A Direct Tire Capacity Estimation Method Using Deep Learning With On-Board Sensor Signals

Abstract

Accurate tire capacity information is crucial to maintaining vehicle handling stability and optimizing safety control systems. Tire capacity, which refers to the grip margin of the tire, represents the maximum additional force that a tire can exert under current conditions. It determines whether the vehicle is capable of executing additional steering, braking, and driving maneuvers while ensuring safety. This study proposes a direct tire capacity estimation method, which represents an advance in the field of tire capacity identification method. This method utilizes an end-to-end architecture to train the network, enabling direct utilization of signals from on-board sensors for estimating tire capacity, irrespective of whether the tires operate under pure or combined slip conditions. Additionally, it mitigates the issues of error transfer and accumulation often found in traditional layered estimation methods that overly rely on intermediate states. Outputs from the single-track (2-DOF) vehicle model are integrated as features to enhance the network's learning capability and interpretability. Additionally, a long-short-term memory (LSTM) based network is constructed to capture the nonlinear relationship and address the phase difference between features. The results of the validation and test sets show high identification accuracy, while the cosimulation of real driving scenarios is used for further verification. The results indicate that the proposed method accurately captures tire capacity information during whole tire working regions and exhibits robust generalization ability.