Research on fault diagnosis of wind turbine generator bearings under variable operating conditions

The operating conditions of wind turbine (WT) bearings are constantly changing due to various factors such as wind speed, air pressure and wind direction. The distributional differences of vibration signals under different working conditions significantly hinder the accuracy of intelligent diagnostic models for bearings. To address this, a fault diagnosis method for WT generator bearings under variable operating conditions is proposed, based on a domain adaptive sparse self-attention convolutional neural network (DA-SSACNN). The DA-SSACNN first employs sparse filtering (SF) to effectively remove background noise from the vibration signals of WT generator bearings. Secondly, a feature extractor with a self-attention mechanism is utilised to learn deep features. Next, domain adaptation is applied to reduce the distribution differences between bearing signals under different working conditions. Finally, the classifier is used to identify bearing faults. Through comparative experiments, the DA-SSACNN has been shown to provide the best diagnostic performance, both under variable operating conditions and in environments with significant background noise, achieving a diagnostic accuracy of 97.2%, which surpasses other methods. This paper offers a theoretical foundation for the intelligent diagnosis of WT bearings.