Evaluation of Different Bearing Fault Classifiers in Utilizing CNN Feature Extraction Ability

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Condition Monitoring

Background:

• Bearing fault diagnosis is critical in heavy industries (aerospace, marine) to enhance machine lifespan, minimize economic losses, and prevent safety incidents.
• Traditional fault diagnosis methods struggle with effective feature extraction from raw bearing data.
• Deep neural networks offer automated feature extraction, reducing reliance on expert knowledge.

Purpose of the Study:

• To evaluate the effectiveness of hybrid models combining Convolutional Neural Networks (CNNs) with different classifiers for bearing fault diagnosis.
• To identify classifiers that best leverage CNN's feature extraction capabilities for improved fault detection accuracy and efficiency.

Main Methods:

• Development and comparison of four hybrid models integrating CNNs with various individual classifiers.
• Comparative analysis focused on fault detection accuracy and diagnostic efficiency of each hybrid model.
• Investigated the synergy between CNN feature extraction and classifier performance.

Main Results:

• The study demonstrated that hybrid models significantly improve bearing fault diagnosis compared to traditional methods.
• Random Forest (RF) and Support Vector Machine (SVM) classifiers were found to effectively utilize CNN-extracted features.
• These specific hybrid models showed superior performance in fault detection accuracy and efficiency.

Conclusions:

• Hybrid models integrating CNNs with classifiers like RF and SVM are highly effective for bearing fault diagnosis.
• The choice of classifier is crucial for maximizing the performance of CNN-based feature extraction in bearing fault detection.
• These findings contribute to more reliable and efficient machine health monitoring in industrial applications.