Infrared Thermography for the Diagnosis of Incipient Faults in High-Efficiency Motors

The high percentage of industrial production and use of induction motors to meet the needs of the population demand has made these machines work for long periods, exposed to faults and causing delays in manufacturing production. This is why the need arises to evaluate the internal faults of these driving machines. This study aims to detect incipient failures of induction motors using infrared thermography images through the Google Teachable Machine extension. This trains a neural network with four classes for analysis: bearings and rotor bars in good and bad condition, respectively. The trained model is exported to an MPU in .h5 and .txt format to start the system. There is a controlled environment cabinet where the parts are placed for analysis and a one-way communication architecture so that an operator can visualize the neural network’s output on a dashboard and identify the anomaly’s exact location. The results show the excellent accuracy of the neural network to diagnose faults, both when using the confusion matrix of the system and the random control developed in the plant, showing that the epoch losses are very close to 0, which indicates that the learning level was almost perfect.