Bearing fault diagnosis method based on WSST and ISSA-MCNN-BIGRU

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearings are critical in rotating machinery, but their fault diagnosis faces challenges in feature extraction and accuracy.
• Existing methods often rely heavily on expert experience, limiting automated diagnostic capabilities.

Purpose of the Study:

• To develop a robust and accurate fault diagnosis framework for rolling bearings.
• To overcome limitations in feature extraction and recognition rates of current diagnostic methods.

Main Methods:

• A hybrid framework combining Wavelet Synchrosqueezed Transform (WSST) for signal representation, Multi-Scale Convolutional Neural Network (MCNN) for spatial feature extraction, and Bidirectional Gated Recurrent Unit (BiGRU) for temporal dependency learning.
• An Improved Sparrow Search Algorithm (ISSA), incorporating chaotic Tent mapping, Gaussian mutation, and Levy flight, was used for adaptive hyperparameter optimization of the MCNN-BiGRU network.
• The framework was validated using bearing datasets from Case Western Reserve University and Southeast University.

Main Results:

• The proposed ISSA-MCNN-BiGRU model achieved a maximum fault diagnosis accuracy of 99.75%.
• The model demonstrated superior performance compared to baseline models (GRU, BiGRU, MCNN-BiGRU, PSO-MCNN-BiGRU, GA-MCNN-BiGRU) in accuracy, stability, and generalization.
• The framework exhibited strong robustness and significantly higher accuracy in various noise environments.

Conclusions:

• The developed hybrid diagnostic framework offers a highly accurate and robust solution for rolling bearing fault diagnosis.
• The integration of WSST, MCNN, BiGRU, and ISSA effectively addresses challenges in feature extraction and diagnostic accuracy.
• This approach shows significant potential for improving the reliability and safety of large-scale rotating machinery.