Fault Diagnosis of Rotating Machinery Using Supervised Machine Learning Algorithms with Integrated Data-Driven and...

Area of Science:

• Mechanical Engineering
• Machine Learning
• Signal Processing

Background:

• Rotating machinery is critical in industrial applications.
• Effective fault diagnosis is essential for operational reliability and safety.
• Traditional methods often struggle with complex operating conditions and fault types.

Purpose of the Study:

• To develop a supervised machine learning framework for vibration-based fault diagnosis.
• To integrate data-driven and physics-informed features for improved diagnostic performance.
• To evaluate the framework's robustness and generalization capabilities.

Main Methods:

• Acquired vibration data under variable load and operating conditions.
• Applied parallel signal processing (time, frequency, baseband, envelope analysis).
• Extracted statistical, spectral, and physics-based features, creating integrated feature sets.
• Trained and evaluated multiple classifiers (SVM, Random Forests, Gradient Boosting, ensemble) using standard metrics.
• Implemented a two-layer classification strategy for fault identification and coexisting fault detection.

Main Results:

• Models using integrated features outperformed those with purely data-driven features.
• The two-layer classification strategy effectively identified faults and coexisting imbalance.
• Analysis revealed the impact of different feature groups and measurement axes on performance.
• Validation on a new dataset confirmed the framework's robustness and generalization.

Conclusions:

• The proposed framework effectively diagnoses faults in rotating machinery.
• Integrating data-driven and physics-informed features significantly enhances diagnostic accuracy.
• The framework demonstrates strong generalization capabilities for real-world applications.