Optimizing Induction Motor Fault Detection with Transfer Learning: A Comparative Analysis of Deep Learning Models

With the advancements of Industry 4.0, condition monitoring maintenance has become essential for preventing equipment failures and operational disruptions. Motor Current Signature Analysis (MCSA) is commonly utilized for condition monitoring to detect and diagnose various faults in Induction Motors (IMs). Despite its popularity, there is limited research comparing deep learning models for Induction Motor fault detection and classification with traditional approaches. This study explores the detection and classification of Induction Motor faults using three Transfer Learning (TL) models: InceptionV3, ResNet152, and VGG19. The research began by modeling a Squirrel Cage induction motor in MATLAB to simulate healthy, single- phasing, and double-phasing conditions, capturing time- domain stator current signatures (current spectrum) to identify fault characteristics. The data were then used to assess the effectiveness of the TL models in detecting and classifying motor faults. Around 500 datasets were created from these simulated conditions, labeled accordingly, and used to train and validate the TL models, each incorporating additional convolutional layers to enhance performance. Model evaluation utilized metrics such as the multiclass confusion matrix, precision, recall, and F1-score across various fault scenarios. Results indicate that stator current signatures can effectively reveal individual faults, with ResNet152 outperforming the other models in classification accuracy. These findings highlight that applying transfer learning techniques with a limited amount of current signature data can support predictive maintenance in industrial settings, potentially reducing costly equipment shutdowns and disruptions in production.