Detección de fallas en motores eléctricos utilizando termografías y redes neuronales

Abstract

The project addresses the detection of faults in electric motors through thermography and neural networks. It involves the study of concepts about alternating and direct current motors, their possible electrical and mechanical failures, the use of infrared thermography, the design and application of artificial and convolutional neural networks. The Cat S61 smartphone is used to generate thermographic reports classifying failures into severity levels. The neural network designed in Python, which includes 83 thermographic images, 3 convolutional layers, kernel filters, activations, and Softmax classifier, achieves a training accuracy of 59% and a validation accuracy of 71%. It uses TensorFlow's Keras interface and a Tkinter graphical interface to facilitate the use of the application. Tests show that the network is especially effective at detecting faults of mild and critical severity. This approach integrates advanced technologies to improve the diagnosis and maintenance of electric motors.