Deep Learning Based Sensor Integrity Monitoring for Robust AUV Navigation
Year: 2026
Type: conference-paper
Abstract: Autonomous Underwater Vehicles (AUVs) operating in shallow and cluttered environments are highly susceptible to sensor anomalies that can degrade navigation performance and compromise mission reliability. Sensors such as the Doppler Velocity Log (DVL), pressure sensor, and Inertial Measurement Unit (IMU) are affected by disturbances including bottomlock loss, acoustic interference, and inertial drift. Distinguishing between short-term degradation and persistent sensor faults in real time remains a critical challenge. This work presents a data-driven framework for detecting sensor faults and short-term degradation from sequential measurements. A synthetic dataset is generated using a high-fidelity 6-DOF AUV simulator with physically grounded disturbance models across multiple sensors. A multi-head Long Short-Term Memory (LSTM) network is trained to classify each sensor at every timestep as clean, degraded, or faulty, enabling temporal reasoning over evolving sensor behavior. To evaluate performance at the mission level, a scenario-based framework is introduced, covering 20 representative combinations of degradation and fault conditions. Results demonstrate reliable detection of both isolated and combined anomalies, with consistent performance across diverse disturbance patterns while remaining computationally efficient for real-time deployment.
Institution: Indian Institute of Technology Madras
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7201848380
