Reinforcement learning for multi-agent systems under semantic and perceptual uncertainties

Autonomous systems are expected to revolutionize vehicle industries, from automotive and commercial vehicles to agricultural machinery. This shift will contribute to sustainability by optimizing driving behavior, reducing energy consumption and emissions, and efficient use of transportation systems. Full delegation of responsibility from the conventional human-sense driving to the vehicles autonomous systems requires replacement of human perception concerning mechanical integrity to be able to emergency stop in the event of severe failures and making reliable decisions based on self-identification and diagnosis of influencing mechanical faults.

The goal is to develop fault diagnosis algorithms to predict, detect and diagnose vehicle anomalies and faults through monitoring of dynamic behavior. Due to complex structural dynamic behavior of vehicle systems, a reliable fault diagnosis that is capable of successfully distinguishing a faulty- from non-faulty condition under different driving scenarios, may be challenging based on only measurement data. To address this challenge, a hybrid data-driven- & structural-dynamic model-based approach is developed for fault diagnosis of mechanical systems. Furthermore, an innovative method is developed for structural fault identification within continuous structural components using dynamic model-based fault diagnosis approach.