TFN: An interpretable neural network with time-frequency transform embedded for intelligent fault diagnosis

摘要

Convolutional neural network (CNN) is widely used in fault diagnosis of mechanical systemsdue to its powerful feature extraction and classification capabilities. However, the CNN is atypical black-box model, and the mechanism of CNN’s decision-making is not clear, whichlimits its application in high-reliability-required fault diagnosis scenarios. To tackle this issue,we propose a novel interpretable neural network termed as time-frequency network (TFN),where the physically meaningful time-frequency transform (TFT) method is embedded intothe traditional convolutional layer as a trainable preprocessing layer. This preprocessing layernamed as time-frequency convolutional (TFconv) layer, is constrained by a well-designed kernelfunction to extract fault-related time-frequency information. It not only improves the diagnosticperformance but also reveals the logical foundation of the CNN prediction in a frequencydomain view. Different TFT methods correspond to different kernel functions of the TFconvlayer. In this study, three typical TFT methods are considered to formulate the TFNs and theirdiagnostic effectiveness and interpretability are proved through three mechanical fault diagnosisexperiments. Experimental results also show that the proposed TFconv layer has outstandingadvantages in convergence speed and few-shot scenarios, and can be easily generalized toother CNNs with different depths to improve their diagnostic performances.