Demonstration of the In-Time Learning-Based Safety Management for Scalable Heterogeneous AAM Operations

Our research team proposes the design, development and demonstration of an in-time learning-based aviation safety management system (ILASMS) for scalable heterogeneous advanced air mobility (AAM) operations. The proposed innovations include: 

• An integrated safety management system architecture spanning across onboard systems, ground control stations, and cloud-based services; 
• Data-driven learning-based detection, assessment and prediction of vehicle anomalies and mission risks; 
• Verifiable automated operational mitigation for resilience and scalability; and 
• The closed-loop interface between detection/prediction and mitigation with safety guarantees for mitigation in the presence of imperfections and possible detection errors introduced by learning-based approach.

Vehicle self-awareness and health monitoring during flight is the key for individual aircraft safety. In this project, we integrate model-based and data-driven methods for diagnosis and prognosis of vehicle component faults and degradation in the electric propulsion system (i.e., the battery and motors) and the GPS navigation system. We are currently developing a Reinforcement Learning-based fault adaptive control mitigation scheme that takes into account degradation in battery and motors as well as varying environmental conditions caused by wind disturbances.