Bearing Faults Diagnosis Method Based on Stacked Auto-Encoder With Graph Regularization for Wind Turbines

ObjectivesIn order to solve the problems of low efficiency of fault feature extraction, inaccurate feature representation, and difficulty in adapting existing methods to complex signal requirements in wind turbine bearing fault diagnosis, a fault diagnosis method based on graph regularization was proposed. The method helps to improve the analysis ability of vibration signals, thereby achieving accurate classification and reliable diagnosis of different fault types.MethodsThe technology based on graph regularization auto-encoder was adopted, combined with the idea of graph embedding, to guide the stack-supervised auto-encoder to carry out feature extraction. In the fault feature extraction stage, the diagnostic signal was first graphically represented, and then graph regularization terms were added to the stacked auto-encoder to ensure that the embedded low-dimensional features maintain the manifold structure, thereby extracting complex geometric features deep in the data.ResultsThe extracted features can accurately classify different fault types, showing significant advantages in fault feature capture. Experimental results show that this method has higher diagnosis accuracy and reliability in the application of actual wind farm data, and effectively improves the extraction efficiency and classification accuracy of fault features.ConclusionsThe proposed method has shown significant effectiveness and advantages in the field of wind power bearing fault diagnosis, and provides reliable technical support for practical applications.