Fault Diagnosis of Wind Turbine Gearbox Based on Multiscale Residual Features and ECA-Stacked ResNet

In the fault diagnosis of a wind turbine planetary, the use of multiscale features (MFs) in high scales cannot comprehensively describe fault information. This limitation generally leads to a low fault diagnosis accuracy. Therefore, a fault feature extraction method based on multiscale residual features (MRFs) is proposed. Coarse-grained signals with residual information are obtained through multiscale residual processing. This method initially amplifies the signal dimensions of each scale and enriches fault information. Then, the MRFs are obtained using the relevant feature extraction method. To study the MRP effectiveness, the method is introduced into the spectral feature (SF) and the permutation entropy (PE). The multiscale residual SF and multiscale residual PE are obtained. These MRFs are placed in a classifier based on a 1-D convolutional neural network to train the diagnostic model. To further enrich the input feature information, the efficient channel attention (ECA)-Stacked residual neural network (ResNet) is proposed. The features of each layer are stacked to obtain the multichannel fault features. Using ECA, the weight of features under each channel is obtained through training to further improve the diagnostic performance of the model. Gearbox fault signals are collected by the Wind Turbine Drivetrain Diagnostics Simulator. The experimental results show that the proposed method can improve the accuracy of the gearbox fault diagnosis and, thus, has certain engineering application values.

Automated summary

In the fault diagnosis of wind turbine planetary, a fault feature extraction method based on multiscale residual features (MRFs) is proposed, which amplifies signal dimensions and enriches fault information. The MRFs are obtained and placed in a classifier to train the diagnostic model. The proposed method improves the accuracy of gearbox fault diagnosis and has engineering application values.