A bearing fault diagnosis method based on hybrid artificial intelligence models

Area of Science:

• Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearing performance is critical for industrial equipment.
• Extracting incipient weak fault signals is challenging for accurate diagnosis.
• Existing optimization algorithms may converge to local optima.

Purpose of the Study:

• To propose an efficient bearing fault diagnostic technique using hybrid artificial intelligence models.
• To enhance the accuracy and generalization capabilities of bearing fault detection.
• To address the limitations of conventional optimization algorithms in feature extraction.

Main Methods:

• Maximum Second-order Cyclostationary Blind Deconvolution (CYCBD) for noise filtering.
• Improved Harris Hawks Optimization (IHHO) with differential evolution mutation and nonlinear escape energy.
• A hybrid IHHO-DBN-ELM model optimizing Deep Belief Networks and Extreme Learning Machines structure.
• Application to the Western Reserve University's bearing fault dataset.

Main Results:

• Successful extraction of fault characteristics from raw time-domain vibration signals.
• Demonstrated enhanced diagnostic accuracy compared to conventional methods.
• Achieved superior generalization capabilities in bearing fault detection.

Conclusions:

• The proposed IHHO-DBN-ELM approach offers an effective solution for rolling bearing fault diagnosis.
• The hybrid model overcomes limitations of traditional methods in weak signal feature extraction.
• This technique significantly improves the reliability and performance of industrial equipment.