ResFaultyMan: An intelligent fault detection predictive model in power electronics systems using unsupervised learning isolation forest

Abstract

Intelligent fault detection considered as a paramount importance in Power Electronics Systems (PELS) to ensure operational reliability along with rising complexities and critical application demands. However, most of the developed methods in real-world scenarios can have better detection, and accurate diagnosis. In this regard, ResFaultyMan, a novel unsupervised isolation forest-based model, is presented in this paper, for real-world fault/anomaly detection in PELS. Capitalizing on the dynamics of faults, ResFaultyMan utilizes a tree-based structure for effective anomaly isolation, demonstrating adaptability to diverse fault scenarios. The test bench, comprising a load, Triac switch, resistor, voltage source, and Pyboard microcontroller, provides a dynamic setting for performance evaluation. The integration of a Pyboard microcontroller and a Python-to-Python interface facilitates fast data transfer and sampling, enhancing the efficiency of ResFaultyMan in real-time fault detection scenarios. Comparative analysis with OneClassSVM and LocalOutlierFactor, utilizing Key Performance Indicators (KPIs) of Accuracy, Precision, and Recall, as well as F1 Score, manifest ResFaultyMan's fault detection capabilities for fault detection in PELSs, and its performance in the related applications.