Intelligent System Identification for an Axis of Car Passive Suspension System Using Real Data Dirman Hanafi l Mohd. Fua'ad bin Rahmar Zainal Abidin bin Ahmad 3 Amran bin Mohd Zaid 4 1,3,4 Department ofMechatronic and Robotic Engineeering, Faculty ofElectrical and Electronic Engineering, Universiti Tun Hussein Onn Malaysia, 86400 Parit Raja, Batu Pahat, Johor, Malaysia Tel: +60-7453-7631, Fax: +60-7453-6060 E-mail: dirman@uthm.edu.my 2 Instrumentation and Control Engineering Department, Faculty of Electrical Engineering Universiti Teknologi Malaysia, 81310 Skudai, Johor Darul Ta'zim Phone: 07-5535900, Fax: 07-5566272, E-mail: drfuaad@lke.utm.my Abstract This paper presents an intelligent system identification using multilayer perceptron neural network algorithm for an axis car passive suspension model. Nonlinear AutoRegressive with eXogenous input (NARX) model were assumed for the system in order to determine the multilayer perceptron neural network structure. The intelligent system identifcation contructed for NARX model used real input output data acquired by driving a car on a special road event. The results show that the method proposed is suitable for modeling a quarter car passive suspension systems. Keywords: Intelligent system identification, Multilayer perceptron, NARX model, real input output data. Introduction The great interests in the vehicle research field is improving the car ride quality and handling performance. The performance improvement can be done by design and analyze the car suspension system controller [1,2,3], thus the high fidelity mathematical model capturing realistic dynamic of the car suspension system is necessary [4]. As in [4,5,6,7], the best way to determine the high fidelity mathematical model of system is using system identification. Since the successful applications of neural networks contribute to widely use in field of system identification as intelligent system identification [8,9]. This paper presents an axis car passive suspension system as a quarter car and its model identifies using intelligent system identification [5,6]. Westwick and George applied traditional feed forward neural network for identifying the nonlinearity contribution in the nonlinear part of a quarter car model [5]. The intelligent system identification used only for identified part of a quarter car model not for whole model. The nonlinear part of the quarter car model is identified based on simulation data. Buckner and Schuetze used the structured artificial neural networks to be the intelligent parameter estimation to continually adapt the lumped 978-1-4244-3481-7/09/$25.00 ©2009 IEEE parameters of a linear quarter-car suspension model [6]. The parameter estimation process is done using simulation data. While, this paper we focus on the intelligent system identification to identify whole model of a quarter car passive suspension system where the intelligent system identification algorithm were based on iterative weighted least square neural network. Next the networks structure develops based on an axis car suspension model. A quarter car passive suspension is assumed has nonlinear model of Nonlinear AutoRegressive with eXogenous input (NARX) structure. This relates with the dynamics of real suspensions that inherently nonlinearity. Moreover this model structure is simple and widely used in control system. The NARX model of a quarter car passive suspension system is identified using real input output data. The data are collected by running car on special road event where the road event is made using bump wood and construct as bumpy road surface. A Quarter Car Modeling The main functions of the ground vehicle suspension is to support the car weight, keep the wheel on the ground, minimize transient force to the body, maintain good ride comfort and enhance handling performance [10,11]. The suspension systems are influenced by the excitations due to the road unevenness and variable velocity. In earlier work on the analysis of car response, the car velocity was considered as constant, since it was difficult to introduce the effect of variable velocity formulation of the model [11]. The physical constraint of the suspension system is the limited suspension travel. A basic passive suspension consists of a spring with a parallel damper at each wheel. In this paper, a generic of quarter car passive suspension model has been modified for analysis [3,4,5,6,7]. Related to the current trend using four independent suspensions on a single car, a quarter car systems offer a quite reasonable Authorized licensed use limited to: KOLEJ UNIVERSITI TEKNOLOGI TUN HUSSEIN ONN. Downloaded on December 21, 2009 at 20:31 from IEEE Xplore. Restrictions apply.