A Neural Network Approach for Roughness-Dependent Update of Tyre Friction

Abstract

The paper presents a unique application of artificial neural networks (ANN) for replicating the predictions of Persson's flash-temperature viscoelastic friction model. This allows the efficient calculation of the coefficient of friction between a tyre and the road surface, given the operating conditions, the tread material properties, and the power spectral density of road roughness. This is important because it allows the update of friction in a computationally efficient manner and therefore presents an opportunity for carrying out vehicle simulations on several road surfaces, without the requirement for tyre testing on every surface. To this end, a method is also proposed for integrating the most successful ANN configuration with a tyre model that had its baseline parameters identified by flat-track testing on a sandpaper surface. It is shown that the ANN-enhanced tyre model operates several times faster than real-time, predicting reduced peak and asymptotic tyre forces, as expected in most cases when moving from sandpaper to regular asphalt. The enhanced tyre model is further integrated with a vehicle model to illustrate the significant effect of reduced friction in stopping distance and handling dynamics.