Machine Learning Based Fault Detection in Induction Motor using Thermal Imaging

In thermal imaging research, the majority of the work involves assessing external faults and detecting failures in equipment's outer parts. This work presents a condition monitoring system for assessing internal faults to detect failures in an induction motor using infrared thermal imaging. A dataset consisting of 540 infrared thermal images of an induction motor, which includes five faulty conditions and one healthy condition, is considered for analysis. The histogram and textural features are extracted from the thermal images after the data is acquired, processed, and analyzed. Next, images are classified into hot and cold, and the region of interest is located by applying a sliding window technique with an auto-thresholding function and ground truth. Machine learning classifiers such as decision trees and random forest and ensemble learners such as Ada boost and gradient boosting trees are used to detect and classify faults in both hot and cold conditions. To examine the performance of each classifier, evaluation metrics are used.