Fault diagnosis method of rolling bearing based on GADF and embedded image segmentation in ResNet34

Traditional fault detection classification algorithms for vibration data are trained using one-dimensional time series signals. There are problems such as large number of algorithm parameters, slow computation speed and low accuracy rate. In order to apply the advantages of two-dimensional data processing to one-dimensional data, this study proposes a bearing fault data diagnosis and classification method based on the Gramain Angular Field (GAF) and Residual Neural Network (ResNet34) model with Patch embedding. The method converts one-dimensional vibration signals of rolling bearing fault data into two-dimensional images. Through image segmentation and embedding techniques, these 2D images are transformed into sequences consisting of multiple Patches, and the expanded vectors are mapped to fixed dimensional inputs for residual neural network processing. In addition, a channel attention mechanism is introduced to enhance the model performance. The method is validated on the Case Western Reserve University (CWRU) bearing dataset. The ablation experiments show that the proposed method has significant advantages over the original ResNet34 model, 60% increase in training speed and accuracy improved by 5%. The method performs well in terms of training speed and accuracy compared to traditional deep learning models and is particularly suitable for vibration data fault detection and classification tasks.