Application of IEEE Std. 1459-2010 and Hjorth Descriptors for the Identification of Stator Faults in Induction Motors

Abstract

Stator interturn short-circuit (ITSC) faults represent a significant fraction of failures in three-phase induction motors (TIMs), with industrial studies reporting occurrences of up to 36%. Current-based diagnostic techniques are widely employed; however, their effectiveness degrades under inherent supply asymmetries, manufacturing tolerances, and moderate voltage imbalances, which introduce negative-sequence and harmonic components that mask incipient fault signatures and can produce current components similar to those generated by internal electromagnetic asymmetries. Diagnostic capability is investigated using two families of features extracted from stator electrical signals: power quantities defined by IEEE Std. 1459-2010 and Hjorth time-domain descriptors. These feature families correspond to distinct physical manifestations observable in the measured currents: macroscopic redistribution of electrical power under nonsinusoidal and unbalanced operating conditions, and local variations in waveform geometry associated with circulating currents and early flux asymmetries. Model performance is evaluated through stratified $k$-fold cross-validation using five supervised classifiers applied to a labeled laboratory dataset comprising multiple ITSC severities (healthy, 1%, 3%, 5%, and 10%), controlled voltage unbalance, multiple mechanical loading conditions, and supply-frequency variation. Hjorth parameters present higher sensitivity to low-severity faults and achieve classification accuracies above 99%, whereas IEEE Std. 1459-2010 quantities yield accuracies between 78% and 95%, reflecting reduced sensitivity to weak distortions but intrinsically phase-invariant behavior. A reduced hybrid feature set combining selected IEEE Std. 1459-2010 descriptors with the Hjorth parameters activity and mobility reduces phase dependence and produces classification accuracies exceeding 99.5% across all evaluated classifiers. Correlation analysis and feature-importance rankings indicate limited redundancy between descriptor groups and reveal complementary diagnostic information associated with electrical power redistribution and waveform-geometry variations. The obtained results indicate that early-stage ITSC faults are identifiable using a compact descriptor space derived from physically interpretable electrical quantities, without dependence on high-dimensional time-frequency representations or deep learning models.