Digital Twin and Deep Learning-Based Approach for Detecting Faults in Induction Motors

Abstract

In this study, a digital twin and deep learning-based method is proposed to detect potential faults in induction motors. The developed methodology utilized the three-phase stator current data obtained from the digital twin, subjected to noise reduction, and the PVM signal was obtained using the Park Vector Modulation (PVM) approach. Then, upper and lower envelope signals were obtained by applying an envelope to the resulting PVM signal. The resulting upper envelope signal was then converted to RGB images with the Recurrence Plot (RP) technique. These images were classified using the MobileNetv2 model. The proposed method offers significant potential in detecting induction motor faults accurately and effectively. The results show that this deep learning-based approach can provide high accuracy rates in early detection of induction motor faults. This study makes a significant contribution to the field of maintenance and fault detection of induction motors, revealing an innovative approach that allows early detection of motor faults in industrial applications.