Beruflich Dokumente
Kultur Dokumente
Ph.D.
Principal Engineer
Eaton Corp Innovation Center,
26201 Northwestern Highway,
Southfield, MI 48076
e-mail: xubinsong@eaton.com
Mehdi Ahmadian
Ph.D.
Professor and Director
Advanced Vehicle Dynamics Laboratory,
Department of Mechanical Engineering,
Virginia Polytechnic Institute and State
University,
Blacksburg, VA 24060-0238
Steve Southward
Ph.D.
Lane R. Miller
Ph.D.
Director
Introduction
hop mode control. Both Delphi and Lord Corp Cary, NC work
together and have commercialized magnetorheological MR
damper based MagneRide suspension.
Adaptive control systems refer to the class of systems that are
able to adapt to the changes that may occur in the system. Adaptive control techniques have been successfully applied to a wide
variety of engineering problems, ranging from motion control,
communication problems, noise cancellation systems, speech
analysis and synthesis, and many other signal processing problems. The primary impetus for using adaptive control techniques
is that often the systems to be controlled are noisy i.e., are subjected to disturbances, or have unknown or unmodeled characteristics. Adaptive control methods for such systems are advantageous over fixed systems, because the controller parameters can
be adjusted or tailored to the unknown and/or varying characteristics of the system. However, due to the nonlinearity of semiactive devices, existing adaptive control for suspension control
usually requires a kind of linearization as will be shown in the
following review.
In order to take advantage of the features of semiactive and
adaptive systems, several studies in the past have discussed semiactive adaptive systems. The main effort of most of these studies
has been on linear or can-be-linearized systems that use semiactive devices and/or a skyhook control policy or its variations,
Shoureshi 7, Henry et al. 8, Venhovens 9, Boyle et al. 10,
and Bubhardt and Isermann 11, to name a few. Shoureshi proposed using both the self-tuning regulator and fuzzy logic control
to design an intelligent vibration control system 7. Some required preconditions, however, can make implementation difficult,
such as linearizing a dynamic system and requiring a reference
signal. In the Henry et al. patent, they try to use more available
measured signals to tune the damping for a full car suspension. It
is a derivative of the skyhook control law developed in 1. Different signals reflect different driving situations, thus if the damp-
ers are tuned properly, the design objective can be achieved. In the
Venhovens and Boyle et al. adaptive semiactive algorithms, some
parameters of the suspension systems such as mass are required to
be known. Actually mass is always varying in such as passenger
vehicles. Thus, it is not practical to design an adaptive suspension
in this way regardless of other issues. Bubhardt and Isermann 11
and Blankenship and Polyakov 12 apply model reference adaptive control MRAC to vehicle suspension systems using nonlinear shock dampers. In Bubhardt and Isermann study, the shock
damper is linearized with the MRAC approach and the suspension
system is designed based on linear system theory. Blankenship
and Polyakov assume that the road roughness can be measured as
a reference signal. Actually this measurement is not only difficult
but also expensive with the current sensing technology. Finally,
wed like to mention the work that has been done by Rubenstein
and Allie 13. They modify the configuration of least-mean
square LMS adaptive filters to be applicable for semiactive systems. The modified algorithm is a feedforward system, which still
requires a reference signal. Further, the plant is modeled using
Finite Impulse Response FIR filters.
The vast majority of semiactive devices, however, are nonlinear
as mentioned above. For instance, hydraulic shock absorbers,
electrorheological ER dampers, and magnetorheological MR
dampers exhibit nonlinearities of such as bilinear characteristics
and hysteresis. As mentioned earlier, although previous studies
have addressed semiactive adaptive systems, none has addressed
how to design semiactive adaptive control directly to deal with the
nonlinearities of semiactive devices as subjected to immeasurable
nonstationary vibration sources. The other objective is to try to
preclude some dynamics issues inherent with skyhook control as
mentioned before 1417. This study intends to address these
issues with the application of MR dampers, which have been attracting academic research and industrial applications.
where K is the spring constant and y 1 y 2 is the relative displacement across the spring. It is worth noting, however, that the adaptive control formulation that will be described next is not necessarily confined to linear systems or isolation systems with linear
springs.
2.2 Nonparametric MR Damper Model. There are a number of modeling approaches to build MR damper models 1721.
In this paper, a nonparametric model from 17,18 will be used to
develop the proposed adaptive algorithm. In 19,20, a parametric
model has been developed based on the Bouc-Wen hysteresis
model. A FIR model is suggested for modeling hysteresis in 22
for formulating a nonlinear adaptive algorithm. However, the
Bouc-Wen model based MR damper model and the FIR-type hysteresis model have big disadvantages for real-time control system
implementation 17,22. Some other approaches by applying such
as the Preisach hysteresis model 2325 may be applicable to this
case too. But in this paper, a nonparametric model that can convert the hysteresis into a memoryless nonlinearity is highlighted
for this application, because this conversion can be used to prove
the adaptive system stability.
The nonparametric MR damper model is briefly described in
the following four equations that will be used to formulate the
semiactive adaptive control algorithm, but readers can refer to
17,18 for more details about this nonparametric model, and alternative formulations.
FMR = Fh + Fbias
Background
AMRI =
aI
i
i=0
vides the non-parametric model for MR dampers. Here for completeness and brevity, this non-parametric MR model is compared
with a parametric model 19,20 that is based on the Bouc-Wen
types of hysteresis model 21. Both are compared with experimental data as shown in Figs. 2 and 3. Table 1 presents one set of
parameter values for an optimized nonparametric MR damper
model. Please note that b2 is a function of the current I to the MR
damper in Table 1.
Figures 2 and 3 shows that the parametric model overestimates
the damping force when the applied damper current goes to 1.5
and 2 A. Furthermore, from our simulation experience, the parametric model needs an integration step size as small as 105 s
while the nonparametric model can use a much larger step size of
0.01 s. Since the step size determines the model simulation speed,
obviously such a parametric model cannot be included in the realtime control algorithm. Thus for the adaptive algorithm proposed
in this paper, the nonparametric model will be applied. In Fig. 4,
the hysteresis of the optimized MR damper can be further characterized as Fs, which is represented by MR=Fs and the delay
function Gds of Eq. 5. MR is a memoryless nonlinearity
that satisfies the sector condition , as follows:
MR y 1 y 2MR y 1 y 2 0
0 and 0 represent the minimum and maximum MR damping coefficient of the shape function Fs with respect to relative
velocity when the current belongs to 0 to 2 A. Actually the maximum current for this study is 1 A as shown later. See Fig. 5. For
this specific case, Gds is derived from Eq. 5 as
Gds =
h0
s + h0
10
Value
Parameter
Value
A0
A1
A2
A3
A4
V0
Fbias
231.0
230.4
5303.1
548.9
1603.8
0
0
b0
b1
b2
h0
h1
h2
h3
h4
2.7183
0
0.2 I + 0.1
111.1
0
0
1
0
11
= y 1ky 12kT
where y 1k is the measured acceleration and
y 12k = y 1k y 2k
represents the measured relative displacement. Further, the parameter vector
Ik + 1 = Ik +
k + 1 = k + k FMR T
12a
k = Pk + T Pk1
12b
1
Pk + 1 = Pk Pk1 + T Pk1T Pk
12c
14
= MKT
includes unknown mass M and stiffness K of the vibration isolator, respectively. The recursive least-square RLS algorithm for
the parameter estimation can be written as
13
J
Ik
J
y 1k
y 12k
= 2py 1k
+ 2qy 12K
Ik
Ik
Ik
15
16
y 1
y 12
+ FMRy 12,I + K
=0
I I
I
17a
y 1 FMR y 12
y 12
FMR
+
=
+K
I
I
y 12 I
I
17b
18
Fig. 12 Effect of system identification input filter on estimating the system mass and stiffness
e = FMR T
33
26
h01 M 22 + K + h0 M 222
h01 M 22 + K + h0 M 222
27a
27b
28
Since the above two Eqs. 27 and 28 hold, the transfer function
ZTs of Eq. 26 is strictly positive real. Thus it can be concluded
that the dynamic system with the related hysteresis as shown in
Fig. 7a is globally absolutely stable. It further means that the
adaptive system in Fig. 6 is globally absolutely stable.
4.2 MR Adaptive System Convergence. Though the adaptive control system is stable, its stability certainly does not guarantee proper convergence of the adaptive controller. Remember
that the adaptive control law is designed to minimize a quadratic
cost function of acceleration and displacement as given by Eq.
14. In order to simplify the problem, the minimization of acceleration will be discussed here and in the later simulation study.
Furthermore, since the h0 about 20 Hz is much higher than the
resonant frequency about 1.4 Hz and the MR suspension system
has only one equilibrium point of 0, 0, we can argue that the
following linearized system can be used to discuss the studied
system convergence:
My 1 + Cy 1 y 2 + Ky 1 y 2 = 0
29
where
C=
FMR
y 1 y 2
1 + 2r2
y1
=
y2
1 r22 + 2r2
30
1/2
31
where
r=
,
n
n =
K
,
M
and =
C
2M
1 + 2r2
=
1 r22 + 2r2
1/2
4r21 r22 1
1 r22 + 2r22
32
1
1,
1 r2
0 0,
=0
and lim
1
at = 0 for r 2, +
1 r2
Thus in the high frequency range, i.e., r 2, the transmissibility is limited by an upper bound limit of one even if
the damping coefficient becomes infinity very stiff
damper, when the suspension excitation frequency is beyond the critical frequency. In other words, the vibration
response cannot be amplified regardless of damping level
for high-frequency vibrations. Furthermore, for any frequency, is the minimum value if becomes the smallest.
Thus, it can be concluded that if the MR suspension system
is subjected to vibration with exciting frequencies greater
than the critical frequency, the smallest current is desired
for the minimum transmissibility.
The above analysis is consistent with the engineering design
criteria of shock absorbers for suspension systems. Next we further illustrate our specific case to demonstrate the algorithm convergence. The MR suspension system resonant frequency is designed to be 1.4 Hz as shown in the next section, and International
Organization for Standardization ISO 27,28 is selected to be
the excitations in Fig. 8 as well as 1.5 and 3.5 Hz tonal excitations. Therefore, it is clear that, compared to the 1.4 Hz resonant
frequency, ISO1 and the 1.5 Hz pure tone represent a low frequency vibration and ISO4 and the 3.5 Hz tone can be considered
high frequency vibration sources. In Fig. 9, the cost function, as
represented by RMS root mean square acceleration i.e., q = 0 is
plotted with respect to constant current to the MR damper for a
variety of different excitation spectrums.
For this study, the current is limited to a maximum of 1 A,
considering the possible saturation limit in practice. As indicated
Transactions of the ASME
in the earlier sections, the larger the current the more damping the
MR damper produces. Thus, the current is directly proportional to
the damping force 17,18. Furthermore, it can be observed that
the optimal relationship between acceleration RMS value and current in Fig. 9 matches the convergence prediction as presented
above.
Figure 9 clearly shows that, for these excitation spectrums, the
cost function has a unique global minimum corresponding to a
single optimal current command to the damper. Figure 9 also indicates that when the excitation spectrum is concentrated above
the isolation frequency, for instance the 3.5 Hz pure tone and
ISO4, the optimal current is 0 A. When the excitation spectrum is
concentrated below the isolation frequency, such as the 1.0 Hz
pure tone and ISO1, the optimal current is max full scale. From
this analysis, it can be concluded that unique global minima exist
for certain classes of excitation spectra and therefore an associated
unique optimal control will also exist. What remains is to show
that the adaptive algorithm converges to the optimal control solutions. This result will be presented in the next section through
numerical simulation studies. In addition, the dynamic accelerations obtained from the adaptive approach are compared with soft
and hard damping cases in the time domain in order to clearly
show how the adaptive algorithm works.
Simulation Results
Fig. 11
Fig. 12 Effect of system identification input filter on estimating the system mass and stiffness
e = FMR T
33
Fig. 16 Comparison of the seat acceleration induced by adaptive control and passive damping
posed adaptive control can provide significant advantages at frequency ranges that are not around the critical frequency. Then
different unknown excitations are used to test the proposed adaptive control by comparing to soft and hard damping. The simulation results show that the developed adaptive control can achieve
the optimal performance in the broadband vibration frequency.
However, it is still important to find ways to improve the accuracy of system identification, though such errors cannot destabilize the adaptive damping-tuning system. Another important aspect about model-based algorithms is always about model
uncertainty and its effect on the control system performance.
These as well as other issues, such as nonlinear model based
convergence study and expansion of the proposed adptive control
algorithm to multiple degree-of-freedom MDOF suspension systems are recommended for future research.
Acknowledgements
The authors would like to acknowledge the financial support
provided by the Lord Corporation, the National Science Foundation, and the Advanced Vehicle Dynamics Laboratory of Virginia
Tech throughout this study.
References
1 Karnopp, D. C., and Crospy, M. J., 1974 System for Controlling the Transmission of Energy between Spaced Members, U.S. Patent 3,807,678, April.
2 Ahmadian, M., Reichert, B. A., and Song X., 1997, Harmonic Analysis of
Semiactive Suspensions, Proceedings of DETC97, Sacramento, CA, September 1417, ASME International, New York.
3 Ivers, D. E., and Miller, L. R., 1989, Experimental Comparison of Passive,
Semiactive on/off, and Semiactive Continuous Suspensions, SAE, Truck and
Bus Meeting and Exposition, Charlotte, NC, November 69, Society of Automotive Engineers, Warrendale, PA.
4 Ivers, D. E., and Miller, L. R., 1991, Semiactive Suspension Technology: An
Evolutionary View, DE-VOL. 40, Advanced Automotive Technologies, ASME,
New York.
5 Margolis, D. L., 1983, A Procedure for Comparing Passive, Active, and Semiactive Approaches to Vibration Isolation, J. Franklin Inst., 3154, pp. 225
238.
6 Karnopp, D. C., 1990, Design Principles for Vibration Control Systems Using
Semiactive Dampers, J. Dyn. Syst., Meas., Control, 112, pp. 448455.
7 Shoureshi, R., 1995, Method and Apparatus for Intelligent Active and Semiactive Vibration Control, U.S. Patent No. 5,418,858, May 23.
8 Henry, R. R., Applebee, M. A., and Murty, B. V., 1995, Full Car Semiactive
Suspension Control Based on Quarter Car Control, U.S. Patent 5,475,596,
Dec. 12.
9 Venhovens, P. J. Th., 1994 The Development and Implementation of Adaptive
Semiactive Suspension Control, Veh. Syst. Dyn., 23, pp. 211235.
10 Boyle, F. P., Petek, N., and Smith, D. P., 1996, Method For Controlling
Motion Using an Adjustable Damper, U.S. Patent No. 5,582,385, December
10.
11 Bubhardt, J., and Isermann, R., 1992, Realization of Adaptive Shock Absorbers by Estimating Physical Process Coefficients of a Vehicle Suspension System, ACC/WM1, Proceedings of the American Control Conference, American
Automatic Control Council, New York.
12 Blankenship, G. L., and Polyakov, R. G., 1993, Nonlinear Adaptive Vehicle
Suspensions, Proceedings of the American Control Conference, San Francisco, CA, June, American Automatic Control Council, New York.
13 Rubenstein, S. P., and Allie, M. C., 1991, A Switching Element Adaptive
Control Algorithm For Nonlinear Systems, Symposium on Recent Development of Acoustic and Vibration Control, Virginia Tech, Blacksburg, VA.
14 Ahmadian, M., Reichert, B. A., Song, X., and Southward, S. C., 2000; No
Jerk Semi-active Suspension System, U.S. Patent 6,115,658, September 5.
15 Ahmadian, M., Reichert, B. A., and Song, X., 2001, System Nonlinearities
Induced by Skyhook Dampers, Shock Vib., 82, pp. 95104.
16 Song, X., and Ahmadian, M., 2004, Study of Semiactive Adaptive Control
Algorithms with Magneto-Rheological Seat Suspensions, SAE 2004-01-1648,
2004 SAE World Congress, Detroit, MI, March, Society of Automotive Engineers, Inc., Warrendale, PA.
17 Song, X., 1999, Design of Adaptive Vibration Control Systems with Application of Magneto-Rheological Dampers, Dissertation, Virginia Tech, December.
18 Song, X., Ahmadian, M., and Southward, S. C., 2005, Modeling MagnetoRheological Dampers with Application of Non-parametric Approach, J. Intell.
Mater. Syst. Struct., 165, pp. 421432.
19 Spencer, B. F., Dyke, S. J., Sain, M. K., and Carlson, J. D., 1996, Phenomenological Model of a Magnetorheological Damper, J. Eng. Mech., 123, pp.
230238.
20 Spencer, B. F., Dyke, S. J., Sain, M. K., and Carlson, J. D., 1996, Modeling
and Control of Magnetorheological Dampers for Seismic Response Reduction, Smart Mater. Struct., 5 pp. 565575.
21 Wen, Y. K., and Asce, M., 1976, Method for Random Vibration of Hysteretic
Systems, J. Eng. Mech. Div., 102EM2, pp. 249263.
22 Isermann, R., Lachmann, K. H., and Matko, D., 1992 Adaptive Control Systems, Prentice-Hall, Englewood Cliffs, NJ.
23 Par, T. E., and How, J. P., 1998, Robust Ability and Performance Analysis of
Systems with Hysteresis Nonlinearities, Proceedings of the American Control
Conference, Philadelphia, Pennsylvania, June, American Automatic Control
Council, New York.
24 Par, T. E., Hassibi, A., and How, J. P., 1999 Asymptotic Stability for Systems with Multiple Hysteresis Nonlinearities, Proceedings of the 1999 American Control Conference, American Automatic Control Council, New York.
25 Gobert, R. B., Morris K. A., and Wang, D. W. L., 2001, Passivity-Based
Stability and Control of Hysteresis in Smart Actuators, IEEE Trans. Control
Syst. Technol., 91, pp. 516.
26 Khalil, H., 1996, Nonlinear Systems, Prentice-Hall, Englewood Cliffs. NJ.
27 SAE Standard, 1992, Measurement of Whole Body Vibration of the Seated
Operator of Highway Work, SAE J1013 AUG92, Society of Automotive Engineers, Inc., Warrendale, PA.
28 SAE Standard, 1993, Vibration Performance Evaluation of Operator Seats,
SAE J1384 JUN93, Society of Automotive Engineers, Inc., Warrendale, PA.