Adaptive Fuzzy Wavelet Control Strategy for Full Car Suspension System

Laiq Khan, Rabiah Badar, Shahid Qamar May 31, 2011

Abstract The direct focus of this work is to explore the potentials of modern and intelligent control techniques for the purpose of designing novel adaptive control systems in the area of vehicle suspension control. The detailed mathematical modeling of dierent adaptive softcomputing techniques has been developed and successfully applied to a full car model. The robustness of the presented techniques has been proved on the basis of dierent performance indices. Unlike, the conventional PID, the proposed algorithms have been compared with each other and Adaptive PID (APID) controller. The simulation results and their analysis reveal that proposed Adaptive Fuzzy Wavelet Neural Network Controller (AFWNNC) gives better ride comfort and vehicle handling as compared to passive or semi active and APID control. The performance of the active suspension has been optimized in terms of seat, heave, pitch and roll displacement and acceleration.

Introduction

In the last few years, dierent linear and non-linear control techniques have been applied by many researchers on the vehicle suspension system. The basic purpose of suspension system is to improve the ride comfort and better road handling capability. All the ongoing research in this area mainly caters the following issues to improve the suspension control; minimize the eect of road and inertial disturbances, on human body, caused by cornering or braking. minimize the vertical car body displacement and acceleration. good control on all the four wheels of the car for their optimal contact with road. All the above objectives lead to rapidly changing operating conditions and the passive suspension system is not as ecient to cope with them simultaneously by adapting its parameters. So, there would always be a compromise between comfort and safety for passive suspension system. Therefore, a good control for active car suspension is unavoidable to get better ride comfort and vehicle control. Many approaches on active quarter car suspensions system have been carried out by [1, 2], but does not reveal robustness of the system. Various control techniques such as optimal state-feedback, model reference adaptive control, backstepping method, fuzzy control, and sliding mode control have been presented for optimized control of the active suspension system. The active control of seat for full car model is examined by [3]. The PID controller is applied on active suspension system by [4]. A variety of simulations showed that the fuzzy logic control is procient to give a better ride quality than other common control approaches for example, skyhook control [5]. However, it is still complicated to design suitable membership functions and fuzzy linguistic rules of the fuzzy logic controllers to give suitable learning rate and weighting distribution parameters in the self-organizing fuzzy controller. The fuzzy control for active suspension system is presented by [6], but only considered the ride comfort. Then adaptive fuzzy controller for semia-ctive suspension systems was presented by [7] which shows only the acceleration of the vehicle compared to the passive suspension. [8] proposed a tuneable fuzzy logic controller, on active suspension system without taking into account the nonlinear features of the suspension spring and shock absorber, also, the robustness problem was not discussed. The neural 1

network control system applied on active suspension system has been discussed by [9] but does not give enough information about the robustness and sensitivity properties of the neural control towards the parameter deviations and model uncertainties. Also, sliding mode neural network inference fuzzy logic control for active suspension systems is presented by [10], but did not give any information about the rattle space limits. The aforementioned fuzzy logic and neural network controllers on active models, do not give enough information about the robustness, sensitivity and rattle space limits. In this study, fuzzy wavelet neural network control has been proposed for the active suspension control of full car model. The proposed AFWNNC integrates the ability of wavelet to analyze the local details of the signal with that of fuzzy logic to reduce system complexity and with the self learning capability of neural networks, which makes the controller ecient for controlling unknown dynamic plants. Takagi-Sugeno-Kang (TSK) fuzzy model has been modied with wavelets in the consequent part. Dierent combinations of membership functions in antecedent part and wavelets in the consequent part have been used which increases the computational control. The following form of fuzzy rules has been used; If x1 isAn1 and x2 is An2 and ... xm is Anm
m

(1)

T hen yn =
i=1

win in

Where x1 , x2 , ..., xm are the input variables and yn is the corresponding output variable, Aij is the membership function of ith rule and jth input, whereas in is the wavelet basis function. The update parameters include mean and variance of the membership functions and the translation, dilation and weights of wavelets. The proposed controller has a layered architecture as follows; Layer 1: The nodes of this layer accept input values to transmit to the next layer. Layer 2: The neurons of this layer fuzzify the input to be used in the antecedents part of the linguistic fuzzy rules. Layer 3: In this layer each node represents a fuzzy rule. In order to compute the ring strength of each rule min operation has been used to estimate the output value of the layer. Layer 4: The output of this layer is the output of wavelet functions. Layer 5: This layer estimates the weighted consequent value of a given rule, i.e., the wavelet functions are multiplied with the third layers output value. Layers 6,7: Defuzzication process is carried out in these layers to calculate the output of the entire network using centroid defuzzication. Backpropagation learning algorithm and gradient decent rule has been used to update the parameters of the controller by minimizing the following cost function. 1 J= 2
n d u (yi yi )2 i=1

(2)

Where, y d and y u are the desired and actual outputs of the network, respectively, and n is the total number of outputs which is one in our case.

Simulation Results

Four dierent types of fuzzy wavelet neural network control techniques have been applied to full car suspension model. It has been found that active control oers good vehicle control and better passenger comfort as compared to the passive or semi-active control. AFWNNC-1 and AFWNNC-2 use maxican hat as wavelet in the consequent part and gaussian and triangular as membership function in the antecedent, respectively. AFWNNC-3 and AFWNNC-4 use morlet as wavelet in the consequent part and gaussian and triangular as membership function in the antecedent, respectively. Figure 1 shows the morlet and maxican hat wavelets alongwith their mathematical expressions. Three types of road proles have been examined to check the robustness of the applied algorithm. These road proles have been used in context of roll, pitch, heave and seat displacement and acceleration. Road prole 1 involves one pothole and one bump, each having duration of one second with a 2

Figure 1: Wavelet functions time delay of 8 secs., for front and rear tyres. Road prole 2 has been taken as two potholes of dierent amplitudes whereas road prole 3 is white noise. MATLAB/Simulink environment has been used to simulate an eight-degree-of-freedom full car model. Figure 2 to Figure 4 give the time domain simulation results for the three road proles for heave displacement. It can be seen that AFWNNC-2 has better response than AFWNNC-1 and APID and AFWNNC-4 has better response than AFWNNC-3 and AFWNNC-2 but at the end, its steady state response degrades whereas AFWNNC-3 shows good steady state performance.

Figure 2: Heave displacement for road prole 1

Figure 3: Heave displacement for road prole 2

Figure 4: Heave displacement for road prole 3

The performance index used for evaluation of dierent algorithms is given by, J= 1 2
T T (ZP QZP ) 0

(3)

where, Zp is the vector for displacement or acceleration, Q is the identity matrix, and Root Mean Square (RMS) value for displacement and acceleration of heave has been calculated by, RM S = 1 T 3
T

[h(t)]2
t=0

(4)

Table 1 shows the comparative results for the applied algorithms against each road prole for heave displacement and acceleration. It can be seen that AFWNNC-4 has over all good performance on the basis of RMS value and performance index for all the three road proles. The table also shows that passive or semi-active control has larger performance degradation for more rough road prole as compared to AFWNNC-4. From all the presented results, it can be concluded that proposed AFWNNC provides ecient vehicle handling capability maintaining the passenger comfort by reducing the settling time and improved steady state response as compared to passive, semi-active and APID control. Table 1: Performance Comparison for Heave
Road Prole Control Algo. Passive Semi-active APID FWNN-1 FWNN-2 FWNN-3 FWNN-4 Passive Semi-active APID FWNN-1 FWNN-2 FWNN-3 FWNN-4 Passive Semi-active APID FWNN-1 FWNN-2 FWNN-3 FWNN-4 Performance Index 6.03515 5.3037 4.92835 4.6587 4.3854 1.9529 1.8006 4.2056 3.8635 3.1469 2.9745 3.0190 1.2864 1.1303 34.0666 33.4707 14.2367 12.5589 12.8224 8.0526 5.1264 RMS Disp. Acc. 0.0404 3.476 0.03729 3.2567 0.0166 3.1395 0.0158 3.0524 0.0148 2.965 0.00403 1.9763 0.004 1.8977 0.0339 2.9 0.09855 2.778 0.01542 2.5087 0.01397 2.439 0.0144 2.453 0.0036 1.604 0.0040 1.503 0.0682 8.254 0.06684 8.1815 0.0239 5.336 0.02607 5.0117 0.0262 5.064 0.0318 4.013 0.00956 3.202 % Improvement w.r.t. Passive Semi-active 70 65 75 65 77 62 93 93 94 89 57 36 80 72 60 47 83 78 87 82 64 55 65 64 64 53 79 73 87 83

Road Prole 1

Road Prole 2

Road Prole 3

References
[1] Hac, A.,Adaptive control of vehicle suspensions, Vehicle System Dynamics, Vol. 16, pp. 57-74, 1987. [2] Yue, C., Butsuen and Hedrick J. K., Alternative control laws for automotive active suspensions, Journal of Dynamic Systems, Measurement and Control, Vol. 111, pp. 286-291, 1989. [3] Rahmi, G., Active control of seat vibrations of a vehicle model using various suspension alternatives, Turkish J. Eng. Env. Sci., Vol. 27, pp. 361-373, 2003. [4] Kumar, M. S., Development of Active Suspension System for Automobiles using PID Controller, in Proc. the World Congress on Engineering, London, UK, Vol. 2, July 2 - 4, 2008. [5] Bigarbegian, M., Melek, W. and Golnaraghi, F., 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. [6] Yester, J. L. Jr., Fuzzy Logic Control of Vehicle Active Suspension, M.S. thesis, Electrical and Computer Engineering Department, Wayne State Univ., Detroit, MI, 1992. [7] Lieh, J. et al., Adaptive fuzzy control of vehicle semi-active suspensions, in Proc. ASME Dynamic Systems Control Division, Vol. 61, pp. 293-297, 1997. [8] Rao, M. V. C. et al., A tunable fuzzy logic controller for vehicle-active suspension system, Fuzzy Sets and Systems, Vol. 85, pp. 11-21, 1997. [9] Moran, A. et al., Optimal active control of nonlinear vehicle suspensions using neural networks, JSME International Journal, Vol. 37, pp. 707-718, no.4, 1994. [10] Al-Holou, N., Lahdhiri, T., Joo, D. S., Weaver, J., and Al-Abbas, F., Sliding Mode Neural Network Inference Fuzzy Logic Control for Active Suspension Systems, IEEE Transaction on Fuzzy System, Vol. 10, no.2, pp. 234-236, APRIL 2002.