Health Monitoring System-Informed Control Barrier Function Filter for Safety-Critical Systems

Aerial vehicles are increasingly deployed in safety-critical missions within uncertain environments, where disturbances, system faults, and external obstacles can compromise mission safety and success. Ensuring that commanded actions remain within the true, possibly time-varying capabilities of the vehicle is therefore a critical challenge. This paper proposes a general framework that integrates a machine learning-based Health Monitoring System (HMS) with a Control Barrier Function (CBF)–based safety filter that ensures continued safe operations. The HMS, originally developed for fault detection and diagnosis, is reconfigured to learn and define the vehicle’s safe set (flight envelope). The learned decision boundary is then interpreted as a CBF, establishing a state-dependent constraint. Based on this constraint, a control input is optimized such that the CBF condition is satisfied, thereby ensuring that the applied command remains compatible with the vehicle’s health state and performance limitations. The proposed framework is demonstrated on a Crazyflie 2.0 quadrotor, using both high-fidelity simulation and flight experiments. Results show that the HMS can generate a data-driven flight envelope from nominal flight data and that the learned decision boundary effectively serves as a CBF constraint, enabling enforcement of health-aware safe operation.