Ensuring aviation safety through intelligent fault diagnosis of bearing by interpretable deep transfer learning based on wavelet transform

While ensuring the aviation safety, application of convolution neural network is largely dependent on the labels of the training data collected from a bearing. However, in real life application, it is difficult to obtain health status labels for bearing data collected in actual operating conditions. Furthermore, convolution kernel used in the feature extraction process lacks physical meaning. To solve the aforementioned problem, this paper proposes an interpretable deep transfer learning algorithm based on wavelet convolution kernel. The proposed algorithm uses wavelet convolution kernels to extract features from labeled simulation data while simultaneously utilizing dynamic simulation signals to train deep network models. Finally feature transfer learning methods have been applied to achieve fault diagnosis. Experimental verification shows that the proposed method can transfer the knowledge learned from the labeled bearing dynamics simulation data to experimentally measured data for performing fault classification. Additionally, the proposed method effectively solves the problem of data label scarcity in the fault diagnosis process and can perform better clustering and fault classification of measured data.