Sang-Jae Yeo, Zoo-Ho Son, Jee-Yeon Jeon, Hyung-Taek Kim, Journal of Internet Computing and Services, Vol. 26, No. 2, pp. 43-50, Apr. 2025
10.7472/jksii.2025.26.2.43, Full Text:  HTML
Keywords: Anomaly Detection, Machine Learning, Current signal, Induction motor, Statistical distance

Abstract

With the growing interest in electric propulsion systems for naval ships, the need for reliable anomaly detection of rotating machinery has also increased. This study aims to improve the accuracy of induction motor anomaly detection based on current signals by introducing a two-stage detection algorithm. The first stage employs an Isolation Forest model using 36 features extracted from three-phase current signals, while the second stage integrates statistical analysis of the cumulative distribution of detection results. To validate the proposed model, experiments were conducted using a test motor to simulate nine types of anomalies, including bearing defects, mechanical looseness, rotor faults, and stator faults. Compared to the first-stage model, which achieved a recall of 54% for anomalous data, the second-stage model incorporating statistical analysis demonstrated a recall of 100%, confirming the effectiveness of the proposed algorithm.

Statistics
Cite this article