Fourier Series Neural Network Analysis of a Steer-by-Wire System

Abstract

Automotive steering system research has traditionally focused on improving vehicle handling and safety, as well as investigating lateral dynamic issues. The emergence of hybrid vehicles provides a motivation for steer-by-wire technology in terms of power source availability and improved performance. From a design perspective, steering systems are difficult to accurately model due to the inherent nonlinearities present in the steering assembly, chassis, wheels, and tire/road interface. One modeling strategy that merits further attention is the Fourier Series Neural Network (FSNN) which has been proven effective for the characterization of dynamic systems. A neural network can approximate nonlinear functions to a high degree of accuracy, given an adequate network structure and sufficient training. In this paper, a Fourier Series activation function neural network will be studied to identify a steer-by-wire system. A behavioral model has been developed for the driver interface and directional control assembly of the rack and pinion steer-by-wire system. Representative numerical results are presented to demonstrate the FSNN’s ability to predict the system’s overall transfer function. This engineering tool may provide an attractive alternative to rigorous system modeling, and inherently captures the response characteristics due to the nonlinear behavior.