Fault Detection and Diagnosis Based on Interactive Multi-Model Moving Horizon Estimation and Neuro-Tire Model

Accurate fault detection information is essential for active fault-tolerant control to ensure the safety of vehicles when experiencing actuator functional failures. This article addresses a fault detection and diagnosis (FDD) technique for four-wheel independent drive electric vehicles. Several factors influence the accuracy of in-wheel motor FDD. First, the parameters of the classic empirical tire models used for FDD are challenging to identify and vary with tire wear, pressure, and other usage conditions. A neural network longitudinal tire model is trained utilizing vehicle state samples captured during healthy use, freeing the tire model from parameters that are difficult to identify and change dynamically. Additionally, an interactive multi-model moving horizon estimation (IMM-MHE) method is proposed, effectively addressing the issue of model mismatch while considering constraints imposed by the wheel system state. The moving horizon framework also enhances the estimation performance by incorporating historical observation data. The developed FDD framework is evaluated by aggressive single-lane-change maneuvers and acceleration conditions that elicit diverse tire dynamic behaviors. Simulation and hardware-in-loop tests demonstrate that the IMM-MHE combined with the neuro-tire model significantly improves the accuracy of FDD even under intense usage conditions.