Spatio-Temporal Attention-based Neural Network for Wind Turbine Blade Cracking Fault Detection

Cracking of wind turbine blades is one of the main faults affecting the safety and efficiency of wind turbines. Because the unit has multiple operating conditions, under different working conditions, the influence weight of each dimensional feature on the final result is different. To solve this problem, a network model based on self-attention mechanism and long and short-term memory network is proposed, which can not only dynamically adjust each feature in space according to different input data, but also capture temporal correlation of data. The model is divided into three stages. First, the self-attention mechanism is used to obtain the correlation between time and space, and the weight of each feature is dynamically adjusted according to the different input data. Then the processed features are segmented and the self-attention mechanism is used for local sequential processing. Finally, the global temporal correlation of the segmented data is obtained through LSTM network. Based on the actual problems of unit and time, this model presents an end-to-end fault warning scheme. The model integrates the self-attention mechanism and LSTM network together, which is more conducive to deeply mining the characteristics of data in time and space, so as to realize accurate warning of blade cracking fault. Finally, the experimental results show that the proposed model is superior to the traditional classification model in detection performance.