Multi-Output Physically Analyzed Neural Network for the Prediction of Tire–Road Interaction Forces

Abstract

This article introduces an innovative method for predicting tire–road interaction forces by exclusively utilizing longitudinal and lateral acceleration measurements. Given that sensors directly measuring these forces are either expensive or challenging to implement in a vehicle, this approach fills a crucial gap by leveraging readily available sensor data. Through the application of a multi-output neural network architecture, the study focuses on simultaneously predicting the longitudinal, lateral, and vertical interaction forces exerted by the rear wheels, specifically those involved in traction. Experimental validation demonstrates the efficacy of the methodology in accurately forecasting tire–road interaction forces. Additionally, a thorough analysis of the input–output relationships elucidates the intricate dynamics characterizing tire–road interactions. This research underscores the potential of neural network models to enhance predictive capabilities in vehicle dynamics, offering insights that are valuable for various applications in automotive engineering and control systems.