A Novel Multistep Wavelet Convolutional Transfer Diagnostic Framework for Cross-Machine Bearing Fault Diagnosis

Transfer learning has emerged as a potent technique for diagnosing bearing faults in environments with fluctuating operational parameters. Nevertheless, the majority of current transfer-learning-based fault diagnosis approaches focus primarily on adapting to varying conditions within the same machine. In real-world applications, there is a frequent need to extend these diagnostic techniques to machines that differ significantly in both function and structural design. Due to the different mechanical structures of different machines, the signal transmission paths are vastly different, and the distribution of collected data varies greatly, making it difficult for existing transfer fault diagnosis methods to meet diagnostic needs. Therefore, a multistep wavelet convolutional transfer diagnostic framework (MSWCTD) is proposed to realize cross-machine bearing fault diagnosis. Firstly, a multistep time shift wavelet convolutional network (MTSWCN) based on the multiscale technique and wavelet transform is proposed to explore the diversity information regarding original vibration data and enhance the feature expression ability. Secondly, a confusion transfer method based on multi-view learning is designed to extract diagnosis knowledge that is transferable, which reduces the discrepancy between machines. Three bearing datasets are utilized to evaluate the MSWCTD, with the MSWCTD showing excellent performance on cross-machine bearing fault diagnosis task.