A novel fault diagnosis method for gear based on dual-channel ResNet34

In response to problems such as the scarcity of fault samples and the complexity of data in the industrial sector, a dual-channel feature fusion-based intelligent diagnosis method for gears is proposed. For ensuring the diagnostic accuracy of the model, one-dimensional vibration signals are converted into images of Markov Transition Field and Continuous Wavelet Transform. A feature fusion layer is designed to effectively combine the critical features from the two types of images, innovatively designed a dynamic weight adaptive fusion module that enhances the capability to extract fault-related information. To validate the effectiveness of the proposed method, fault classification experiments were conducted on two datasets, and the performance was compared with single-channel network such as AlexNet, LeNet5, ConvNeXt, and ResNet34. Experimental results show that the proposed method can effectively process fault data under varying operating conditions and significantly improve diagnostic accuracy. The method exhibits strong generalizability and provides a promising solution for industrial fault diagnosis.