Deep Transfer Learning-Based Broken Rotor Fault Diagnosis for Induction Motors

Abstract

Induction motors, vital in industrial systems, are prone to electrical faults like broken rotor bars. Early detection is crucial to prevent failures. This study uses an IEEE dataset of motor signals to classify faults under varying loads via time-frequency analysis (STFT spectrograms) and deep learning. Three pre-trained CNNs (ResNet50, AlexNet, and VGG16) were tested, with ResNet50 achieving the highest accuracy (99.73%). Evaluated using accuracy/loss curves, confusion matrices, and precision/recall metrics, ResNet50 proved most robust for early fault detection, ensuring reliability across operational conditions.