Deep Learning Approaches and Motor Current Signature Analysis in Detection of Broken Rotor Bar Faults

Abstract

Induction motors are preferred in industrial applications due to their simple and robust structure, cost-effectiveness, self-starting capability, high efficiency, and reliability. However, faults like broken rotor bars occasionally encountered in these motors can lead to reduced performance and increased operating costs. Deep learning models are increasingly being used for the early detection of such faults. These models can recognize complex patterns in motor data to identify potential faults in advance, allowing for timely intervention, extending motor life, and ensuring production continuity. In this study, the diagnosis of broken rotor bars in induction motors was performed using four different deep learning models. Binary classification was conducted based on images obtained from current signals using a pre-existing dataset. The study achieved over 90% accuracy, thereby proving the effectiveness of deep learning models on induction motors.