Fault Feature Extraction Method of Rotating Machinery Based on DeepLabV3+ Semantic Segmentation

The requirements for the health status of rotating machinery equipment continue to increase, and traditional fault diagnosis methods cannot efficiently and intelligently process the collected large-capacity rotating machinery fault test data. In the paper, we combine the advantages of time-frequency representation and deep learning semantic segmentation, and propose the Fault Feature Extraction Method of Rotating Machinery Based on DeepLabV3+ Semantic Segmentation. First, the continuous wavelet transform is used to represent the fault vibration signals in different health states in time-frequency, so as to fully represent the time-varying and non-stationary fault characteristic information of different faults and the time-varying frequency characteristic information. Then in order to efficiently and intelligently identify these features and use them as segmentation targets, a DeepLabV3+ deep learning semantic segmentation model is constructed to accurately extract information of different fault categories. Finally, the model is trained and tested with the rolling bearing failure data, and the influence of the number of training samples on the performance of the model is analyzed. The accuracy of the background and fault features are 87.20% and 92.21%, respectively. The total accuracy of the DeepLabV3+ model is 89.5%, the Kappa coefficient is 0.7895, and the loss is 0.1109. The results prove that the proposed method can accurately extract fault features from a large number of time-frequency representations.