Real-Time State Perception and Adaptive Fault-Tolerant Control of Electric Drive Systems for New Energy Vehicles Based on Multi-Source Sensor Fusion

Abstract

Electric drive systems in new energy vehicles (NEVs) depend on multi-source sensors for real-time state perception, yet harsh automotive environments induce sensor faults that degrade fault-tolerant control (FTC) performance. Existing architectures treat sensor fusion, fault diagnosis, and controller reconfiguration as separate sequential processes, allowing corrupted measurements to enter the fusion stage before reliability is assessed. This paper proposes a confidence-driven multi-source sensor fusion (CDMSF) framework comprising three sequential modules: sensor confidence evaluation, adaptive weighted fusion, and adaptive fault-tolerant control. The framework evaluates sensor reliability through prediction residuals before and during fusion, dynamically weighting each sensor's contribution to prevent corrupted data from reaching the controller. Hardware-in-the-loop validation demonstrates 62.3% reduction in fault detection latency (8.4 ms versus 22.3 ms), 47.6% improvement in torque tracking accuracy under simultaneous multi-sensor faults, and guaranteed stability margins exceeding 6 dB. The CDMSF framework closes the critical convergence gap between perception and control in safety-critical NEV applications.