Health Condition Estimation of Bearings with Multiple Faults by a Composite Learning-Based Approach.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence

Background:

• Bearings are essential components in rotating machinery, and their health is critical for industrial operations.
• Multiple bearing faults present significant diagnostic challenges, including fault masking and complex signal interference, which existing methods struggle to address.
• Current deep learning models often lack sufficient feature diversity for reliable multiple fault diagnosis.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework for accurate multiple bearing fault diagnosis.
• To address the limitations of existing methods in handling complex fault scenarios and noisy data.
• To improve the generalized diagnostic capability of bearing health management systems.

Main Methods:

• Utilized extended feature sets within three homogenous deep learning models to increase data diversity.
• Employed blending ensemble learning to fuse the outputs of multiple models for complementary solutions.
• Validated the approach using vibration datasets specifically containing multiple bearing faults.

Main Results:

• Achieved a diagnostic accuracy of 98.54% for multiple bearing faults.
• Demonstrated a 2.74% improvement in overall effectiveness compared to single deep learning models.
• Showcased enhanced generalized diagnostic capability compared to existing technologies.

Conclusions:

• The proposed deep learning framework with extended features and ensemble learning effectively diagnoses multiple bearing faults.
• This approach overcomes challenges like fault submergence and non-Gaussian noise, offering a more robust solution.
• The findings indicate a significant advancement in bearing health management and predictive maintenance.