Research on Gearbox Fault Diagnosis Method Based on ECA Improved VGG16

To overcome the limitations of traditional neural network training samples that require a large amount of data, the idea of transfer learning is utilized to migrate the VGG16 model within the field of image recognition to this field, and a deep transfer learning model is constructed to complete the task of wind turbine gearbox fault diagnosis. The data structure is changed to transform the time-domain vibration signal into a two-dimensional time-frequency map using the discrete wavelet transform, and then the Efficient Channel Attention (ECA) mechanism is used to realize the reduction of feature dimensionality loss while capturing the feature channel information interactions to improve the accuracy of gearbox fault diagnosis. Finally, the image after changing the data structure is input into the deep migration learning model for training, and the degree of weight migration and the model structure are fine-tuned during the experimental process, finally, the diagnostic capability of the model for the fault diagnosis and classification task is evaluated, and the corresponding validation experiments are accomplished on the public dataset of Southeast University.