You are on page 1of 9

948

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

An H-Infinity-Based Control Design Methodology Dedicated to the Active Control of Vehicle Longitudinal Oscillations
D. Lefebvre, P. Chevrel, and S. Richard

AbstractThis paper deals with the problem of active damping of vehicle oscillations. A complete methodology based on an optimization is presented. Several dedicated analysis tools are used in order to analyze the behavior of the closed-loop system. Two tuning parameters allow the design of a controller managing the compromise between performance (in terms of oscillation attenuation and limit cycles) and robustness. Finally, a feedforward component is designed in order to improve the performance without changing the robustness properties or the controller complexity. Simulation results obtained with an experimentally validated model show the efficiency of the resulting controller.

Index TermsDescribing functions, driveability, frequency domain analysis, H-infinity control, road vehicle control, robustness.

I. INTRODUCTION MBEDDED processors in current vehicles allow the implementation of advanced control strategies in order to improve vehicle behavior. However, the lack of methodology restrains the use of such strategies and few papers demonstrate this interest in the context of vehicle dynamics control [1]. This study forms part of a more general project undertaken by the French automobile manufacturer PSA Peugeot-Citron focusing on control methodology. The control design stage has to be more and more efficient to cope with the increasing complexity of the control strategies as the development time for a new vehicle becomes shorter. The vehicle driveline allows the transfer of engine torque to the wheels. Resonance in the elastic parts of the driveline must be handled carefully. In particular, one significant aspect of driveability is the attenuation of the first torsional mode which produces unpleasant (0 to 10 Hz) longitudinal oscillations of the car, known as shuffle. These oscillations are a cause of discomfort and generally occur during transient changes in driver demand (tip-in and tip-out) or on clutch engagement. The longitudinal oscillations are reduced by means of a controller acting on the engine torque. More generally the problem of flexible transmission in the context of electrical engineering has already been studied, e.g., in [2][4]. On the other hand, the problem
Manuscript received January 14, 2001. Manuscript received in final form February 7, 2003. Recommended by Associate Editor M. Jankovic. D. Lefebvre is with the Automatic Control Department, Institute of Research in Communication and Cybernetics (IRCCyN), UMR CNRS 6597, 44321 Nantes Cedex 3, France and also with the Research Department (DRIA), PSA Peugeot Citron, 78140 Vlizy Villacoublay, France (e-mail: Damien.lefebvre@mpsa.com). P. Chevrel is with the Automatic Control Department, Institute of Research in Communication and Cybernetics of Nantes (IRCCyN), France and also with the Ecole des Mines, 44307 Nantes Cedex 3, France (e-mail: philippe.chevrel@irccyn.ec-nantes.fr). S. Richard is with the Research Department (DRIA), PSA Peugeot Citron, France (e-mail: Sebastien.Richard@mpsa.com). Digital Object Identifier 10.1109/TCST.2003.815552

Fig. 1. Simplified model of the powertrain.

with a Spark Ignited engine has been rarely examined. One can however note [5] in which is explained the problem for the first time and [6] explains works on trucks. A special effort on methodology has been carried out in order to facilitate the controller design. Indeed, the only way to ensure the success of a control design method in the automobile domain is to propose to the user only a few tuning parameters with clear effects. Some dedicated analysis tools were developed to evaluate the quality of the feedback action. It will then be possible to take them into account during the control design stage. The feedback action is not sufficient to ensure good performance and robustness. The introduction of a feedforward action based on optimization can improve performance. an This paper is organized as follows: first, the model of the powertrain and the control problems will be presented. Then a set of analysis tools to be used in Section VI will be introduced in Section III before the feedback and feedforward design (Section IV and Section V). The article will end with an application on a detailed non linear model of the powertrain, and some elements are considered to manage the implementation (Section VII). II. MODEL OF THE POWERTRAIN The model of the powertrain is a high order system including many non linearities. The first stage, before the controller design is model reduction. The use of classical model reduction tools (modal reduction, balanced realization [7]) which do not keep the physical meaning for the parameters of the simplified model is not appropriate. Preferably, a mechanical reduction is performed which leads to the 3-state model of Fig. 1 [8]. This simple model retains the major part of the activity index [9] of the system in the bandwidth considered for driveability. It corresponds to an inertia in rotation involving another inertia by way of a flexible connection and a backlash. Some assumptions were made to establish this simplified model. It is supposed that the

1063-6536/03$17.00 2003 IEEE

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

949

Fig. 2.

Bode diagram of

G(s) function engine speed and G

( ) function acceleration.

clutch is stuck and that the differential is blocked. The inertias of the gearbox driveshafts and of the wheels are neglected, thus, the high frequency dynamics are not modeled. (see Fig. 1) corresponds to the The input of the model of the torque provided by the engine. The angular velocity is the image of the engine speed while first inertia represents the vehicle longitudinal acceleration. Aerodynamic forces and the road profile are not considered. The linear model is considered to design the controller. The backlash is introduced afterwards to analyze the closed-loop system. Each parameter of this model is related to characteristics of and correspond to the inertia of the vehicle. the engine flywheel and to the equivalent inertia of the vehicle, (Nm/rad) is an aggregate stiffness related to respectively; the driveshafts, the clutch, the gearbox, and the tire stiffness; and (Nm/(rad/s)) include all existing frictions; (m) is the equivalent radius of the wheel through the reduction of the gearbox and of the back axle. Numerical values of each parameter are derived from the experimental data analysis and from information provided by the manufacturer. Each coefficient is uncertain. The frictions and are rough approximations. The value of the engine inis precisely known while the equivalent inertia of the ertia can change depending on the vehicle load. vehicle

Therefore, the engine speed and the vehicle acceleration are related to the torque input by the two transfer functions defined in (1).

(1) The polynomial can be rewritten as

The expressions of the damping ratio and of the natural frequency are approximated by (2). (2) They are valid as long as

950

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

2) Dedicated Margins: The robustness against dynamic unstructured uncertainties outside the bandwidth control can be checked thanks to given by the following expression: (3) It should be noted that shown in [11].
Fig. 3. Representation of the closed-loop.

overestimates the delay margin as

B. Nominal Performance Indicators The reduction of the oscillations is equivalent to a decrease in the resonance peak of the acceleration transfer function defined in (1) (see Fig. 2). For the nominal system, the reducof this peak is representative of the level of perfortion ratio mance. Its expression is given by

The corresponding Bode Diagrams (Fig. 2) highlight the risk of persistent oscillations near the resonant frequency (around 15 rad/s). The prototype concerned by this work has a specific hierarchical structure. The close control of the motor, a four-stroke engine, plans its action for the next motor cycle in order to optimize the performances in terms of consumption and pollution. As a consequence, the required torque is realized with an average delay corresponding to the time needed by the crankshaft to complete four half turns. This delay varies from 20 to 200 ms depending on the engine speed. A method to derive a discrete controller (taking into account the variable delay) from a fixed maximal delay continuous-time design has been proposed in [8]. The design will focus on the continuous controller by assuming the engine as a constant delay of 200 ms. The control scheme is defined in Fig. 3. The only measured . The controller modifies the signal is the engine speed torque required by the driver in order to limit the rise of osof the transfer cillations. More precisely, the damping ratio is modified by the feedback action. function The objectives of the feedback are to attenuate the unpleasant driveline oscillations; to preserve the dominant real mode; and to ensure performances despite uncertainties.

(4) where is the damping ratio of the flexible mode [see (2)]. is introduced to normalize the expression with regard to the magnitude of the peak. of the system must not be modified by the The constant controller. The controller action has to be selective at middle frequencies. The feedback control is chosen as active when the is modified by more than 10% by the corinput torque (see Fig. 3). Thus, its activity at a specific rective torque frequency can be measured through the complementary sen. Let us also intrositivity function as , with and defined by duce (5), to quantify the frequency bandwidth where the controller is specially active.

III. ANALYSIS INDICATORS The analysis indicators are classified into three main categories: the nominal performance and the robustness properties concerning stability and performance. , and are the sensitivity function, the complementary sensitivity function and the input sensitivity function, respectively. These indicators are detailed in [10]. A. Indicators of Stability Robustness 1) Traditional Margins: The gain, phase and modulus , ) are significant indicators margins ( to guarantee the stability of the closed-loop. The delay margin is representative of the robustness of the-loop with regard to parasitic pure delays and neglected time constants. This point is all the more important to be checked that the engine and its close-loop control introduce a time-varying delay.

(5) The engine speed is used for feedback (see Fig. 3). These data are measured on the engine flywheel with some parasitic noise. The closed-loop sensitivity to the measurement noises can be analyzed with the help of the dimensionless indicator defined in [12] (6) characterizes the capacity of The function the closed-loop system to transmit the noise when, roughly is representative of speaking, the control is active, while the total transmission of the noise. The lower the value of , the less sensitive the closed-loop is to the measurement noise. C. Robust Performance Indicators The following computations are based on the analysis. See [13] for the theory and for instance in [14], [15], for the compu. tation of 1) Computation of Guaranteed Damping Ratio: The small theorem on generalized stability [16], is applied to compute

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

951

Fig. 5. Standard interconnection structure.

Fig. 4.

Control design methodology.

the damping ratio of the close-loop poles which is guaranteed for worst-case parametric uncertainties. Conversely, the small theorem can be applied to determine the variation intervals of parameters, which will ensure the stability. 2) Nonlinear Analysis: The describing function method gives good results on this model [17]. But the result is only valid for the set of parameters considered. The concept of parametric robustness is not met. To fulfill this requirement, the robustness margin defined by (7) and introduced by Ferreres [18] is used. This consists in calculating the size of the uncertainties that lead to a , given value of the limit cycle. The weaker the margin the closer the system is to the limit cycle of characteristic . characterizes the distance of the closed-loop . It allows the deduction of the system to the limit cycle way in which the limit cycle will be modified when the model parameters change. (7) is the describing function of the nonlinearity and is the linear part of the model. IV. CONTROLLER DESIGN control is particularly well suited to the problem of frequency shaping. Some works demonstrate its interest in the context of flexible joints [2], [19] or active control of vibrations [20]. The task now consists in defining a methodology dedicated to the problem of active damping of vehicle drivetrain oscillations. Fig. 4 represents the whole procedure proposed here which will be detailed in the following. The feedback control analysis is a direct application of Section III. A. The Framework

The problem is then reduced to a standard optimization one as defined in [21], [22] and solved in [23] or in [24] with weakened assumptions. The standard scheme is recalled in Fig. 5. represents the signal to be constrained, is the external input. is the command and is the measured signal for the control. The corresponding standard optimization problem is to which internally stabilizes find the transfer matrix norm of the the closed-loop system and minimizes the closed-loop transfer function between and . The Linear Fractional Transformation [13], [25] allows the compact expression of the closed-loop transfer function. For methodological reasons, only the suboptimal problem will be concerned which, for a given , consists in finding such that (8) Having recalled this basic result, the methodology can now be presented. B. Weighting Functions As recalled in part 2, the feedback aims to actively damp the oscillations. However, the controller must guarantee the stability of the closed-loop despite neglected dynamics, parasitic delays, and uncertain gains. 1) Specifications: The control objectives cannot be definitive specifications a priori without knowing what is attainable. The specifications (Si) are mainly qualitative and concern: Performance: the first torsional mode of oscillation must be attenuated (S1) while the dominant time delay has to be preserved (S2). The Spark-ignited engine must not be solicited at high frequency (S3). The backlash effects must be reduced as much as possible (S4). Constraints: the control law has to work for all cars of the same type throughout their life cycle. This may be translated in robustness requirements, which include the robustness against neglected dynamics (S5), against parasitic gain, phase and delays (S6), and also against parametric uncertainties (S7). In particular, the zero of the simplified model is related to the weight of the vehicle which may vary. Moreover, the single measured signal is the engine speed. 2) Design: A direct design, taking into account a priori all the control objectives leads necessarily to overspecification and conservatism. Therefore, only main constraints are explicitly

control design consists in reformulating the control The criterion. The smaller the criterion, objectives through an the better the control objectives will be satisfied.

952

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

(in percent), and the desired unstructured robustness coefficient . It is a bound on the complementary modulus margin

is related to by the relation (11) where is the damping ratio associated to the resonant mode. It limits the resis onance peak to be reduced: a reduction of 70% usually sufficient. (11)
Fig. 6. Standard structure for the damping problem.

formulated and the number of tuning parameters is voluntarily limited to define a clear methodology for the designer. is The complementary sensitivity function constrained to ensure the stability margin. Indeed the maximum is related to classical margins. value of is included in the criterion The transfer function to specify performance. The oscillations of the powertrain are which can be associated related to the weak damping ratio to the maximum value of the transfer functions defined in (1) (see Fig. 2). Minimizing this maximum value will reduce the oscillations. is constrained to The input sensitivity function ensure robustness against additive and parametric uncertainties of the-loop. It allows, in particular, to shape in some point the . As the backlash frequency response of the controller effects are reduced when the feedback action decreases, it will be possible to reduce the magnitude of the controller at certain frequencies. These transfer functions are constrained thanks to three dif, , ferent transfer functions. They are, respectively, . The corresponding standard and the product : scheme (Fig. 6) is associated to the 4-block criterion (9)

is related to by . is the weighting function associated with T(s). As a , and the complementary consequence, the phase margin modulus margin will be guaranteed

(12) The robustness against additive uncertainty, and indirectly the robustness against some parametric uncertainties, is automati. Indeed, parametric uncertainties cally adjusted through (the vehicle weight for instance) induce an uncertainty on the . prevents a pole-zero canceltransmission zero of is lation, hence increasing robustness. A transfer function to reduce the effect of introduced in the expression of the feedback action in the frequency bandwidth where the limit has a high gain for a cycles appear. The transfer function specific pulsation and the width of the peak is adjustable. (13) All specifications from S1 to S7 are related to the weighting functions. The attenuation level of resonance with function answers S1. S2 and S3 are fulfilled by introducing a time conand to limit the action of stant in functions the controller at low frequencies. Specifications S4 to S7 are . The largest is achieved by the value of the function , the smallest is the function . Hence, the ro. bustness of the-loop is guaranteed by the function 3) Use of the Constraint Functions: With this in mind, the proposed methodology consists of solving a suboptimal problem with weighting functions automatically derived from the tuning parameters: choice of the performance level via the ; choice of a desired compleresonance peak value with or of a phase margin which immentary modulus margin ; maximization of the value of the static poses a value of of in order to obtain the greatest robustness with gain respect to additive uncertainties; and maximization of the time to limit the action of the controller at low constant of frequencies. criterion is near The iterations are stopped when the . This first controller being obtained, it can be analyzed through the analysis indicators defined previously. If one of them does not suit the designer, it is possible to modify one

with

The external inputs correspond to the input torque and to the noises on the measured signal (the engine speed). Signal contains the control input and the acceleration of the vehicle. The control signal is a torque which modifies the reference torque problem (8) required by the driver. Then, the suboptimal leads to applied to

(10) , and is not straightforThe choice of ward for a non expert designer. The proposed methodology makes use of only two tuning parameters. Here are their definitions. The minimal expected oscillation attenuation

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

953

Fig. 8. Tuning refinement.

Fig. 7.

Initial tuning and H

1 optimization.
Fig. 9. Structure for the prefilter design.

tuning parameter in order to converge to the desired robustness or performance. For instance, if the limit cycle characteristics are too large, it is possible to reduce them by introducing in the constraint , the function parameterized by the results function of the analysis tools. However, some other characteristic values will be degraded. The procedure is detailed in Fig. 7 and Fig. 8. The structure for the prefilter design is shown in Fig. 9. They optimization and the tuning refinement correspond to the stage of the control design methodology (Fig. 4). Some constraints come into conflict with others. It is necessary to reach a compromise between performance and robustness. The designer does this through manual adjustment of the parameters. Despite the efforts, not all oscillations are removed. A feedforward part, which will filter the reference input, will help to ensure performances without degrading the robustness. V. FEEDFORWARD DESIGN The feedforward design is performed here in order to norm) between the minimize the difference (in terms of

closed-loop system including and a reference model . Moreover, we are doing so that the feedforward part does not increase the complexity of the whole controller (order and compuand tational complexity for implementation). For that, are constrained to share the same dynamics:

(14) is the error between the reference model and the closed-loop system output, filtered by the transfer function . The mathematical formulation of the problem is derived from the bounded real lemma and can be reduced to the resolution of a linear matrix inequality (LMI) as shown in [26]. suboptimal prefilter synthesis problem has a soluThe such tion if and only if there exists a symmetric matrix that the constraint (14) and where under must be asymptotically stable.

954

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

TABLE I VALUES OF THE MODEL PARAMETERS

TABLE III RESULTS OF THE  ANALYSIS

TABLE II LIST OF THE INDICATORS

is unknown. All others matrices are derived from the state and [see (15)]. space representation of

(15) The matrix contains the free parameters of the prefilter: and . Thanks to this method, the feedback part is will only change not modified. So the feedforward part the zeros of the closed-loop system, not its poles. The robust stability properties are so unchanged. VI. APPLICATION The procedure detailed in Fig. 7 is now applied to the model of the powertrain represented in Fig. 1. The minimal admissible is taken as 0.765. As a complementary modulus margin consequence [see (12)] the phase margin will be greater than required is 70%. After optimizing 45 . The attenuation level and , the first part of the design is carried out the values of with the following constraint functions

Fig. 10. Step response of the vehicle acceleration with and without the controller.

Values of the model parameters are shown in Table I. The majority of the indicators are satisfying (see Table II). The required performances are achieved. For the computation of the feedforward part, the reference with an model is derived from the open-loop transfer . The optimization is carried increased damping ratio . The maximum value of the error between out with

and after optimization is about and the static gain is preserved. of the closed-loop system The results of the analysis are described in Table III. The type A results express the parameter intervals that ensure the stability. The type B results show the intervals for which the feedback system respects a damping ratio of 0.15 and the type C are obtained with the respect of constraints of the closed-loop equivalent to oscillation reduction function of at least 70%. All the intervals are large enough to include the admissible uncertainties. For instance an acceptable vehicle variations on load variation is equivalent to . The simulation model is composed of a simplified SI engine model and an accurate model of the powertrain including the This model gearbox, the clutch, engine mounting systems has been validated with real experiments [8] on a French car 406 of PSA Peugeot-Citron. So the results are considered to be very close to the real dynamics of the vehicle. This model differs from those of Fig. 1 simply through high frequency dynamics and nonlinearities thus the general dynamics is kept. A torque . The vehicle is in 1st gear step of 50 Nm is applied at and was stationary. Fig. 10 depicts a simulation of the vehicle acceleration during transients with and without the controller. The 1st oscillation is reduced whereas the others one are removed.

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

955

The computation of the controller observer-state feedback form from one given state space realization is carried out using a generalized Riccati equation [27], [28]. The controller poles must be classified between three categories, which are the observation gain, the filter gain and the Youla parameter. The following pole selection procedure has been adopted: the fast and unobservable poles are assigned to the estimation gain; and the poles that are close to the physical system and the uncontrollable poles are assigned to the state feedback gain. Hence, the controller sums up to three gain matrices ( , and ) and two transfer functions ( and ) associated to the observer state space representation. VIII. CONCLUSION A complete methodology has been proposed in order to deal with the problem of active control of driveline oscillations. optimization and analysis, the control design Based on procedure requires few tuning parameters with clear effects. An iterative procedure allows the designer to refine the first controller obtained from the performance and robustness analysis tools. The backlash is explicitly taken into account thanks to a robust describing function analysis during the control design. A prefilter design completes the procedure in order to improve the performance without changing the robustness properties and the controller complexity. The controller is realized in an observer-state feedback form for implementation. The entire procedure has been validated on a complete high order model of the driveline. The design method can be generalized to multi input systems, as can the majority of the analysis tools. This will be important for future prototypes which will include additional torque providers such as electric motors, piloted gearbox, etc. ACKNOWLEDGMENT This work is the result of a collaboration between PSA Peugeot Citron and the University of Nantes, France. REFERENCES
[1] D. Hrovat and E. Tobler, Bond Graph Modeling of Automotive Power Trains: Ford Motor Co., 1991, pp. 623663. [2] S. Font and G. Duc, H control of an electromechanical drive with nonlinearities using a multi-block criterion, Contr. Eng. Practice, vol. 2, no. 6, pp. 10191030, 1994. [3] G. Brandenburg and W. Kaiser, On PI control of elastic systems with notch filters in comparison with advanced strategies, in Proc. IFAC Motion Contr., 1995, pp. 863875. [4] F. Altpeter, F. Ghorbel, and R. Longchamp, Control of drives with flexible transmission, in Proc. IFAC Motion Contr., 1995. [5] C. Mo, A. Beaumount, and N. Powell, Active Control of Driveability, SAE, Tech. Paper series 960 046, 1996. [6] M. Petterson, L. Nielsen, and L. Hedstrm, Transmission-Torque Control for Gear Shifting With Engine Control, SAE, Tech. Paper series 970 864, 1997. [7] G. Obinata and B. Anderson, Model Reduction for Control Design. London, U.K.: Springer-Verlag, 2001. [8] S. Richard, Improvement of the driveability by active damping of the longitudinal vehicle oscillations, Thse de doctorat, IRCCyN Nantes, 2001. [9] L. Louca, J. Stein, and G. Hulbert, A physical-based model reduction metric with an application to vehicle dynamics, in Proc. IFAC Non Linear Contr. Syst. Symp., Enschede, 1998.

Fig. 11.

Step response of the vehicle acceleration.

Fig. 11 shows the acceleration response for different situations of the vehicle. Three percent road slope and a load of 200 kg were chosen in order to show the robustness of the controller. The general behavior is not affected, the first oscillation remains and the response time is kept, the robustness analysis is then confirmed with time simulations.

VII. IMPLEMENTATION Before discretization, the controller layout must be shaped in order to ensure a good numerical conditioning. Several layouts exist: the well-known transfer function structure corresponds to the companion state space realization of the system. The main advantage of this structure is the minimization of the execution time. However, the numerical conditioning can be unfavorable. Also, the state feedback observer structure seems more advantageous. It allows an enrichment of the observer model with a physical meaning. Indeed, these states estimate the states of the physical system. It improves the readability of the signals and the numerical conditioning will then be a priori better than the transfer function form. The states initialization of the controller is based on the physical states of the system, so the starting and the commutations to one controller to another (when the gear ratio changes for example) will be facilitated. For that, the state space realization of the third-order system [see (1)] together with a fourth-order Pade approximation of the delay in a balanced state space realization is used. The state space representation of the system is named ( , , ). The controller is then written in the observer-state feedback form: .

is the torque reference input from the driver. is is the state feedback gain and is the the engine speed. is the estimation estimation gain for the controller, while gain for the prefilter. The Youla parameter is represented by for the feedback and for the transfer functions the prefilter. These parameters exist only if the controller is of higher order than the observer.

956

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 11, NO. 6, NOVEMBER 2003

[10] D. Lefebvre, P. Chevrel, and S. Richard, Control analysis tools for active attenuation of vehicle longitudinal oscillations, in Proc. 10th IEEE Conf. Contr. Applicat., Mexico City, 2001. [11] Y.-P. Huang and K. Zhou, Robust control of uncertain time delay systems, in Proc. 38th Conf. Decision Contr., 1999, pp. 11301135. [12] P. De Larminat and S. Puren, Robust poles placement via LTR, in 3rd Robust Control Workshop, Poitiers, France, 1997. [13] K. Zhou, J. C. Doyle, and K. Glover, Robust and Optimal Control. Englewood Cliffs, NJ: Prentice-Hall, 1996. [14] P. M. Young and J. C. Doyle, Computation of  with real and complex uncertainties, in Proc. 29th Conf. Decision Contr., Honolulu, Hawaii, 1990, pp. 12301235. [15] G. Balas, J. Doyle, K. Glover, A. Packard, and R. Smith, -analysis and synthesis toolbox, in Users Guide With MATLAB, 1995. [16] G. Ferreres, V. Fromion, G. Duc, and M. MSaad, Application of real/mixed  computational techniques to a missile autopilot, Int. J. Robust and Nonlinear Contr., vol. 6, no. 8, pp. 743769, 1996. [17] H. K. Khalil, Nonlinear Systems, 2nd ed. Englewood Cliffs, NJ: Prentice-Hall, 1996, pp. 450468. [18] G. Ferreres, A Practical Approach to Robustness Analysis With Aeronautical Applications. New York: Kluwer Academic/Plenum, 1999, pp. 163181.

[19] B. Wie and D. S. Bernstein, Special issue on 2 mass benchmark problem, Int. J. Robust and Nonlinear Contr., vol. 5, no. 1, 1995. [20] W. K. Gawronski, Dynamics and Control of Structures. New York: Springer-Verlag, 1998. [21] B. A. Francis, A course in H control theory, in Lecture Notes in Control and Information Sciences. New York: Springer-Verlag, 1987, vol. 88. [22] B. A. Francis and J. C. Doyle, Linear control theory with an H optimality criterion, SIAM J. Contr. Opt., vol. 25, pp. 815844, 1987. [23] J. C. Doyle, K. Glover, P. Khargonekar, and B. A. Francis, State space solutions to standard H2 and H control problems, IEEE Trans. Automat. Contr., vol. 34, pp. 831847, 1989. [24] P. Gahinet and P. Apkarian, A linear matrix inequality approach to H control, Int. J. Robust and Nonlinear Contr., vol. 4, pp. 421448, 1994. [25] J. C. Doyle, A. Packard, and K. Zhou, Review of LFTs, LMIs and , in Proc. 30th IEEE Conf. Decision Contr., 1991, pp. 12271232. [26] A. Giusto, A. Neto, and E. Castelan, H and H2 design techniques for a class of prefilters, IEEE Trans. Automat. Contr., vol. 41, pp. 865870, 1996. [27] R. A. Bender, Some considerations for estimator-based compensator design, Int. J. Contr., vol. 41, pp. 15771588, 1985. [28] D. Alazard and P. Apkarian, Exact observer-based structures for arbitrary compensators, Int. J. Robust Nonlinear Control, vol. 9, pp. 101118, 1999.

You might also like