This paper presents a new fuzzy controller for semi-active vehicle suspension systems. The proposed fuzzy controller has only nine fuzzy rules, whose performance is equivalent to the existing fuzzy controller with 49 fuzzy rules. Simulation results show that the ride comfort and road holding are improved by 28% and 31deg.4, respectively, with the fuzzy controlled semi-active suspension system.
This paper presents a new fuzzy controller for semi-active vehicle suspension systems. The proposed fuzzy controller has only nine fuzzy rules, whose performance is equivalent to the existing fuzzy controller with 49 fuzzy rules. Simulation results show that the ride comfort and road holding are improved by 28% and 31deg.4, respectively, with the fuzzy controlled semi-active suspension system.
This paper presents a new fuzzy controller for semi-active vehicle suspension systems. The proposed fuzzy controller has only nine fuzzy rules, whose performance is equivalent to the existing fuzzy controller with 49 fuzzy rules. Simulation results show that the ride comfort and road holding are improved by 28% and 31deg.4, respectively, with the fuzzy controlled semi-active suspension system.
International Conference onMechatronics and Automation
August 9 - 12, Changchun, China Fuzzy Control of Semi-Active Automotive Suspensions Aws Abu-Khudhair, Radu Muresan, and Simon X. Yang School ofEngineering University ofGuelph Guelph, Ontario LiG 2Wi, Canada {aabukhud, rmuresan, syang}@uoguelph.ca Abstract - This paper presents a new fuzzy controller for semi-active vehicle suspension systems, which has a significantly fewer number of rules in comparison to existing fuzzy controllers. The proposed fuzzy controller has only nine fuzzy rules, whose performance is equivalent to the existing fuzzy controller with 49 fuzzy rules. The proposed controller with less number of fuzzy rules will be more feasible and cost-efficient in hardware implementation. For comparison, a linear quadratic regulator controlled semi-active suspension, and a passive suspension are also implemented and simulated. Simulation results show that the ride comfort and road holding are improved by 28% and 31.4, respectively, with the fuzzy controlled semi-active suspension system, in comparison to the linear quadratic regulator controlled semi-active suspension. Index Terms - Passive Suspension System, Semi-Active Suspension System, Linear Quadratic Regulator, Fuzzy Control. I. INTRODUCTION The main function of suspension systems is to isolate the disturbances from the road to the passenger, while maintaining good vehicle handling and road holding. Currently most commercial vehicles employ a standard passive suspension system consisting of purely passive elements (spring and damper). Inherent to the passive suspensions is a trade-off between ride quality and road holding characteristics. Soft springs and dampers allow for good ride quality at the expense of increased vertical wheel and body motion. Meanwhile to provide good road holding and handling, large spring and damper coefficients are required [1]. Hence, a trade-off must be made by choosing coefficients large enough to reduce wheel and vehicle motion, yet small enough to provide an acceptable ride quality [1, 2]. One solution to this trade-off problem is the addition of a controllable element (variable damper) to the passive suspension system; this setup is referred to as semi-active suspension. In order to achieve the required response from the semi-active suspension system, a suitable control strategy must be implemented to provide the necessary damping coefficient for the variable damper in the semi-active suspension. The focus of this paper is to provide a suitable control method to generate the required damping coefficient. Fuzzy logic control (FLC) has been extensively studied and implemented in various systems such as: cruise control, engine speed control, air conditioning units, torque generation for electrical vehicles, and most recently self-parking vehicle 978-1-4244-2693-5/09/$25.00 2009 IEEE 2118 system [3-5]. A main advantage for this type of control strategy is that an accurate description of the system is not necessary, also since the system relies on various sensors, imprecisions inherent in the measurements obtained by the various sensors are unavoidable, fuzzy logic control is ideal for this type of situation, since accurate and precise input is not required. Due to these advantages, many researchers choose to look into this type of control strategy to eliminate the trade-off between the ride quality and vehicle handling. In [6], the authors proposed a fuzzy controller tuned to enhance mainly the ride comfort of the vehicle. Through various simulations they were able to show that the FLC is able to provide a superior ride quality in comparison to other common control strategies such as skyhook control [10, 13]. In [8] a FLC that is capable of providing a reasonable performance with respect to both ride quality and vehicle handling was developed. The authors achieved this by using a total of 49 rules in their FLC. This paper investigates the performance enhancement a fuzzy logic controller is able to provide. The simulations will be performed on a quarter car suspension model. In order to fully understand the performance enhancement the FLC is able to provide, two additional suspension systems are modelled and simulated using the same test conditions, these are the standard passive suspension system [2, 9] and a LQR controlled semi-active suspension system [1, 2, 9]. The remainder of this paper is structured as follows. Section II outlines the automotive suspension model used to test the performance of both passive and semi-active suspension systems, and the implementation of the standard LQR control method. The proposed fuzzy control method is given in Section III. Section IV presents the simulation results. Finally, the conclusion and future work are provided in Section V. II. AUTOMOTIVE SUSPENSION MODEL AND LINEAR QUADRATIC REGULATOR A. Quarter Car Model A quarter car suspension model with two degrees of freedom is shown in Fig.I. The model represents the vehicle suspension system as it relates to one of the four wheels. This model assumes that the vehicle tire does not leave the ground, the vertical displacement of both the vehicle body and the tire are measured from their respective equilibrium positions, and the vehicle body (sprung mass) is a rigid body. ----.J z, ----.J z. ----.J z, Fig. 1 Quarter car passive suspension system. Fig. 2 Quarter car semi-active suspension system. k, = 16000 N/m, b s =1000 Ns/m, k, = 160000 N/m, m, = 240 kg, m ll =36 kg. The model described above must be extended to properly implement the functionality of the semi-active suspension [1, 9, 13]. This is accomplished by adding a variable damper in parallel with the passive elements (spring and damper). By connecting the variable damper in this manner the performance of the system is enhanced, by providing an additional support to the passive elements, and allowing for increased stability of the system. Fig. 2 shows a schematic diagram of the system. State Output o 11m, o -1/ m u Fig. 3 Block diagram ofthe semi-active suspension. L-----1 A z, .: 000 o 1/ m u 0 Ro ad Disturbance Input B. Linear Quadratic Regulator Linear quadratic regulator theory has been studied, and used by many researchers in an attempt to design an optimal controller for active and semi-active suspension systems [2, 9, 12]. This type of controller works towards generating an optimal actuator force that leads to the minimization of a performance index. This performance index (5) is related to the ride quality, suspension deflection (Xl) and tire deflection (X3)' J I 1 E[T f " 2 2 .2 2 .2] (5) = ;'!!,T c, 0 Z, +PIXI +P2 z, +P,x, +P4 z u By introducing the variable damper to the passive system equations (I) to (3) are modified to account for the additional element. Equation (4) shows the modified state space representation of this system. X=AX+NXb"",; + Li, (4) where b Semi is the damping coefficient of the variable damper (controlled parameter) in the design, and The block diagram representation of the state space equations used in the implementation of the quarter car semi-active suspension system is shown in Fig.3. By comparing (3) and (4), and inspecting Fig. 3, It IS evident that the passive suspension model is equivalent to the semi-active suspension model, with the assumption that the damping coefficient of the variable damper is zero for the passive suspension. This fact allows for a simple simulation of both the passive and semi-active systems. (3) ,L =[ , -I o -I bLm, 1 -b, I m. o o o In the quarter car model shown in Fig. I, k, and b s represent the spring constant and the damping coefficient, respectively; m, is the sprung mass that represents the quarter car equivalent vehicle body mass; mil represents the unsprung mass that is the equivalent to the mass of the vehicle axel and tire; and ZII, z, and z; represent the vertical displacements from static equilibrium of the sprung mass, unsprung mass, and the road, respectively. The linearized equations of motion are written as [9] m,=, +b,(i, -iu)+k,(z, -z.,)=O, (I) mu=u- b,(i, - i u) - k,(z, - zu)+ k,(z, - z,) =O. (2) In order to implement this model, a state space representation is constructed as [9] X=AX +Li" where A=[-/1m, -b,11m, o 0 k, I m u b, I m u - k, I m u [ XI [Z' . : X 2 Z X= = s , X 3 Zu -z,. X 4 i u where Xl is the suspension deflection; X2 is the absolute velocity of the sprung mass; X3 is tire deflection; and X4 is the absolute velocity of the unsprung mass. For this system the coefficients are described as [9] 2119 PO 0,4 0.3 LG (c) (a) ZR (b) ME Sl\I 500.0 1000.0 1500.0 2000.0 2500.0 3000.0 3500.0 4000.0 Dampi ng Coefficient 0,2- -0.3 -0',2 0.'0 0:1 0.'2 Suspe nsi on Defl ecti on (em) ZR 1.0- 0.8 - 0.6 - /I 0.4 - 0.2 - 0.0 0.0 ,. Fig. 6: Membership Functions for the proposed Fe. (a) Suspension deflection Input MFs ; (b) Body velocity Input MFs; (c) Damping coefficient output MFs. ,t III. THE PROPOSED NEW Fuzzy CONTROLLER FLC is one of the most efficient and popular control methods that currently exist for controlling the variable damping system in semi-active suspensions. A standard FLC is composed of a fuzzification interface, fuzzy rule base, decision making logic, and defizzification interface [14, 15]. The Fuzzification interface is the process which transforms measurements into fuzzy sets which are then used in the fuzzy rule base, and decision making process. The Defuzzification interface is the process which converts the fuzzy output into crisp values to be used by the plant. The input and output functions used in the controller depend on the working dynamics of the plant to be controlled, and are subject to the designers knowledge and method of design. In the proposed FLC two inputs are implemented, they are the suspension deflection (SD, Fig. 6.a), and the sprung mass body velocity (BY, Fig. 6.b). Each of these inputs is represented by using three triangular membership functions (MFs) with the following linguistic variables: Negative (NE), Zero (ZR), and Positive (PO). The output membership function (Damping Coefficient, Fig. 6.c) is designed as a singleton fuzzy set composed of five membership functions with the linguistic variables : Zero (ZR), Small (SM), Medium (ME), Large (LG), and Extra Large (XL). NE ZR PO
0 6- 0,4- (b) --.J z, --.J z. (a) Fig. 4 Equivalence offorce control and damper control systems . (a) Force control ; (b) damper control. Fig. 5 LQR semi-active control laws. In Fig. 5, the active force is given as F;, = -K1(B T P+S)x, (6) where R= _ _I_ 0 2 ' 2 2 2 , m s m s m s m s 8=[0 11m, 0 l /mJ , P=[PI p, P 3 pJ F b =- - -'- stmJ .\ 0: - X .J The performance matrix (P) above contains the weights used in the LQR control performance; these weights are set depending on which aspects of the suspension system are desired to be enhanced. In this paper the values are picked to reduce the sprung mass acceleration without significantly penalizing the other performance aspects. This is done by using very small coefficients in the performance matrix [9]: P = [0.4 0.16 0.4 0.16] Using the determined active force, the damping coefficient of the semi-active damper can be calculated using the conditions in Fig. 5. Although this control method provides a reasonable performance enhancement to all performance criteria, it requires specific and very accurate road and vehicle dynamics measurements, which in tum require expensive sensors. Depending on the actuator force and the difference between the absolute velocity of the sprung mass and that of the unsprung mass, the semi-active damper (bsemD will take on a specific value using the equations in Fig. 5 [2]. 2120 The control method is implemented in the fuzzy rule base using fuzzy conditional statements. Table I shows the 9 fuzzy rules implemented in the controller. These rules are interpreted by the controller as follows: IF (SD is NE) and (BV is NE), THEN (DC is XL). IF (SD is NE) and (BV is ZR) , THEN (DC is LG). IF (SD is PO) and (BV is PO), THEN (DC is XL) . TABLE ! FLC RULES FOR CALCULATING THE DAMPING COEFFICIENT -40 -'------------------------- Time Interval .... Passive Response - Semi-Active - LOR Response - Semi-Active - FLC Response Fig. 7 Results for ride quality simulati on. t, l' .r., .r., .r., . " . . . . . . . . . 1:.A1 lY :Ai :If; :If; :If; l U , :n ,1 :n ,1 f#
-- -- -- -- -- -- -- - 'l- 0 l I '"
I b I'"
, 'l . 'H . 'l . '.'t ','t ## '.'t . . , . . . . . V V V V -0.05 -0.04 -0.03 0.04 0.05 0.03 :[ 0.02 .20.01 0 ;: Q) C -0.01 e i= -0.02 Rule # Suspension Body Damping Deflection Velocity Coeffi cient (SD) (BV) (DC) I NE NE XL 2 NE ZR LG 3 NE PO ME 4 ZR NE SM 5 ZR ZR ZR 6 ZR PO SM 7 PO NE ME 8 PO ZR LG 9 PO PO XL IV. SIMULATION R ESULTS By implementing the memberships of Fig. 6, along with the rules in Table I, a suitable damping coefficient can be determined to stabilize the system efficiently. -0.5 -'----- - - - - - - - - - - - - - - - - - - - - 0.5 ,--- - - - - - - - - - - - - - - - - - - - - Time Intreval 50 ..0.. - - -:0..-- - - 0.4 t--:..._- - - ..-- - - .. EO.3 -0.4 +----' "-- ---' '------' "-- ---' "-- ---' .. .. Passive Response -r-e-r-Semt-Acnve - LOR Response - Semi-Active - FLC Response .. .. Passive Response Semi-Active - LOR Response - Semi-Active - FLC Response Fig. 8 Results for road holdin g simulation. Time Intreval Fig. 9 Results for support of static weight simulat ion. By analysing the magnitude of the various responses obtained from the simulation, it is clear that fuzzy logic control is able to provide a superior performance enhancement over the LQR controller, and the passive suspension system. Table II contains a summarization of the observed magnitude response by each system. In this section, the quarter car model is used to test the performance of the proposed fuzzy controller. The performance of the proposed controller is compared to that of the LQR control method for semi-active suspensions, as well as to a standard passive suspension system. In order to obtain a clear indication of the performance enhancement FLC is able to provide for semi-active suspensions, three functions of the modeled suspension system are simulated [2]: 1. Ride quality (Fig. 7) In general the ride quality is optimized by minimizing the vertical acceleration of the passenger's location. 2. Road holding and handling (Fig 8) This function is characterized by the vehicles cornering, breaking, and traction abilities. The performance of the suspension in regards to these characteristics can be improved if the variations in the normal tire loads are minimized. 3. Support of static weight (Fig 9) This function is enhanced by minimizing the rattle space requirement of the vehicle (suspension deflection). In all three simulations, the road disturbance is modelled as a sine wave with a frequency of 1Hz and amplitude of 0.01m. The inputs to the FLC are assumed to be readily available, either through direct measurement; using proximity and velocity sensors, or through a standard observer system. 2121 TABLE II PERFORMANCE DIFFERENCES BETWEEN THE PASSIVE AND SEMI-ACTIVE SUSPENSIONS Type of Ride Road Holding Support of Static Quality Suspension (m/s 2 ) (m) Weight (mm) Passive 54 0.088 7.7 Suspension LQR Controlled 44 0.076 5.5 SASS FL Controlled 31 0.052 2.5 SASS Using the data in Table II, it is evident that the proposed fuzzy logic controller offers a significant improvement in performance with comparison to the optimal (LQR) control method. The simulation results show that the three suspension functions are enhanced with the fuzzy controlled semi-active suspension system in comparison to the optimal controlled semi-active suspension system by 28%, 31% and 54% respectively. From these results it's evident that the proposed fuzzy logic controller is outperforming the LQR controller by at least 28%, which is a significant improvement. V. CONCLUSION In this paper, a fuzzy logic control strategy for semi-active suspension systems is developed. The fuzzy logic controller is used to generate the damping coefficient required by the variable damper in semi-active suspension systems. An optimal controller based on the linear quadratic regulator for semi-active suspensions, and a passive suspension system, are adopted for comparison with the proposed fuzzy logic controller. Simulation results show that the proposed fuzzy logic controller performs significantly better than the linear quadratic regulator control method for the same semi-active suspension system, as well as a standard passive suspension system. Analysis of the results show that the trade-off problem suspension systems suffer from is significantly reduced by using the proposed new fuzzy logic controller. Furthermore, the performance enhancement that the proposed fuzzy logic controller provides, is not only limited to a single suspension function as in the controller proposed by the authors in [6], but is also tuned to enhance the performance of all three basic functions of the suspension system. Unlike the work presented in [8] where the authors used a total of 49 rules to achieve a 40% improvement over the passive suspension, the proposed fuzzy controller is able to achieve a 42% improvement over the passive system using only 9 rules. This is a significant improvement over the design in [8], since the implementation of the 49 rules in hardware will add to the complexity, cost, as well as the size of the design. 2122 To further demonstrate the effectiveness of the proposed controller, future work will involve the implementation of the fuzzy controller using the Spartan 3E FPGA board, while simulating the quarter car model using the LabVIEW design and simulation environment. Further research would also be performed to look into the performance of the controller, when using an observer system to predict the state of the vehicle suspension system. This would reduce the dependency of the control strategy on the various expensive sensors that are needed to determine the state of the suspension, which is needed by the control system. REFERENCES [1] A.1. Barr and 1.L Ray, "Control of an active suspension using fuzzy logic," in Proc. of5th IEEE Int!. Conference on Control and Automation, New Orleans, USA, 1996, pp. 42-48. [2] R. Rajamani, "Vehicle Dynamics and Control, " Springer, Berlin, 2006. [3] R. Riadi, R. Tawegoum, A. Rachid, and G. Chasseriaux, "Decentralized temperature fuzzy logic control of a passive air conditioning unit," in Proc. of 15th IEEE IntI. Mediterranean Conference on Control and Automation, Athens, Greece, 2007, pp. 1-6. [4] Q. Zeng and J. Huang, "The design and simulation of fuzzy logic controller for parallel hybrid electric vehicles," in Prec. ofthe IEEE IntI. Conference on Automation and Logistics, Shandong, China, 2007, pp. 908-912. [5] R. Muller and G. Nocker, "Intellignet cruise control with fuzzy logic," in Proc. of IEEE IntI. Intelligent Vehicles '92 Symposium, Detroit, USA, 1992, pp.173-178. [6] C.F. Nicolas, 1. Landaluze, E. Castrillo, M. Gaston, and R. Reyero, "Application of fuzzy logic control to the design of semi-active suspension systems," in Proc. of 6th IEEE IntI. Conference on Fuzzy Systems, Barcelona, Spain, 1997, pp. 987-993. [7] L. Zheng, Y. N. Li, J. Shao, and X. S. Sun, "The design of a fuzzy- sliding mode controller of semi-active suspension systems with MR dampers," in Proc. of 4th IEEE IntI. Conference on Fuzzy Systems and Knowledge Discovery, Hainan, China, 2007, pp. 514-518. [8] Z. Li, Y. Yang, X. Gong, Y. Lin, and G. Liu, "Fuzzy control of the semi- active suspension with MR damper based on uGA," in Proc. of IEEE Int!. Vehicle Power and Propulsion Conference, Harbin, China, 2008, pp. 1-6. [9] T. Butsuen, "The Design of semi-active suspensions for automotive vehicles," Ph.D. diss., M.LT., 1989. [10] M. Bigarbegian, W. Melek, and F. Golnaraghi, "A novel neuro-fuzzy controller to enhance the performance of vehicle semi-active suspension systems," Vehicle System Dynamics, vol. 46, no.8, pp. 691-711,2008. [11] L.C. Felix-Herran, J.1. Rodriguez-Ortiz, R. Soto, and R. Ramirez- Mendoza, "Modeling and control for a semi-active suspension with a Magnetorheological damper including the actuator dynamics," in Proc. of IEEE IntI. Electronics, Robotics and Automotive Mechanics Conference, Morelos, Mexico, 2008, pp 338-343. [12] L.R. Ray, "Robust linear-optimal control laws for active suspension systems," ASME Journal of Dynamic Systems, Measurement, and Control, vol. 114, no. 4, pp. 592-598, 1992. [13] M. Ahmadian, and C.A. Pare, "A quarter-car experimental analysis of alternative semi-active control methods," Journal of Intellignet Material Systems and Structures, vol. 11, no. 8, pp. 604- 612, 2000. [14] K. Tanaka, H.O. Wang, "Fuzzy Control Systems Design and Analysis: A Linear matrix Inequality Approach," John Wiley & Sons, New York, 2001. [15] M. E. Dupre, "GA Optimized Fuzzy Control of an Autonomous Mobile Robot," MSc Thesis, University of Guelph, 2007.