Detección de Fallas en Motores de Inducción a Través de Descriptores Estadísticos Utilizando Señales de Vibración

Resumen

Operational reliability in modern industry largely depends on the continuous operation of induction motors, as they power most essential production processes. A method is proposed for early fault diagnosis that extracts statistical descriptors from vibration signals, second to fourth-order centered moments, RMS, and crest/shape factors, and classifies motor states with a shallow decision tree. Tri-axial vibration data from a three-phase induction motor (1.1 kW, 220/380 V, 5 A, 2800 RPM, 50 Hz) were recorded under different operating conditions, segmented for uniformity, and transformed into the described features. Several supervised models were evaluated on a stratified 90/10 train–test split; the decision tree achieved the best accuracy while remaining easy to interpret. The findings indicate that simple statistical features coupled with an interpretable classifier can deliver accurate diagnosis with modest computational cost, supporting predictive-maintenance adoption in resource-constrained environments.