Thermal Image Analysis and Detection of 3-Phase Induction Motors Using Machine Learning

The heat of the induction motor results in power loss, lower efficiency, and shorter life. For inspection, it is often done by touching or measuring the temperature, which can easily have measurement errors. Therefore, this research presents the analysis and inspection of thermal images of induction motors using machine learning techniques that are accurate and correct in detecting abnormal heat in the motor. The thermal image of the motor is trained and processed through the CiRA CORE program to classify and learn the motor heat in normal, warning, and problem conditions. The training of thermal images in the V4-tiny and V5-tiny motor models, which are CNN models, is compared to find the accuracy and loss. The analysis and processing are done by sending notifications via TGNotify and connecting ESP32 to send siren notifications. The experimental results show that the accuracy of the V5-tiny model is 0.88, which is higher than the V4-tiny model, which is 0.85. On the other hand, the loss of the V4-tiny model is 15%, which is higher than the V5-tiny model, which is 12%. V5-tiny has the highest accuracy and stability because it is trained with high-resolution images. For notification via the Telegram app and siren, the notification can be sent accurately, and the notification transmission time is not more than 3 seconds. In the application of machine learning to analyze and classify the thermal image of the motor, it can prevent motor damage and help in deciding on maintenance.​​​​​​​