Isolation Forest and Local Outlier Factor-Based Anomaly Detection on an Induction Machine

This paper is aimed at alleviating the data imbalance challenges with data-driven condition monitoring on rotating electrical machines, more specifically in the case of an experimental wound rotor induction generator. The investigation is based on the use of unsupervised learning for the purpose of anomaly detection. The combined stator voltages and rotor current signatures are used in the development of isolation forest and local outlier factor-based anomaly detection models. These models are evaluated to determine the suitability and performance of these techniques for anomaly detection. This investigation is based on the brush anomaly, stator, and rotor inter-turn anomalies at no load. The results indicate that the local outlier factor outperforms the performance of the isolation forest in detecting the investigated faults on the machine.