Hybrid Diagnostic Framework for Interpretable Bearing Fault Classification Using CNN and Dual-Stage Feature Selection

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning, especially Convolutional Neural Networks (CNNs), excels in vibration-based fault classification for rotary machinery.
• Limited interpretability of deep learning models hinders their use in safety-critical industrial settings.
• Timely fault diagnosis is crucial for system reliability and minimizing downtime.

Purpose of the Study:

• To develop a hybrid diagnostic framework integrating CNN transfer learning with interpretable supervised classification.
• To enhance both predictive accuracy and model transparency in fault diagnosis.
• To provide explainable and reliable fault diagnosis solutions for industrial environments.

Main Methods:

• A dual-stage feature selection process using Analysis of Variance (ANOVA) and Permutation Feature Importance (PFI) was employed.
• Deep features were extracted from a pre-trained VGG19 network and refined.
• SHapley Additive exPlanations (SHAP) were utilized for model interpretability.

Main Results:

• The proposed framework achieved 100% classification accuracy on the Case Western Reserve University (CWRU) bearing dataset.
• The dual-stage feature selection effectively reduced dimensionality and improved classification performance.
• SHAP analysis provided insights into influential features driving fault classification.

Conclusions:

• The hybrid framework successfully combines high performance with transparent decision-making for fault diagnosis.
• The approach demonstrates strong potential for explainable and reliable fault diagnosis in industrial settings.
• Integrating interpretable methods with deep learning enhances the practical applicability of AI in machinery diagnostics.