Bearing Fault Diagnosis Method Based on CWT-ResNet18

A bearing fault diagnosis method based on CWT-Resnet18 is proposed to address the inability of one-dimensional vibration signals as the input of convolutional neural networks to fully utilize their advantages in image processing. The collected time-series vibration signals are transformed into feature images through CWT (Continuous Wavelet Transform) and then input into the Resnet18 network for fault signal recognition and diagnosis. The CWRU (Case Western Reserve University) public bearing dataset is used to test the fault diagnosis performance of the CWT-Resnet18 model and compared with other common convolutional neural network models such as CWT-CNN (Convolutional Neural Networks) to validate the proposed model. The experimental results show that the proposed model shows better fault diagnosis performance than other common neural networks under the influence of complex noise. The model converges rapidly. After three epochs, the accuracy of the validation set has reached over 90%, with a maximum accuracy of 99.7%.