Intelligent Anomaly Detection for Hydraulic System in Deep-Sea Submersible

Abstract

The hydraulic system used in the deep-sea submersible of China works in the extreme conditions of the deep-sea environment, where the pressure is high, the temperature is low, and the landscape is complicated, which contributes to the risks of critical failures. Anomaly detection is essential to the safety of crews and mission success. However, traditional techniques have two problems: small samples of faults and severe class imbalance (with few faults hidden in the vast sea of normal activity). To cope with these issues, we suggest the Aux-BiGAN, an unsupervised anomaly detector that is specific to the hydraulic system. The approach enhances learning normal data distributions and increases the sensitivity of anomaly detection but traines on normal data only by auxiliary losses. It is tested on actual hydraulic system defects and artificial data of hard failures and incipient soft anomalies. In real-world conditions, it is able to identify severe failures, including hydraulic oil leakage, electronic control unit burnout, and sticking solenoid valves, and its results are supported by postdive maintenance reports by pilots. Experimental results demonstrate the effectiveness of the proposed method, which achieves an area under the receiver operating characteristic curve of 0.9104 on hard anomalies. It outperforms the baseline BiGAN across all metrics. It surpasses state-of-the-art methods in terms of accuracy, F1-score, and false alarm rate across both failure types, validating its enhanced reliability for deep-sea missions.