Architectural Search-Driven Hybrid Pipelines for Diagnosing Incipient Stator Inter-Turn Faults in Induction Motor Under Power Quality Disturbances

Line-fed induction motors (IMs) are essential in transportation systems, powering traction systems, hoists, elevators, escalators, and conveyor belts. Stator inter-turn faults (SITFs) pose a significant threat, potentially causing severe motor damage and operational failures. Diagnosing SITFs is further complicated under practical conditions where IMs are subjected to power quality (PQ) disturbances, deviating from ideal balanced sine wave operation. This paper presents two hybrid diagnostic machine learning (ML) pipelines, developed through architectural searches, for fault diagnosis in IMs operating under PQ disturbances. The first pipeline predicts the severity of SITFs using stator voltage and current signals, while the second estimates the type of PQ event based solely on stator voltages. A dedicated experimental test-bed was developed to emulate various fault severities under various PQ disturbances and load conditions. The diagnostics pipelines which achieved an <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">R</i><sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score of 0.9699 on training and 0.9533 on testing for severity estimation. The threshold based fault classification achieved accuracies of 96.92% (training) and 93.42% (testing). PQ event classification accuracies reached 98.87% and 96.71%, respectively. Additionally, comparisons with models trained on raw time-series data were conducted to demonstrate the effectiveness of the proposed diagnostic approach. This research uniquely integrates domain knowledge with ML techniques to diagnose incipient SITFs in IMs under PQ disturbances—an area largely unexplored in existing literature.