Fault Detection and Diagnosis for Thermal Management System in Electrified Aircraft Propulsion

Abstract

The thermal management system (TMS) of electrified aircraft propulsion (EAP) is critical to safety and efficiency, especially for hydrogen fuel-cell and battery configurations. While architecture optimisation has been widely studied, fault detection and diagnosis (FDD) of TMS components remains underexplored. Here, a model-based FDD framework that leverages analytical redundancy via a bank of unknown-input observers (UIOs) is proposed. Each UIO is designed on a linearised TMS model of a Polymer Electrolyte Membrane Fuel Cell to reject known disturbances and non–target actuator faults; residuals are formed as innovations and shaped by a left-nullspace projection to enforce selectivity. Detection uses a causal moving-average envelope and adaptive, robust thresholds based on median absolute deviation with persistence, improving detectability during take-off while curbing false alarms. For a realistic scenario, measurements are simulated with Gaussian sensor noise. Simulation-based single-fault injections show consistently high detection rates, short delays relative to system dynamics, negligible false alarms under influence of disturbance, and no cross-detections across observers, evidencing reliable isolation. A sensitivity limit is observed for a coolant-leak type fault, where detection rate reduces for fault magnitude below 50% of the coolant flow.