FAILURE DETECTION IN THREE-PHASE MOTORS USING DEEP LEARNING IN VIBRATION ANALYSIS

Abstract

Induction motors play a fundamental role in the industrial sector and require proper monitoring to avoid unscheduled production downtime. Vibration monitoring stands out in the condition diagnosis of these motors due to its effectiveness in detecting faults, especially in bearings. This work proposes the use of Deep Learning techniques to automate bearing fault detection through a comparative analysis between Long Short Memory (LSTM) and Convolutional Neural Network (CNN) models. Using the University of Cincinnati's Intelligent Maintenance System (IMS) dataset, the LSTM model was trained with statistical descriptors extracted from the vibration signals, while the CNN operated directly on the raw data. The results demonstrate the superiority of the LSTM model, which achieved 98% accuracy, compared to the CNN, which achieved 86%.

DOI: 10.56238/edimpacto2025.041-003