Data-Driven Predictive Diagnostics and Fault-Tolerant Control for Fuel Cell Electric Vehicle Powertrains Under Uncertainty

Fuel-cell electric vehicles (FCEVs) are vulnerable to sensor drift, actuator degradation, and air/water management faults that can rapidly impair efficiency, drivability, and durability. This work proposes a control-oriented, data-driven framework that tightly integrates predictive diagnostics with fault-tolerant control (FTC) for FCEV powertrains operating under uncertainty. A multi-sensor feature pipeline based on stack voltage/current, cathode pressure, compressor speed, hydrogen flow rate, DC-link voltage, and traction power is first constructed, and health indicators are generated using a probabilistic sequence model that captures temporal dependencies. The diagnostic module performs online early fault detection and isolation, including fault magnitude estimation and remaining useful time prediction to anticipate constraint violations. An active FTC layer then reconfigures the powertrain controller in real time based on fault estimates, enforcing fuel-cell ramp-rate limits, air-path constraints, and energy-storage limits while ensuring that fast power transients are buffered by the battery/supercapacitor and traction power tracking is maintained. Uncertainty is addressed through disturbance modeling and confidence-weighted adaptation to avoid overreacting to noisy diagnostic outputs. Simulation studies on urban and aggressive drive cycles with imposed sensor bias, compressor efficiency loss, and hydrogen starvation demonstrate superior robustness over a non-tolerant baseline, with earlier fault detection, no constraint or emissions violations, and reduced hydrogen consumption while preserving drivability.