Acoustic Diagnosis of Stator Inter‐Turn Short Circuits in Induction Motors Using MEMS Microphones and Group‐Aware Validation

ABSTRACT

Stator inter‐turn short circuits (ITSCs) can evolve from local insulation damage to severe winding failure, but their acoustic signatures remain less established than current‐ and vibration‐based indicators. This study investigates whether compact acoustic descriptors can support ITSC diagnosis whereas accounting for dependence among segmented recordings, microphone variability and motor‐domain shift. Five analogue microelectromechanical systems (MEMS) microphones sampled at were placed around a three‐phase induction motor tested under healthy operation and nominal ITSC levels of , , and , across 10 torque levels and balanced or modified phase‐voltage conditions. Seventy‐one statistical, temporal, spectral and cepstral descriptors were extracted from non‐overlapping windows and evaluated with six classifiers using acquisition‐file group‐aware validation. The best binary and five‐class models achieved and accuracy, respectively. Calibration‐free leave‐one‐sensor‐out multiclass macro‐F1 was , increasing to after healthy‐reference calibration. Mixed‐domain training with a WEG W22 Plus motor achieved macro‐F1, whereas leave‐one‐motor‐domain‐out performance fell to . Feature extraction required per window and microphone. The findings support low‐cost, non‐contact screening within represented sensing and machine domains. Deployment requires sensor calibration, industrial‐noise validation, multi‐machine transfer assessment, pulse‐width modulation and variable‐frequency drive testing and target‐hardware qualification.