A real-time vertical plane flight anomaly detection system for a long range autonomous underwater vehicle

Abstract

For autonomous underwater vehicles (AUVs) to be successful in long duration deployments, they must be reliable in the face of subsystem failure and environmental challenges. The ability to detect performance anomalies and unexpected events in real time, especially in the vertical plane, is critical for the vehicle's survivability (the AUV must surface for recovery) and important for planning and vehicle operations. To this end, we have developed a vertical plane flight anomaly detection algorithm capable of comparing observed vehicle performance to references of expected behavior onboard the Tethys class long-range AUV in real time. The detection algorithm operates based on statistical characterization of training datasets that represent normal vertical plane performance. These datasets are taken directly from previous long-range AUV field operations. From this analysis we have derived a series of conditional tests that monitor representative components of the vehicle state (e.g., depth rate, pitch angle, and stern plane angle). In the months of January, February and March 2015, we conducted a series of tests in Monterey Bay, CA. The Daphne long-range AUV ran the algorithm to detect and flag vertical plane performance anomalies in real time. The AUV was successful in discriminating between expected vertical plane flight performance and anomalies during long-duration deployments lasting more than 11 days.