Estimating Tire Forces with an Artificial Neural Network Model

We herein propose a method to estimate roll/pitch and car body slip angle using an unscented Kalman filter and to estimate tire longitudinal/lateral force using an artificial neural network (ANN). Existing methods cannot accurately estimate tire longitudinal/lateral forces owing to the nonlinearity and limitations of physical sensors. Hence, we aim to address this problem by training an ANN with various simulated driving data obtained from racing circuits experiencing excessive slips. The proposed method is learned and verified using CarSim, which is a vehicle dynamics simulation tool, and different racing circuits are used for learning and verification. This study may facilitate future relevant studies via the acquisition of actual vehicle data pertaining to excessive driving. Additionally, the feasibility of the proposed method for future actual-vehicle-based studies is verified and its limitations are analyzed.