Research on an interpretable fault diagnosis method for wind turbine bearings under variable operating conditions

Abstract

During the operation of wind turbine bearings, the rotational speed continuously fluctuates with changes in wind intensity. At different rotational speeds, bearing faults exhibit distinct characteristic patterns and the accuracy of intelligent diagnostic methods significantly decreases. Regarding this challenge, a multi-scale fusion transfer model based on HigherHRNet (MSFTM-HigherHRNet) is proposed in this paper for bearing fault diagnosis of wind turbine generators. The proposed MSFTM-HigherHRNet model can effectively reduce the distribution differences of fault data at different speeds and achieve high-precision fault diagnosis. Firstly, to enhance the interpretability of the model, an S-transformation is performed on the vibration signals from both the source and target domains to extract features in the time-frequency domain. Secondly, HigherHRNet is utilised to extract multi-scale features. Feature extraction and reconstruction are performed separately for each scale, followed by multi-scale feature fusion to obtain a deep fault feature. Furthermore, a subdomain alignment module (SAM) is constructed. The unsupervised classification results of samples using K-means yields subdomains under different categories and transfer learning is employed to align the subdomains, thereby reducing the distribution differences at different rotational speeds. Finally, the effectiveness of this method is verified using actual wind field test data. The results show that its diagnostic accuracy exceeds 95%, which is superior to existing fault diagnosis methods at present. This research provides an effective solution path for the fault diagnosis of wind turbine bearings under variable working conditions.