Few-shot bearing fault detection based on multi-dimensional convolution and attention mechanism

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Bearing failures pose significant risks to industrial safety and production.
• Acquiring large datasets for bearing fault detection is challenging due to small sample sizes.
• Existing methods may fail to capture crucial features in bearing vibration signals.

Purpose of the Study:

• To develop an effective bearing fault detection method for small sample scenarios.
• To improve the feature extraction capabilities for bearing vibration signals.
• To enhance the safety and reliability of industrial physical systems.

Main Methods:

• A multichannel preprocessing technique for utilizing bearing vibration signal information.
• Multidimensional convolution and attention mechanisms for enhanced feature extraction.
• Nonlinear mapping into a metric space for improved sample similarity measurement.

Main Results:

• The proposed method demonstrates robust fault detection performance with limited data.
• Improved feature extraction leads to more accurate bearing fault identification.
• The approach effectively measures sample similarity, enhancing detection accuracy.

Conclusions:

• The developed few-sample learning method effectively addresses challenges in bearing fault detection.
• The technique offers significant benefits for reducing machine downtime and economic losses.
• This contributes to ensuring the safe operation of industrial equipment.