coulomb friction in
suspension
Description
Nonlinearity
16.5
Quadratic stiffness in
suspension
Cubic stiffness in
suspension
Coulomb friction in
suspension
22.5
Frequency permutation,
f 2 2 f 1
Worsening trend in
coherence
27.5
2250
Fig. 14 Transmissibility and coherence function between sprung mass response, x 2 t , and
unsprung mass response, x 1 t for a vehicle speed of 60 mph on an urban highway
Fig. 15 Coherence function between sprung mass response, x 2 t , unsprung mass response, x 1 t ,
, and normalized power spectrum of relative motion between the unsprung
mass and sprung mass, , for a vehicle speed of 60 mph on an urban highway. Two frequency
ranges; i 015 Hz and ii 1530 Hz.
Fig. 16 Transmissibility and coherence function between sprung mass response, x 2 t , and
unsprung mass response, x 1 t for a vehicle speed of 35 mph on an urban highway
Fig. 17 Coherence function between sprung mass response, x 2 t , unsprung mass response, x 1 t ,
, and normalized power spectrum of relative motion between the unsprung
mass and sprung mass, , for a vehicle speed of 35 mph on an urban highway. Two frequency
ranges; i 015 Hz and ii 1530 Hz.
tude is lower as can be seen from the power spectra in Figs. 13,
15, and 17. It is apparent that for low speeds there is more energy
at low frequencies, and for high speeds the energy is distributed
over a larger-frequency range. Because Coulomb friction is more
significant at lower amplitudes, it is more apparent in this case.
It can be concluded from the above discussion that the suspension has both quadratic and cubic stiffness nonlinearities and Coulomb friction. As mentioned earlier, complex nonlinearities can be
approximated by polynomials. The types of complex nonlinearities generally observed in vehicle suspension systems are hysteretic stiffness with backlash dominant in low-frequency range,
Coulomb friction damping, and bilinear damping dominant in
high-frequency range. Hence, the drops in coherence attributed to
polynomial nonlinearities are possibly also due to the presence of
such complex nonlinearities. An important thing to note is that
information about the nonlinear system helps to identify the underlying nominally linear system. Good coherence around the unsprung mass resonance helps to identify it. Subsequently, drops in
coherence around this frequency help to identify the sprung mass
resonance. For example, a drop around 14 Hz can be attributed to
a frequency permutation f 1 f 2 . Knowing the approximate unsprung resonance, the sprung resonance can be identified.
Thus, measured transmissibility functions, which do not require
full knowledge or measurement of the force or motion excitation
at the tire patch, can be used to characterize nonlinearity in spite
of not knowing the exact underlying linear nature of the tirevehicle suspension system. This ability to characterize nonlinearity without all of the input-output information is a result of the
fact that the nonlinearity itself acts as an internal force that is
incoherent with the linear external applied excitation, thereby
causing drops in coherence.
sue that needs to be investigated is the coupling between the vertical, lateral, and longitudinal degrees of freedom in the suspension. The need for these measurements may require more instrumentation on the vehicle during road tests as well as other types of
testing.
Nomenclature
B(s)
Fnl(S)
f b (t)
H a,b ( j )
T a,b ( j )
X( j )
X(s)
x(t)
(s)
,
Impedance matrix
Laplace transform of vector of nonlinear terms
Excitation time history
Frequency response function between input b and
output a
Transmissibility between a and b
Fourier spectrum of the output time history
Laplace transform of the output time history
Measured output time history
Characteristic polynomial
Scalar nonlinear parameters for nonlinear elements
References
1 Kim, C., and Ro, P. I., 2000, Reduced-Order Modeling and Parameter Estimation for a Quarter-Car Suspension System, Proc. Inst. Mech. Eng., 214D,
pp. 851 864.
2 Sharp, R. S., and Hassan, S. A., 1986, An Evaluation of Passive Automotive
Suspension Systems With Variable Stiffness and Damping Parameters, Veh.
Syst. Dyn., 15, pp. 335350.
3 Lin, Y., and Kortum, W., 1991, Identification of System Physical Parameters
for Vehicle Systems With Nonlinear Components, Veh. Syst. Dyn., 20, pp.
354 365.
4 Roberts, J. B., Dunne, J. F., and Debunos, A., 1995, A Spectral Method for
Estimation of Nonlinear System Parameters From Measured Response,
Probab. Eng. Mech., 10, pp. 199207.
5 Yi, K., and Hedrick, K., 1995, Observer-Based Identification of Nonlinear
System Parameters, J. Dyn. Syst., Meas., Control, 117, pp. 175182.
6 Nazaruddin, Y. Y., and Yamakita, M., 1999, Neuro-Fuzzy Based Modeling of
Vehicle Suspension System, Proc. of the 1999 IEEE International Conference
on Control Applications, New York, pp. 14901495.
7 Siebert, W. M., 1986, Signals and Systems, McGrawHill, New York.
8 Bracewell, R. N., 1986, The Fourier Transform and Its Applications, 2nd Ed.,
WCB/McGrawHill, Boston, Mass.
9 Thrane, N., 1984, The Hilbert Transform, Hewlett Packard Application Notes.
10 Chui, C. K., 1992, An Introduction to Wavelets, Academic, San Diego, CA.
11 Leontaritis, I., and Billings, S., 1985, Input-Output Parametric Models for
Nonlinear Systems. Part I: Deterministic Nonlinear Systems, Int. J. Control,
412, pp. 303328.
12 Storer, D. M., and Tomlinson, G. R., 1993, Recent Developments in the
Measurement and Interpretation of Higher-Order Transfer Functions From
Nonlinear Structures, Mech. Syst. Signal Process., 72, pp. 173189.
13 Cafferty, S., Worden, K., and Tomlinson, G., 1995, Characterization of Automotive Shock Absorbers Using Random Excitation, J. Autom. Eng. Part D,
209, pp. 239248.
14 Audenino, A. L., and Belingardi, G., 1995, Modeling the Dynamic Behavior
of a Motorcycle Damper, J. Automobile Eng.: Part D, 209, pp. 249262.
15 Ahmed, A. K. W., Rakheja, S., and Richard, M. J., 1993, Frequency Response Analysis of Symmetric and Asymmetric Nonlinear Vehicle Suspension, Trans. Can. Soc. Mech. Eng., 174B, pp. 837 850.
16 Adams, D. E., 2000, A Spatial Approach to Nonlinear Vibration Analysis,
Doctoral Dissertation, University of Cincinnati, OH.
17 Nayfeh, A. H., and Balachandran, B., 1989, Modal Interactions in Dynamical
and Structural Systems, Appl. Mech. Rev., 4211, Part 2, pp. 175201.
18 Nayfeh, A. H., and Mook, D. T., 1979, Nonlinear Oscillations, Wiley, New
York.
19 Adams, D. E., and Allemang, R. J., 2001, Discrete Frequency Models: A
New Approach to Temporal Analysis, ASME J. Vibr. Acoust., 123, pp. 98
103.