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

Area of Science:

• Mechanical Engineering
• Artificial Intelligence

Background:

• Gearbox operational reliability is crucial for industrial systems.
• Vibration analysis using accelerometers and data-driven methods are common for fault diagnosis.
• Deep learning methods offer high accuracy but lack interpretability.

Purpose of the Study:

• To develop an interpretable and accurate intelligent fault diagnosis system for gearboxes.
• To address the "black-box" limitation of existing deep learning approaches.

Main Methods:

• Proposes a knowledge-guided selective wavelet kernel fusion neural network.
• Integrates diagnostic domain knowledge into a Modern Temporal Convolutional Network (TCN) using a multi-kernel convolutional module.
• Employs an attention-based selective wavelet kernel fusion strategy for adaptive kernel merging.

Main Results:

• The proposed method demonstrates enhanced interpretability compared to traditional deep learning models.
• Experimental validation on public datasets shows improved diagnostic accuracy.
• Successfully overcomes the "black-box" limitation in intelligent fault diagnosis.

Conclusions:

• The knowledge-guided approach effectively enhances both interpretability and diagnostic performance.
• The selective wavelet kernel fusion strategy improves adaptive learning capabilities.
• This method offers a promising direction for developing more transparent and reliable intelligent diagnostic systems.