DTW-Based Anomaly Detection and Fault Categorization in Rotating Machinery Using Multi-Modal Sensor Data

Abstract

In industrial rotating machinery, early detection and classification of mechanical faults are critical for ensuring operational safety and minimizing unplanned downtime. This study presents a data-driven condition monitoring framework utilizing multi-modal sensor data to detect rotor imbalance and infer likely fault categories in a radial fan system. The proposed approach leverages Dynamic Time Warping (DTW) to quantify the deviation of current operating conditions from a healthy baseline, using a MinMax-scaled DTW distance to generate an interpretable anomaly score ranging from 0 to 100. By mapping score intervals to standardized fault categories, the system not only detects anomalies but also provides fault interpretability grounded in vibration analysis theory. Experimental validation is conducted on a dataset containing a known imbalance fault, and the results demonstrate that the proposed method accurately identifies the fault type while showing strong temporal alignment with existing fault metrics. The integration of unsupervised anomaly score estimation with fault severity classification charts offers a practical and scalable solution for predictive maintenance in real-world industrial settings. © 2025 IEEE.