Deep Domain Adaptation Network for Drive Train of the Wind Turbine Fault Diagnosis Using Acoustic Emission

The drive train of the wind turbines needs to adapt to the rotational speed changes caused by wind speed fluctuations. The gearbox needs to maintain stable operation under nonsmooth working conditions, resulting in complex sensor signals and difficult extraction of fault characteristics. This paper explores the fault diagnosis method of rotating machinery based on acoustic emission records. Firstly, in order to reduce the dimensionality of the records while enhancing the component state information, the processing methods of framing, windowing, and periodic feature extraction are adopted to generate feature images. Secondly, the ResNet50 model is adopted for deep feature extraction. Meanwhile, an adversarial network is constructed to break the distributional difference between the source and the target domain. Further, the margin disparity algorithm is adopted between the classification and domain discrimination models to constrain the marginal distribution of the classifier and reduce the feature distribution differences between the source domain and the target domain. Experimental research and results show that this method has achieved relatively excellent recognition rates of component status across low‐speed operating conditions. It provides new clues for the fault diagnosis of wind power transmission chains.