Research on Bearing Fault Diagnosis Method for Varying Operating Conditions Based on Spatiotemporal Feature Fusion

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Bearing rotational speed variability causes inconsistent vibration data features, hindering direct model application across operating conditions.
• Limited labeled data in target domains complicates fault diagnosis under changing operational parameters.

Purpose of the Study:

• To propose a Spatiotemporal Feature Fusion Domain-Adaptive Network (STFDAN) for robust bearing fault diagnosis under varying operating conditions.
• To develop a framework capable of handling domain shift and data scarcity in bearing fault diagnosis.

Main Methods:

• Utilized Fast Fourier Transform (FFT) and Variational Mode Decomposition (VMD) for multi-level spectral and modal feature extraction.
• Employed a parallel Convolutional Neural Network (SECNN) and Bidirectional Long Short-Term Memory (BiLSTM) network for spatiotemporal feature extraction.
• Integrated cross-attention for feature fusion and Multi-Kernel Maximum Mean Discrepancy (MKMMD) for domain distribution alignment.

Main Results:

• The STFDAN framework demonstrated high diagnostic accuracy across different load conditions using benchmark datasets (CWRU, JNU, SEU).
• Achieved effective fault diagnosis despite domain shifts caused by varying operating conditions.

Conclusions:

• The proposed STFDAN framework successfully addresses the challenge of bearing fault diagnosis under varying operating conditions.
• The method offers a promising solution for real-world applications requiring reliable condition monitoring of bearings.