Deep Learning-Based Adaptive Neural-Fuzzy Structure Scheme for Bearing Fault Pattern Recognition and Crack Size...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Bearings are critical components in rotating machinery, exhibiting nonlinear behavior.
• Effective condition monitoring and data analysis are essential for bearing fault diagnosis.
• Existing methods often struggle with the complexity of bearing vibration signals.

Purpose of the Study:

• To develop a deep learning-based adaptive neural-fuzzy structure technique for precise bearing fault diagnosis.
• To improve the accuracy of pattern recognition and crack size identification in bearing systems.
• To establish a robust method for analyzing bearing vibration data.

Main Methods:

• Utilizing a support vector autoregressive-Laguerre model to approximate normal vibration signals and extract state-space equations.
• Designing an adaptive neural-fuzzy structure observer for normal and abnormal signal estimation using high-order variable structure techniques.
• Employing a convolution neural network (CNN) algorithm to detect and identify faults based on residual signals.

Main Results:

• The proposed technique demonstrated superior performance on the CWRU bearing vibration dataset.
• Significant improvements in average pattern recognition accuracy (up to 35.29%) were achieved compared to benchmark methods.
• Enhanced crack size identification accuracy was also observed, validating the method's effectiveness.

Conclusions:

• The deep learning-based adaptive neural-fuzzy structure technique offers a powerful tool for bearing fault diagnosis.
• The integration of support vector autoregressive-Laguerre modeling and CNN provides robust signal analysis capabilities.
• This approach represents a significant advancement in condition monitoring for rotating machinery.