Fault-Tolerant Control Architectures for Safety-Critical Unmanned Systems: A Comprehensive Review

Abstract

Safety-critical unmanned systems operating in proximity to humans or performing high-value missions require fault-tolerant control architectures that maintain safe operation despite component failures, sensor malfunctions, or actuator degradation. This comprehensive review examines fault-tolerant control (FTC) approaches spanning fault detection and diagnosis, control reconfiguration, and graceful degradation strategies across unmanned system domains. We systematically categorize FTC architectures into passive approaches employing robust control designs inherently tolerant to anticipated faults, and active approaches that detect faults and explicitly reconfigure control laws. The review analyzes fault detection and isolation methodologies including model-based approaches using observers and parity equations, signal-based methods applying statistical analysis, and data-driven techniques leveraging machine learning. Particular emphasis is placed on diagnostic accuracy metrics—false alarm rates and missed detection rates—and detection latency critical for timely intervention. Control reconfiguration strategies are comprehensively analyzed including control allocation redistributing effort among redundant actuators, adaptive control adjusting parameters to compensate for changed dynamics, and switching control selecting from pre-designed controllers. Hardware redundancy architectures are examined including homogeneous redundancy with voting schemes, heterogeneous redundancy avoiding common-mode failures, and analytical redundancy exploiting mathematical relationships. Application-specific implementations are analyzed for quadrotor UAVs with rotor failures, fixed-wing aircraft with control surface damage, and underwater vehicles with thruster faults. Verification and validation challenges are critically evaluated including combinatorial explosion of fault combinations and certification requirements. Advanced topics include prognosis-enhanced FTC predicting impending failures, learning-based FTC improving through experience, and cooperative FTC where healthy vehicles assist faulty ones. The review identifies research gaps in FTC for cyber-physical attacks and integration with mission planning.