Robust fault classification in rotary machines using recurrence quantification analysis features for machine learning...

Area of Science:

• Engineering
• Data Science
• Machine Learning

Background:

• Fault detection in rotating machinery is crucial for industrial reliability and minimizing downtime.
• Conventional feature extraction methods struggle in noisy, real-world operational conditions.

Purpose of the Study:

• To explore machine learning classifiers (RF, GB, XGB, LGBM) using statistical and Recurrence Quantification Analysis (RQA) features for fault detection.
• To evaluate the robustness of these methods in both noise-free and noisy industrial environments.

Main Methods:

• Benchmarking RQA features against statistical features on synthetic nonlinear systems (Lorenz, Rössler, Hénon, Duffing).
• Applying selected machine learning classifiers to experimental rotary machine data under varying noise conditions (Gaussian, Brownian, impulsive).

Main Results:

• In noise-free conditions, statistical features achieved 99% accuracy, while RQA features yielded 93%.
• With added Gaussian noise (15%-75%), statistical features accuracy dropped to 85%, whereas RQA features remained above 94%.
• RQA features demonstrated consistent superiority across different noise types.

Conclusions:

• Recurrence Quantification Analysis (RQA) features are superior to statistical features for fault detection in noisy industrial environments.
• RQA's ability to capture long-term dynamics makes it a more reliable tool for industrial fault detection.