A strain-based estimation of tire-road forces through a supervised learning approach

Abstract

In the automotive industry, instrumented tires, called smart tires, are widely employed for identifying tire-road contact forces functional for monitoring the operative conditions of vehicles. Recently, the coupling between smart tire technology with artificial intelligence techniques is made due to the possibility of exploiting data obtained from intelligent tires for training Neural Networks able to identify variables functional for vehicle monitoring and control purposes. A supervised learning approach is presented in this paper for estimating in-plane tire-road interactions starting from measurements of the circumferential strain of the tire tread band. Two feedforward Neural Networks are trained by employing synthetic strain data generated through the well-known Flexible Ring Tire Model, avoiding expensive experimental tests for collecting strain data. The results obtained through the trained Neural Networks regarding vertical and longitudinal tire-road forces are compared with the simulated ones related to the Flexible Ring Tire Model, considering three different profiles of the circumferential strains. The proposed approach demonstrates its suitability as a monitoring tool in the automotive field by estimating tire-road forces in real-time at each wheel revolution.