Research on Intelligent Fault Diagnosis Method of Wind Turbine Gearbox Based on Multimodal Voiceprint Fusion

The gearbox fault of wind turbine is the main reason for unexpected shutdown and increased maintenance costs, so accurate diagnosis is essential to ensure the reliability and cost-effectiveness of wind turbine. However, most of the existing diagnosis methods fail to fully extract the spatio-temporal fractal features in the data, ignoring the impact of class imbalance, which limits the accuracy of diagnosis. Therefore, a fault diagnosis model of wind turbine gearbox based on improved convolutional neural network (ICNN) and gated recurrent unit (GRU) is proposed. The ICNN-GRU spatio-temporal feature extraction network is constructed, in which CNN extracts the local spatial features of the data, and the attention shifting mechanism effectively fuses the channel and spatial information to enhance the spatial representation. GRU captures long-term spatiotemporal dependencies. The traditional cross entropy loss is replaced by the focus loss, giving higher weight to the fault samples that are difficult to classify. The experimental results show that the average precision of ICNN-GRU model is 96.36%, which is better than other models.