A gearbox fault diagnosis method based on Vision Transformer-BiGRU parallel network.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Traditional single-modal fault diagnosis methods often lose critical information and have limited representation capabilities.
• This leads to lower fault identification rates in complex machinery like gearboxes.

Purpose of the Study:

• To propose a novel multi-modal gearbox fault diagnosis method to overcome limitations of traditional approaches.
• To enhance the accuracy and reliability of fault identification in gearboxes.

Main Methods:

• Utilized short-time Fourier transform (STFT) to convert 1D vibration signals into time-frequency patterns.
• Developed a parallel network integrating Vision Transformer for image features and Bidirectional Gated Recurrent Unit (BiGRU) for temporal features.
• Fused multi-modal features and employed a Softmax classifier for fault classification.

Main Results:

• The proposed Vision Transformer-BiGRU parallel network achieved the highest fault identification rate.
• Demonstrated an average diagnostic accuracy of 99.69% on a gearbox test bench dataset.
• Outperformed other intelligent diagnostic methods in experimental validation.

Conclusions:

• The multi-modal approach effectively extracts comprehensive fault information, enhancing diagnostic accuracy.
• The proposed method is feasible and shows significant potential for intelligent gearbox diagnosis and practical applications.