Fault Diagnosis Based on Interpretable Convolutional Temporal-Spatial Attention Network for Offshore Wind Turbines

Fault diagnosis of offshore wind turbines is instrumental to its operation and maintenance. To effectively diagnose faults in the very early stage, this study firstly proposes a condition monitoring based sample mining method from SCADA time-series data. Then, based on the convolutional neural network and attention mechanism, an interpretable convolutional temporal-spatial attention network (CTSAN) model is proposed. The proposal can extract deep temporal-spatial features from SCADA data sequentially by (1) a convolution module to extract features based on time intervals, (2) a spatial attention module to extract spatial features considering their weights, and (3) a temporal attention module to extract temporal features considering the weights of intervals. The proposal has the superiority of interpretability by exposing the deep temporal-spatial features extracted in a human-understandable form of temporal-spatial attention weights. The effectiveness and superiority of the proposal are verified by numerical studies on a real offshore wind farm in China.