A Neural Network Sliding Controller for Active Vehicle Suspension

The hydraulic active suspension systems have certain nonlinear and time-varying behaviors. It is difficult to establish an appropriate dynamic model for model-based controller design. Here a novel neural network based sliding mode control is proposed by combining the advantages of the adaptive, radial basis function neural network and sliding mode control strategies to release the model information requirement. It has on-line learning ability for handling the system time-varying and nonlinear uncertainty behaviors by adjusting the neural network weightings and/or radial basis function parameters. It is implemented on a quarter-car hydraulic active suspension system. The experimental results show that this intelligent control approach effectively suppresses the oscillation amplitude of sprung mass in response to road surface disturbances. Introduction A vehicle suspension system should have the capability to reduce sprung mass displacement and acceleration, and provide adequate suspension travel to maintain tyre-terrain contact. Hydraulic and pneumatic components are widely used in semi-active or active suspension systems of commercial vehicles. These dynamic systems exist certain dynamic uncertainty and nonlinear time varying behavior. It is difficult even impossible to derive or establish an accurate dynamic model for model based controller design. Although traditional adaptive control and sliding mode control schemes had been proposed to control dynamic systems with imperfect information. However, it still needs the system model information for the control law calculation. Hence, how to develop a totally model-free adaptive control structure has become an interesting research field. The model-free fuzzy logic control [1,2] and neural network control [3,4] were employed to design the controllers of vehicle active suspension systems for releasing the requirement of complicated dynamic model. Rao and Prahlad [5] proposed a tuneable fuzzy controller for an active suspension system. Huang and Lian [4] proposed a fuzzy and neural network hybrid control scheme to compensate the coupling dynamics for improving control performance. However, these approaches need a complicated learning mechanism or a specific performance decision table, which is designed by a trial-and-error process. Its application still exists certain difficulty. The RBF scheme was first proposed by Hardy [6]. It has been widely used to represent the nonlinear mappings between the inputs and outputs of nonlinear control systems. Sanner and Slotine [7] employed Gaussian basis functions in nonlinear adaptive control, Lu and Basar [8] used RBF to develop a neural-network based identification algorithm, and Chen et al. [9] employed RBFNN to model some unknown nonlinear functions for deriving a feedback linearization control law. Here a novel RBFNN based sliding mode controller is developed for direct control purpose and implemented on a quarter-car active suspension system. This control strategy is based on a radial basis functions structure and combined the advantages of adaptive 1 Materials Science Forum Vols. 440-441 (2003) pp 119-126 online at http://www.scientific.net © (2003) Trans Tech Publications, Switzerland Online available since 2003/Nov/15 All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of the publisher: Trans Tech Publications Ltd, Switzerland, www.ttp.net. (ID: 130.203.133.34-14/04/08,12:16:14) and sliding mode control schemes. The adaptive rule is employed to on-line adjust the weighting of radial basis functions by using the reaching condition of a specified sliding surface. Since this approach has learning ability for establishing and regulating the weightings of radial basis functions continuously, its control implementation can be started with zero initial weighting RBFNN. Active suspension system schema and model Here a serial type quarter-car active suspension system with 2 degree-of-freedom (DOF) is designed and built for investigating the dynamic performance and control effect. This experimental system includes a hydraulic actuating unit, a road condition simulation unit, an active suspension unit, an I/O data interface unit and a PC based control unit. In order to evaluate the dynamic response and control performance of this active suspension system, a 2 DOF dynamic model of this quarter car system is employed. The basic assumption is that the tyre always contacts with the road surface and the function of the tyre is simulated by a spring with spring constant and unsprung mass . The spring, damper and actuator between sprung mass and unsprung mass constitute an active suspension system as Fig. 1. Where and are measured variables representing the sprung mass displacement and the displacement of tyre axis, respectively. The dynamic equations of this suspension system can be derived. t k u m