Você está na página 1de 9

Chul Kim

General Motors Co.

An Accurate Full Car Ride Model Using Model Reducing Techniques


In this study, an approach to obtain an accurate yet simple model for full-vehicle ride analysis is proposed. The approach involves linearization of a full car MBD (multibody dynamics) model to obtain a large-order vehicle model. The states of the model are divided into two groups depending on their effects on the ride quality and handling performance. Singular perturbation method is then applied to reduce the model size. Comparing the responses of the proposed model and the original MBD model shows an accurate matching between the two systems. A set of identied parameters that makes the well-known seven degree-of-freedom model very close to the full car MBD model is obtained. Finally, the benets of the approach are illustrated through design of an active suspension system. The identied model exhibits improved performance over the nominal models in the sense that the accurate model leads to the appropriate selection of control gains. This study also provides an analytical method to investigate the effects of model complexity on model accuracy for vehicle suspension systems. DOI: 10.1115/1.1503065

Paul I. Ro
Department of Mechanical Engineering, North Carolina State University, Raleigh, NC 27695

Introduction

Ride and handling performances of a vehicle suspension system often present engineers conicting design goals. For decades, engineers have studied active control for vehicle suspension systems to resolve this conict. The intent of the system is to replace the classical passive elements by a controlled system that can supply unlimited force to the system. This type of active suspension system has proven capable of achieving improvements over passive systems. For the analysis and the synthesis of suspension systems, model accuracy becomes the most important measure. With the help of multibody dynamic software such as ADAMS, it is possible to obtain a detailed vehicle suspension model. The model is able to incorporate even complicated geometric properties such as roll steer, camber effects and other compliance properties. However, it is not appropriate for the synthesis of control systems due to its complexity. On the other hand, most suspension models used for controller synthesis studies are generally low-order, linear models. A large percentage of the literature on active suspension systems has used the two-mass model. This model has been assumed to be simple yet effective in representing the two dominant modes sprung mass bouncing and wheel hopping of a quarter car system. In most research of active suspension system design, the set of equivalent parameters of the simple model is assumed to be available. For example, the spring stiffness and the damping coefcient of the two-mass model are obtained based on component data. Also, the unsprung mass is calculated using the summation of suspension linkages. However, the simple nominal model may not be as effective as one might expect without considering the inuence of the suspension kinematic structure 1 . The result of the study shows that the suspension kinematic structure has a greater impact as a suspension system becomes more complex e.g., double wishbone, multilink type, etc. . Any inaccuracy in the model can cause an undesirable tuning process when a controller is incorporated into the actual system. Therefore, obtaining both an accurate and simple model for a given system is very important in reducing the development cost and time. For the quarter car system, a parameter identication
Contributed by the Mechanisms Committee for publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received October 2000. Associate Editor: S. K. Agrawal.

method and model reduction methods have been used to obtain a two-mass model with accurate suspension parameters 2,3 . The simple model resulted in accurate prediction of system behavior after incorporating the control law. Here, the focus is turned to obtaining a full car ride model with low-order structure. As mentioned above, a high-order model is needed in order to compensate for the coupling effects of a full vehicle system such as roll and pitch. Several researches have adopted full car or half car suspension models for designing active suspension control laws 4,5 . In the majority of recent vehicle control systems, each subsystem, such as ABS or active suspension, is designed and controlled, separately. However, a high-order model is essential to capture fully the combined effect of these subsystems. For example, the combining effect of ABS and active suspension systems has been shown using the half car model 4 . Also, in order to compensate for the roll characteristics e.g., roll axis, roll stiffness distribution, etc. , a full car model with roll mode is essential as shown in 5 . An extended control law can be obtained using the model, which is not possible by adopting a quarter car model. In most cases, however, the models for the full car system are obtained in the same way as the nominal two-mass model for a quarter car system. For example, a 7-DOF degree-of-freedom model was used in 5 to synthesize an active control law which considers the roll and pitch behaviors. The roll axis was simplied to be horizontal, and the roll stiffness data was embedded into the equation of motion based on component data. Clearly, this kind of nominal model is not able to compensate different suspension structures. This may cause the inaccuracy of the model, which will require unnecessary effort and time. A system identication method has been applied to obtain a set of equivalent parameters 6 . Using several different objective functions, different sets of identied parameters were obtained and compared in the frequency domain in this study. In the study, however, the simulation model used to verify the method has the same model complexity as the identication model. Therefore, it remains uncertain what may happen when the method is applied to more complex systems with suspension linkages included. In this study, an alternative mean of obtaining equivalent parameters is proposed. It is derived from multibody dynamics modeling of a full car system that includes a complex suspension system. Linearization and model reduction techniques are applied to obtain an accurate yet simple model for the full car ride simuDECEMBER 2002, Vol. 124 697

Journal of Mechanical Design

Copyright 2002 by ASME

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

lation and synthesis. Using the proposed approach, it would be more convenient to include the roll and pitch relating characteristics because they are automatically embedded in the reduced-order model. The rest of this paper is organized as follows. In section 2, a reduced-order model for full car ride behavior is developed. In section 3, a set of identied parameter is obtained by comparing the entries of the reduced-order model and the 7-DOF model. In section 4, the design process for active suspension systems is shown in order to illustrate the benets of the accurate, simple model. The conclusion reviews and recommends further topics.

where, X and U are system state vector and control inputs, respectively. Linearizing the equations around an equilibrium position, largeorder linearized equations of motion can be obtained as X AX BU, y CX (2)

where, A f / X Rn n , B f / U Rn 4 , C g/ X 7 n R , and n is the number of states twice the degree of freedom . The control input U is the force vector generated by the four actuators installed parallel to the struts such that, U u1 u2 u3 u 4 T. (3)

2 Reduced-Order Modeling for Full Car Ride Analysis and Synthesis


2.1 MBD Model and Simple Model for Full Car Suspension System. Figure 1 shows a complex multibody dynamic model used to analyze a full car system that includes suspension linkage and connecting elements. With the help of commercial software it is possible to model realistically and simulate the fullmotion behavior of the complex mechanical system. Despite its merit of accuracy, the multibody dynamic model has rarely been used for the purpose of controller synthesis due to its nonlinear property and complexity. For example, the degree of freedom of the full car system shown in Fig. 1 increases signicantly as the kinematic joints are replaced with bushing elements. Also, the equations of motion of the model are highly nonlinear and difcult to manipulate. Instead, a simple linear model is mostly used to design a control system in practical situations. For example, the 7-DOF model shown in Fig. 2 has been frequently used for a full car suspension system, where roll, pitch, and the vertical displacements of sprung mass and four unsprung masses are included. A key issue addressed in the present paper is to determine a set of equivalent parameters that makes the simple model as close to the original system as possible. 2.2 Linearization of the MBD Model. A complex model Fig. 1 is described by a set of nonlinear equations as follows: X f X,U , y g X (1)

The system output y is dened as a 7 1 vector of roll, pitch, and the vertical displacements of the sprung mass and unsprung masses. For simplicity, kinematic joints have been used to build the complex MBD model. The corresponding states of the linearized equations 2 are composed of 20 states as follows. X xs ys
s x y

z s z u1 z u2 z u3 z u4

s s z s z u1 z u2 z u3 z u4 T , where the subscripts s and u stand for the sprung mass and the unsprung mass, respectively. The denitions of variables are shown in Table 1. Notice that the states can be divided into two groups: handlingrelated states and ride-related states. The rst three states x s , y s , and s and their velocity terms are related to the handling performance. The rest of the states are closely related to the ride quality. In fact, the roll angle is coupled with both ride and handling characteristics. In order to compensate for the coupling effect and to obtain a reduced-order ride model, model reduction techniques are applied. 2.3 Reduced-Order Model for Full Car Suspension System. Model reduction has been a major topic in systems theory for decades. Various kinds of techniques have been developed since it rst appeared in the late 1960s 8 10 . Dominant mode techniques were the typical methodologies in the beginning. Singular perturbation was also applied to model reduction 11 . The method is very similar to the dominant mode technique in the sense of separating the modes based on the size of eigenvalues 11 . The balanced realization technique is applied to the model reduction by Moore 12 which is based on measures of controllability and observability of the system. In the previous study 2 , model reduction techniques have been applied to obtain an accurate set of equivalent parameters for the two-mass model. An accurate yet simple handling model of a full vehicle system was obtained using the same method. The results show that the reduced-order models represent the original system more accurately compared to the nominal counterparts. Here, a similar process is applied to a full vehicle system to obtain an accurate 7-DOF ride model. Among the techniques mentioned, singular perturbation is selected in this study. The technique is simple to understand and

Fig. 1 ADAMS complex model for full car ride quality

Table 1 The denitions of variables

698 Vol. 124, DECEMBER 2002

Transactions of the ASME

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

if a dominant mode technique is applied. The balanced realization technique has a similar difculty since it requires the system to be strictly stable. The large-order system can be expressed by separating two state groups as follows. P XR P XH or
Fig. 2 Simple full car ride model 7-DOF model

A 11 A 21

A 12 A 22

XR XH

B1 U B2

(4)

P XR A 11XR A 12XH B 1 U P XH A 21XR A 22XH B 2 U


T T

(5a) (5b) xs ys
s

retains the original denition of the states. This is a very useful characteristic in designing a control system. Since some system states need to be measured for feedback, it is desirable for the reduced-order model to be expressed by the original system states. The dominant mode technique is inappropriate for this case, because the system contains some semidenite modes that need to be remained. Using the dominant mode technique the semidenite modes with zero natural frequencies may remain since choosing the remaining states is based on the size of eigenvalues. For example, some handling of related states with zero eigenvalues, such as longitudinal displacement/velocity are to be remained

where XR

z s z u1 z u2 z u3 z u4 , XH
T XR

XR

T XR T ,

XH

T XH

T XH T ,

and the subscripts R and H represent the ride-related states and the handling-related states, respectively. Using a typical singular perturbation technique, Eq. 5b be comes an algebraic relationship between two state groups, XR and XH . Applying the static relationship into Eq. 5a , a reducedorder model for full car ride characteristic is obtained. P XR A R XR B R U, A 12A 221 A 21 , A 12A 221 B 2 (6)

where, A R A 11 BR B 1 , and the numerical data are shown in Appendix A. Figures 3 and 4 show the corresponding mode shapes of the reduced-order model, which clearly reect the bouncing, pitching, roll, and the vertical modes of the unsprung masses. Figure 5 shows the responses of the large-order MBD model and the reduced system for an actuator input, where a step force is given to the front right actuator. It can be seen from the result that the reduced-order full car model represents the original system with considerable accuracy. In the following section, the nominal model will be obtained based on the component data set, then it will be compared with the reduced-order model in terms of model accuracy.

3 Equivalent Parameters of Full Car Ride Model from Reduced-Order Model


Many of the ongoing research studies about designing a control system for a full car system have used the well-known 7-DOF model shown in Fig. 2 5 . As mentioned in Section 2, the accuracy of this model depends on the selection of the equivalent parameters such as m s , I p , I r , k s f , b s f , . . . , etc. that are provided by part vendors in most cases. However, this way cannot compensate for the change of suspension structures. For example, a different suspension structure with the same component data would result in the same nominal model. This may cause the parameter uncertainty of the nominal model, which leads to a prolonged tuning process. In this section, inaccuracy of the nominal model will be illustrated rst. Then, the response of the nominal model will be compared with the reduced-order model. Finally, a set of identied parameters will be obtained, which will match the 7-DOF model as close to the original system as possible. 3.1 Nominal Model for Full Car Suspension System. The equations of motion for the full car ride model Fig. 2 can be derived as follows. For the bouncing of the sprung mass: m s z s b f z s1 z u1 k f z s1 z u1
Fig. 3 Mode shape of the reduced-order model I sprung mass motions

b f z s2 z u2 k f z s2 z u2

b r z s3 z u3 k r z s3 z u3

b r z s4 z u4 k r z s4 z u4 (7)

u1 u2 u3 u4 For the pitching motion of the sprung mass:

Journal of Mechanical Design

DECEMBER 2002, Vol. 124 699

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 4 Mode shapes of reduce-order model II unsprung mass modes

l p s b f a z s1 z u1 z u4

b f a z s2 z u2

b r b z s3 z u3

b r b z s4

For the rolling motion of the sprung mass: I r s b f T f z s1 z u1 b f T f z s2 z u2 k f T f z s1 z u1 k r T r z s4 z u4 (9) b r T r z s3 z u3 k f T f z s2 z u2

k f a z s1 z u1

k f a z s2 z u2

k r b z s3 z u3 (8)

k r b z s4 z u4

au 1 au 2 bu 3 bu 4

b r T r z s4 z u4 k r T r z s3 z u3

T f u 1 T f u 2 T ru 3 T ru 4 For each side of wheel motion vertical direction : m u f z u1 b f z s1 z u1 m u f z u2 b f z s2 z u2 m ur z u3 b r z s3 z u3 m ur z u4 b r z s4 z u4 where z s1 T f z s3 T r


s s

k f z s1 z u1 k f z s2 z u2 k r z s3 z u3 k r z s4 z u4 z s , z s2 Tf
s

k t f z u1 k t f z u2 k tr z u3 k tr z u4 a Tr
s s

u 1 k t f z r1 u 2 k t f z r2 (10) u 3 k tr z r3 u 4 k tr z r4 zs , b
s

a b

s s

z s , and z s4
1

zs

Rewriting the equations of motion in the form of 6 , we get, XR M


1

C XR M

K XR U G Z r , B

(11)

Fig. 5 The comparison of single actuator u 1 step response ADAMS model and a reduced-order model

where, the detailed expressions of the above matrices are shown in Appendix B. Table 2 shows the parameter set used in the nominal model, which is the same as the component data of the original ADAMS model. Transactions of the ASME

700 Vol. 124, DECEMBER 2002

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Table 2 The parameter values of the nominal model

Figure 6 shows the responses of three different models when the same input is given as in Fig. 5. The gure shows the large discrepancies between the responses of the original system and the nominal model. 3.2 Identied Simple Model for Full Car Suspension System. The goal is to obtain a set of equivalent parameters for the 7-DOF model such that its response is as close to the reducedorder model as possible. In this section, the relationship between two models will be investigated to achieve the goal. This way, the effectiveness of the 7-DOF model will be also veried. The system equations 11 of the 7-DOF model can be expressed as follows. XR XR 0 M
1

I K M
1

XR XR

0 U. B

(12)

The equations of the reduced-order model are obtained from the model reduction process. 8 X R A R R B R U. X (13)

Notice that the entries of matrices of Eq. 13 are lled with numbers. On the other hand, the elements of Eq. 12 are algebraic expressions in terms of suspension parameters such as m s , I p , I r ,

k s f , b s f , . . . , etc. Here, the parameters are to be identied by comparing with the numerical values of Eq. 13 . As seen from the comparison of two sets of elements, it is not easy to calculate the estimated parameters by choosing any element pairs. For example, the counterparts of some zero element of the simple model have some numerical values, which cannot be neglected. This is not surprising, because the effects of suspension geometric characteristics such as the lengths of linkages are included in the reduced-order model. In fact, there can be very different results of parameter identication depending on the choice of elements and their order. A process of obtaining such identied parameters is proposed in this study. The detail of the process is shown in Appendix C. The process is obtained according to the same logic that was used to obtain identied parameters for the quarter car system 2 . Table 3 shows the identied parameter values by matching the reduced-order model to the 7-DOF model. There are some differences between the two sets of parameters, specically in the tread terms such as T f , and T r . These parameters are closely related to the roll stiffness of suspension systems. It can be seen from the results that the original system has less tread with similar roll stiffness than the simple model. This tendency is mainly due to the suspension linkage structure of the original system. That is, the linkage structure acts as a kinematic constraint, which increases the roll stiffness. On the other hand, there is no such constraint in the roll motion of the 7-DOF model. As a result, greater tread values are required for the simple model to match the roll characteristic while keeping the bouncing motion closely to the original system with similar spring stiffness terms, k s f and k sr . System responses of the 7-DOF model with the identied parameters are compared with the ones of the reduced-order model in Fig. 7. The results show that there is little difference between the two responses. The results proves that the model structure of the 7-DOF model is effective enough to represent the full car suspension system. However, determining an accurate set of parameter is very crucial. The similar results can be veried using the frequency related parameters such as the natural frequency and the damping coefcient of each mode Fig. 8 . The gure shows a good matching of the natural frequencies between the original system and the identied simple model.

Design of Controller

Fig. 6 The comparison of single actuator u 1 step response ADAMS model, a reduced-order model and the nominal model

In this section, the merits of the accurate reduced-order model are illustrated in designing active suspension systems. In general, an accurate model leads to a more efcient design process of active systems. In a linear control system, an inaccurate model results in the error between the prediction and the actual responses

Table 3 The identied parameter values from the reduced-order model

Journal of Mechanical Design

DECEMBER 2002, Vol. 124 701

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 7 The comparison of single actuator u 1 step response ADAMS model, a reduced-order model and an identied 7-DOF model

Fig. 9 The comparison of the road step responses Passive, active based on nominal model and active based on identied model

of the system output. Even in the design of a robust control system such as sliding mode control, obtaining an accurate model becomes crucial because the selection of the switching gain depends on the system uncertainties 1,3 . Using the model structure for the 7-DOF model, two different equations of motion can be obtained depending on the sets of suspension parameters i.e., nominal data set or identied data set . P XR A R X R B R U G R Z r . (14)

Here, the control law is to determine the control gain matrix, K, of U KXR , such that, P XR A R B R K XR G R Z r , (15)

where, the closed-loop system in Eq. 15 is assumed to have the desired characteristics. In this section, a typical pole placement method is applied to calculate the control gains. Two sets of eigenvalues of two systems are listed in Table 4. The three values of each column represent the pitch, bounce and roll mode, respectively. Assume that the desired pole locations of the active system are increased by 50% of the rst three modes of the passive system. The third column of Table 4 represents the desired pole locations. Figure 9 shows the response of two active systems compared with the responses of the passive system. The results clearly show the merits of using the identied model for the controller design. Using the identied model, all responses behave in the desired direction in terms of increasing the fastness of the responses and reducing the steady state errors. On the other hand, some responses of active system become worse than the passive system using the nominal model. In real situations, such tendency leads to unnecessary tuning when the control law is incorporated into the suspension system.

Conclusions

Fig. 8 The comparison of frequency characteristics ADAMS model and an identied 7-DOF model Table 4 Eigenvalue sets of identied model and nominal model with desired pole locations

In this paper, an accurate 7-DOF model for the ride analysis of a full vehicle is obtained. Linearization and model reduction techniques are used to nd a simple version of a complex full car model. First, a large-order model for a full vehicle is obtained by linearizing an ADAMS model. Second, the model is divided into two state groups: ride-related states and handling-related states. Among these, the ride-related states are chosen using the singular perturbation method. Then, a reduced-order model with 7 degree of freedom is obtained. Finally, by comparing the reduced-order model and the well-known 7-DOF full car model, a set of equivalent parameters is obtained. It is shown from the results that the 7-DOF model efciently represents the full car behavior only when the equivalent parameters are well chosen. An identied set of equivalent parameters makes the 7-DOF model closely comparable to the original sysTransactions of the ASME

702 Vol. 124, DECEMBER 2002

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

tem. Also, a signicant effect of the kinematic structure on rollrelated characteristics such as roll stiffness, front/rear tread has been revealed. The merits of identifying the parameters are demonstrated by developing active suspension systems based on the identied model and the nominal model. The identied model exhibits improved performance over the nominal models in the sense that the accurate synthesis model provides an accurate prediction of the control system behavior and the appropriate selection of control gains. As can be expected from the relationship between the roll property and the front/rear tread, other interesting relationships may exist between the equivalent parameters and the full vehicle characteristics. This tendency is mainly due to the effect of suspension kinematic structures. Investigating such relationships remains as future work. Also, the merits of the identied simple model need to be investigated further. For example, the model may also provide benets in the process of designing control systems with an observer.

T f , Tr a, b X y U A, B, C xs , x ys ,
s y

s, s b f , br k f , kr k t f , k tr XR , XH XR , X H

M , C, K k f , kr , . . .

Nomenclature
ms m u f , m ur Ip , Ir zs z si z ui sprung mass front/rear unsprung mass pitch/roll moment of inertia vertical displacement of the sprung mass vertical displacement of the sprung mass at each corner (i 1,2,3,4) vertical displacement of the unsprung mass (i 1,2,3,4)

front/rear tread distance to front/rear wheel center from e.g. of sprung mass state variable of large-order system output variable of large-order system actuator control input vector system matrices for large-order system longitudinal displacement/velocity of the sprung mass lateral displacement/velocity of the sprung mass roll, pitch and yaw angle of the sprung mass front/rear damping coefcients front/rear spring stiffness front/rear tire spring stiffness ride and handling related state vector augmented ride and handling related state vector equivalent parameter matrices of the 7-DOF model estimated equivalent parameters

Appendix A Numeric Data of the Reduced-Order Model


From Eq. 6 , XR A R XR B R U 0 A R21 I A R22 XR 0 U B R2 (A1)

1.0706e 02 2.2433e 06 2.5513e 05 A R21 2.7986e 02 2.7986e 02 1.6028e 02 1.6028e 02 1.1298e 01 2.6633e 07 1.8123e 06 A R22 2.8809e 01 2.8809e 01 2.5835e 01 2.5835e 01

2.0515e 04 4.6844e 01 3.8182e 00 4.1932e 02 4.1932e 02 3.6317e 02 3.6317e 02 1.7834e 05 5.9782e 00 1.3917e 00 4.0776e 01 4.0776e 01 5.5747e 01 5.5747e 01

1.7540e 04 1.4745e 00 5.4940e 01 3.7256e 02 3.7256e 02 2.3258e 02 2.3258e 02 1.6983e 05 7.5472e 01 6.5557e 00 3.7274e 01 3.7274e 01 3.4299e 01 3.4298e 01

4.3958e 01 1.0976e 01 2.1392e 01 3.9421e 03 8.6519e 02 9.1650e 00 5.8521e 00 3.5447e 00 8.8263e 01 1.6917e 00 3.8042e 01 7.9831e 03 7.2991e 01 4.8108e 01

4.3955e 01 1.0975e 01 2.1391e 01 8.6596e 02 3.9421e 03 5.8520e 00 9.1649e 00 3.5447e 00 8.8262e 01 1.6917e 00 7.9890e 03 3.8042e 01 4.8108e 01 7.2990e 01

4.9268e 01 2.4035e 01 3.5828e 01 1.5022e 00 6.7459e 01 2.6831e 03 1.3463e 01 3.8421e 00 1.2524e 00 1.5769e 00 9.8189e 02 7.1560e 02 3.5484e 01 3.9428e 01

4.9268e 01 2.4035e 01 3.5828e 01 6.7459e 01 , 1.5021e 00 1.3463e 01 2.6831e 03 3.8421e 00 1.2524e 00 1.5769e 00 7.1560e 02 9.8190e 02 3.9428e 01 3.5484e 01 ,

1.2987e 003 3.2338e 004 6.1982e 004 B R2 1.3658e 002 2.9272e 006 2.6742e 004 17.627e 004

1.2987e 003 3.2338e 004 6.1982e 004 2.9272e 006 1.3658e 002 1.7627e 004 2.6742e 004

1.3470e 003 4.3971e 004 5.5409e 004 3.4445e 005 2.5069e 005 1.2200e 002 1.3889e 004

1.3470e 003 4.3971e 004 5.5409e 004 2.5069e 005 3.4445e 005 1.3889e 004 1.2200e 002 ,

Appendix B The Equations of Motion for the Nominal Model


From Eq. 11 , XR M where, M diag(Ir ,Ip ,ms ,muf ,muf ,mur ,mur), DECEMBER 2002, Vol. 124 703
1

C XR M

K XR U G Z r , B

(B1)

Journal of Mechanical Design

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

2 2b f T 2 2b r T r f

0 2b f a
2

0 2b r b
2

bfTf bfa bf bf 0 0 0 kfTf kfa kf kf 0 0 0 0 0 0 0 0 0 0 ktf 0 0 ktf kf 2b r

bfTf bfa bf 0 bf 0 0 kfTf kfa kf 0 ktf 0 0 0 0 0 0 0 k tr 0 0 0 0 0 0 0 k tr

b rT r b rb br 0 0 br 0 k rT r k rb kr 0 0 k r k tr 0

b rT r b rb br 0 0 0 br k rT r k rb kr 0 0 0 k r k tr ,

0 0 C bfTf bfTf b rT r b rT r
2 2k f T 2 2k r T r f

2b f a 2b r b 2b f bf bf br br 0

2b f a 2b r b bfa bfa b rb b rb 0 2k f a
2

0 0 K kfTf kfTf k rT r k rT r

2k r b kfa kfa

2k f a 2k r b 2k f 2k r kf kf kr kr

2k f a 2k r b

k rb k rb Tf a 1 Tf a 1 0 1 0 0 Tr b 1 0 0 1 Tr b 1 0 0 0 1

1 0 0 0

ktf 0 0 0

Appendix C Equivalent Parameter from the ReducedOrder Model


Rewrite the equation of motion B1 for 7-DOF model as follows: XR XR 0 M
1

I p a/R R 9,1 Ir t b f f /A R 8,12

(C2)

The steady state expression of roll angle for the 7-DOF system can be obtained using Maple.
s ss

I K M
1

XR XR

0 U B

(C1) or tf

k t f k r k tr /t f 1 2 k f k t f k r k tr k r k tr k f

ktf

(C3)

By comparing the elements of two systems A1 and C1 , the following set of identied parameters are obtained. muf m ur bf br a b kf 1/B R 11,1 1/B R 13,3 A R 12,12 m u f A R 14,14 m ur A R 12,9 /A R 12,12 A R 14,9 /A R 14,14 A R 12,3 m u f

tr

1 2
s ss

k t f k r k tr k f k t f k r k tr k r k tr k f ktf

(C4)

References
1 Kim, C., Ro, P. I., and Kim, H., 1999, Effect of Suspension Structure on Equivalent Suspension Parameters, Journal of Automobile Engineering, IMechE, 213 D , pp. 457 470. 2 Kim, C., and Ro, P. I., 2000, Reduced Order Modeling and Parameter Estimation for Quarter Car Suspension System, Journal of Automobile Engineering, IMechE, 214 D , pp. 851 864. 3 Kim, C., and Ro, P. I., 2001, An Accurate Simple Model for Vehicle Handling Using Model Reduction Technique, Transactions of the SME, Journal of Passenger Cars, in press. 4 Alleyne, A., 1997, Improved Vehicle Performance Using Combined Suspension and Braking Forces, Veh. Syst. Dyn., 27, pp. 235265. 5 Williams, D. E., and Haddad, W. M., 1997, Active Suspension Control to Improve Vehicle Ride and Handling, Veh. Syst. Dyn., 28, pp. 124. 6 Demic, M., 1997, Identication of Vibration Parameters for Motor Vehicles, Veh. Syst. Dyn., 27, pp. 65 88. 7 Crolla, D. A., and Abdel-Hady, M. B. A., 1991, Active Suspension Control; Performance Comparisons Using Control Laws Applied to a Full Vehicle Model, Veh. Syst. Dyn., 20, pp. 107120. 8 Davison, E. J., 1966, A Method for Simplifying Linear Dynamic Systems,

k r A R 14,3 m ur ktf k tr A R 12,5 m u f kf

A R 14,7 m ur k r m s 1/B R 10,1

704 Vol. 124, DECEMBER 2002

Transactions of the ASME

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

IEEE Trans. Autom. Control, 11, pp. 93101. 9 Chen, C. F., and Shieh, L. S., 1968, A Novel Approach to Linear Model Simplications, Int. J. Control, 8 6 , pp. 561570. 10 Bonvin, D., and Millichamp, D. A., 1982, A Unied Derivation and Critical Review of Modal Approaches to Model Reduction, Int. J. Control, 35, pp. 829 848.

11 Kokotovic, R. E., OMalley, R. E., and Sannuti, P., 1976, Singular Perturbations and Order Reduction in Control TheoryAn Overview, Automatica, 1, pp. 123132. 12 Moore, B. C., 1981, Principal Component Analysis in Linear Systems: Controllability, Observability and Model Reduction, IEEE Trans. Autom. Control, 26, pp. 1732.

Journal of Mechanical Design

DECEMBER 2002, Vol. 124 705

Downloaded 16 Feb 2009 to 130.237.29.138. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Você também pode gostar