A Comparative Analysis of Deep Learning Models for Rotor Bar Faults Detection in Induction Motors Based on Scalograms Derived from Stator Current and Vibration Signals

In Industry 4.0, deep learning models applied to sensor data analysis in machinery play a crucial role in shaping predictive maintenance strategies. This paper presents two deep Convolutional Neural Networks (CNN) that significantly differ in their architectures, GoogLeNet and RESNET50, in detecting and diagnosing rotor broken bars in three-phase induction motors. The Scalogram images derived from experimental stator current and vibration signals were utilized to assess and contrast their efficacy against the proposed models. The outcomes obtained validate the efficiency and effectiveness of the proposed strategy. The two models perform better when analyzing vibration signal scalograms than stator current scalograms. Such results might suggest that vibration signals carry more diagnostic information about broken bar faults than stator current signals.