Categorization of Asynchronous Motor Situations in Infrared Images: Analyses with ResNet50

Abstract

Asynchronous or induction motors are frequently preferred in industrial applications concerning their cheap supply, strength and easy maintenance. However, the fault recognition of these motors constitutes a comprehensive examination with direct interference. As a consequence, the images obtained from infrared cameras and their analyses gain importance to remotely detect the situation of motors. For this purpose, we handle a transfer learning approach named ResNet50 to categorize 11 different situations of asynchronous motors in infrared camera images. For performance assessment, hyper-parameters of ResNet50 are examined to maximize the success to be achieved. In experiments, two test methods (70%-30% training-test split and 80%-20% training-test split) are utilized to objectively evaluate the parameter adjustments and to obviously reveal the effect of training samples. As a result, it's proven that ResNet50 can achieve 100% classification accuracy for categorization of induction motor situations in experiments with both test methods. © 2022 IEEE.