A Knowledge-Guide Data-Driven Model with Selective Wavelet Kernel Fusion Neural Network for Gearbox Intelligent Fault Diagnosis

The gearbox is a critical component in modern industrial systems, directly determining the operational reliability of machinery. Therefore, effective fault diagnosis is essential to ensure its proper functioning. Modern diagnostic approaches often employ accelerometers to monitor vibration signals and apply data-driven techniques for fault identification, achieving considerable success. However, deep learning-based methods still face limitations due to their "black-box" nature and lack of interpretability. To address these issues, this paper proposes a knowledge-guided selective wavelet kernel fusion neural network. By integrating diagnostic domain knowledge into data-driven modeling, the proposed method enhances both the interpretability and diagnostic performance of intelligent fault diagnosis systems. First, a multi-kernel convolutional module is designed based on domain knowledge and embedded into a Modern Temporal Convolutional Network. Then, an attention-based selective wavelet kernel fusion strategy is introduced to adaptively fuse kernels according to the distribution of different datasets. Finally, the effectiveness of the proposed method is validated on two public datasets. Experimental results demonstrate that the approach not only provides prior interpretability, which overcoming the black-box limitation of deep learning, but also further improves diagnostic accuracy.