Nonlinear Tire Model Approximation Using Artificial Neural Networks

Abstract

This paper presents a comparative study between two different approximation approaches for the traditional Magic Formula tire model based on Artificial Neural Network (ANN), in specific, Radial Basis Function Network and Multilayer Perceptron Network, with the view to approximate longitudinal and lateral friction coefficients curves. Simulation results are considered satisfactory, showing that, the MLP network presented better results compared to the RBF network which indicates that the prediction of the friction coefficients curves is driven close to the reference derived from the Magic Formula. A nonlinear physical-mathematical model is used as the real plant for the comparison between the MLP tire model and the Magic Formula, considering torque and different steering angles as input. Moreover, the results show that the implemented network is adequate for applications in simulated ground vehicles.