DTCWPNet: A Novel Interpretable Intelligent Fault Diagnosis Model for Rotating Machinery Systems Based on Sensor Data

Mechanical transmission systems are extensively applied in various industrial equipment and constitute a fundamental component of modern industrial production. With the increase in operating time, various faults inevitably emerge in their critical components. In recent years, deep learning-based methods have been widely applied in the field of fault diagnosis using sensor data. However, in engineering practice, these methods still face challenges such as poor interpretability and insufficient noise resistance. This paper presents a method that utilizes the dual-tree complex wavelet packet transform (DTCWPT) to replace the "black-box" components in traditional convolutional neural networks (CNNs). Specifically, it substitutes the conventional convolutional kernels, activation functions, and pooling layers with wavelet filters, Bayes shrink soft-threshold activation functions, and energy spectrum pooling layers. Additionally, the suggested method explores the decision-making process of the model by visualizing filters, feature maps, and reconstructed signals. The superiority of the proposed method is validated on two datasets, rolling bearings and planetary gearboxes. In contrast experiments, anti-noise experiments, and ablation experiments, the proposed method achieves the best performance. The model presented in this paper offers a novel approach to detecting machinery faults in practical industrial applications, which combines both noise robustness and interpretability.