Detección de fallas en motores de inducción mediante características basadas en texturas e inteligencia artificial

Resumen

Induction motors are fundamental components in the industry, but their continuous operation makes them susceptible to internal faults that can lead to unscheduled downtime and high repair costs. Timely fault detection is crucial for preventive maintenance. This paper proposes the development of an automated fault classification system by analyzing texture features extracted from thermal images. Unlike proposals that rely on deep neural networks, this project utilizes classic machine learning models, allowing for an implementation with lower computational cost, making it suitable for embedded systems. A K-Nearest Neighbors (KNN) model achieved the highest accuracy at 84.1%, demonstrating that this approach is a fast, non-invasive, and precise solution for a preventive diagnosis of motor failures. The results position this method as an effective and computationally efficient alternative to more complex models.