Estimating Tire Forces Using MLP Neural Network and LM Algorithm: A Comparative Study

Accurately estimating tire forces is crucial for understanding and optimizing vehicle dynamics, particularly in terms of handling and stability. This paper presents a novel approach for estimating tire forces using a multi-layer perceptron (MLP) neural network and emphasizes the significance of carefully selecting the network architecture and training method for achieving optimal performance. The proposed method involves training the MLP neural network using an extensive and diverse dataset encompassing tire force measurements and associated input variables. This comprehensive dataset enables the network to capture and model the intricate and non-linear relationships that exist between input variables and tire forces. Through this process, the network becomes adept at providing accurate and real-time estimations of tire forces, which is essential for understanding and optimizing vehicle dynamics. Additionally, the paper explores the utilization of the Levenberg-Marquardt (LM) algorithm to optimize and fine-tune the weights and biases of the MLP network during the training process. Based on proposed method, a neural network is designed and trained to predict the lateral force of a tire based on the tire's slip angle and longitudinal force. The results of our research demonstrate the superior accuracy achieved by our proposed method in predicting tire forces across a range of scenarios and conditions and explore the effects of different factors which are involved in this study. This could contribute to advancements in traction control, stability control systems, and overall vehicle maneuverability.