Image-Based Detection of Short-Circuit Faults in Drive-Fed Induction Motors: A Comparative Study of Current and Vibration Signatures

Abstract

Detecting stator winding Short-Circuit (SC) faults in Induction Motors (IMs) is essential to prevent severe damage at early stages of fault development. This work investigates SC fault detection in a laboratory IM under various load conditions, considering both direct grid connection and operation supplied by Variable Frequency Drives (VFDs) with different manufacturers, nominal frequencies, and control strategies. Conventional detection methods based on stator current spectral analysis suffer from significant limitations in VFD-fed drives due to dense converter-induced harmonics and speed-dependent shifting of fault components. To address these challenges, a frequency-adaptive, image-based diagnostic framework is proposed, replacing fixed-frequency spectral analysis with visual pattern recognition. In addition, a novel analytical relationship is introduced that links the ellipticity of the Park’s vector directly to the negative-sequence current component, providing a theoretical explanation for the transition from circular trajectories under healthy conditions to elliptical patterns under stator inter-turn fault conditions. A Residual Network (ResNet) classifier is trained on images generated from Park’s Vector Analysis (PVA) of stator currents and Orbit Pattern Analysis (OPA) of vibration signals, enabling a comparative assessment of current and vibration fault signatures under identical VFD conditions. The proposed method is experimentally validated using incipient SC faults under varying load levels, drive configurations, and control strategies. The results demonstrate high classification performance and highlight the effectiveness of morphology-based diagnosis for early-stage SC fault detection in VFD-fed IMs.