Digital-Twin-Driven Predictive Maintenance and Fault-Tolerant Control for Electrified Agricultural Machinery: A Multiphysics and Deep Reinforcement Learning Framework for PMSM In-Wheel Drives

Electrified agricultural machinery increasingly relies on permanent-magnet synchronous motor drives that must remain dependable under variable traction, terrain-induced vibration, thermal cycling, contamination, and intermittent duty. Existing work offers mature methods for motor fault diagnosis, digital twins, predictive maintenance, and distributed-drive fault-tolerant control, but their health information is often not shared across diagnosis, maintenance, and control. This article develops a multiphysics digital-twin framework in which electrical, thermal, mechanical, and vibrational evidence maintains a probabilistic health belief that constrains both supervisory maintenance decisions and torque redistribution. The framework formalizes observability, uncertainty-aware health factors, deterministic safety shielding, multi-rate fallback logic, and agriculture-specific traction constraints. Reproducible low-fidelity numerical studies provide preliminary support for context-conditioned residual diagnosis and the revised thermal-continuity trade-off underlying health-aware derating, including robustness and sensitivity analyses. They do not validate sim-to-real transfer or the reinforcement-learning policy itself; those hypotheses remain open for dynamometer and hardware-in-the-loop testing. The results are therefore interpreted as mechanism-level evidence rather than field-performance validation.