A Novel Cross-Domain Mechanical Fault Diagnosis Method Fusing Acoustic and Vibration Signals by Vision Transformer

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Shifting signal features due to changing operating conditions degrade bearing fault diagnosis accuracy.
• Existing models struggle with feature robustness and generalization in dynamic environments.
• Advanced deep learning architectures are needed for reliable condition monitoring.

Purpose of the Study:

• To propose a novel dual-channel parallel adversarial network (DPAN) for enhanced bearing fault diagnosis.
• To improve the robustness and generalization ability of diagnostic models against operational variations.
• To achieve high diagnostic accuracy exceeding 98% in experimental validation.

Main Methods:

• Developed a dual-channel parallel adversarial network (DPAN) utilizing vision transformers.
• Extracted features from acoustic and vibration signals via parallel network channels.
• Employed adversarial training and Wasserstein distance for feature fusion and domain adaptation.

Main Results:

• The DPAN demonstrated superior diagnostic accuracy compared to existing methods in bearing fault diagnosis experiments.
• Experimental results confirmed the method's effectiveness, achieving diagnostic accuracy above 98%.
• Adversarial training and Wasserstein distance significantly enhanced feature robustness and network generalization.

Conclusions:

• The proposed DPAN effectively addresses the challenge of shifting signal features in bearing fault diagnosis.
• The method offers a robust and accurate solution for condition monitoring in dynamic industrial settings.
• DPAN shows significant potential for improving the reliability of machinery diagnostics.