Autoencoder-based Fault Detection Strategy for Autonomous Underwater Reconfigurable Vehicles

This work presents an advanced fault detection strategy for Autonomous Underwater Reconfigurable Vehicles (AURVs) leveraging Deep Learning techniques, specifically autoencoders. As AURVs are increasingly utilized for complex underwater missions, ensuring their operational reliability is critical. The proposed method addresses the challenges of fault detection in such environments by implementing an autoencoder-based approach to identify anomalies in the vehicle's thruster performance. The strategy's effectiveness is validated through extensive simulations under both 'survey' and 'hovering' configurations. The model demonstrated high accuracy in both the 'survey' configuration and 'hovering' configuration, effectively distinguishing between faulty and non-faulty states with minimal false positives. The results indicate that the autoencoder approach is robust, providing reliable fault detection that can enhance the safety and performance of AURVs in dynamic and unpredictable underwater environments.