A fault diagnosis method based on attention mechanism with application in Qianlong-2 autonomous underwater vehicle

Abstract This study proposed a fault diagnosis method based on deep learning and attention mechanism for autonomous underwater vehicle (AUV). Firstly, a data attention mechanism is proposed to introduce dynamic weighting coefficients of monitoring variables to realize dynamic decorrelation. Then, the automatic feature engineering is realized by a bi-directional gated recurrent unit (GRU) network to acquire the time dynamic characteristics of monitoring variables. Finally, fault detection is implemented via multi-layer perceptron (MLP). With respect to fault identification, this study embeds a spatial attention mechanism in the fault detection network to capture the semantic relationship between monitoring variables and faults, and fault identification result can be obtained by parsing this semantic relationship. We present a new loss function and training strategy for cooperation between the fault detection and identification tasks. The proposed method is validated on the monitoring data of Qianlong-2 AUV obtained during the mission in the South China Sea, which shows the effectiveness and superiority of the method.