Modelling the predictive reliability of rotating machines using Artificial Intelligence.

Abstract

In the context of Industry 5.0, predictive maintenance of rotating machinery is essential for improving operational efficiency, enhancing forecast reliability, and minimising unplanned downtime. This study presents an innovative approach that uses the Audio Spectrogram Transformer (AST) to predict the remaining useful life (RUL) of rotating machinery, leveraging a vibration signal dataset from a three-phase electric motor in the electrical engineering workshop at the IUT of Douala, called Vibration DATA IUT. By transforming vibration signals into time-series spectrograms, the AST model enables the detection of complex temporal patterns indicative of wear and degradation. Our methodology demonstrates superior predictive capabilities, achieving 98.9% accuracy and an impressive root mean square error (RMSE) of 1.2 hours in RUL estimation. These results highlight the effectiveness of integrating deep learning techniques with time-frequency domain analysis, offering an innovative framework for reliable failure prediction. These advances not only advance the state of the art in vibration analysis but also contribute significantly to predictive maintenance strategies in industrial applications, paving the way for increased safety and productivity.