Track: Machine Learning

Abstract

With the increasing focus on process automation and the advent of Industry 4.0, predictive maintenance strategies (PdM) have gained significant attention. These strategies enable maintenance actions to be performed when necessary, reducing downtime costs and enhancing the efficiency and availability of production machinery. The widespread adoption of advanced analytical tools and Machine Learning (ML) technologies has facilitated continuous monitoring of machine operation through the development of PdM models.

In this study, we propose an ML-based approach to develop a prediction model specifically for the manufacturing sector. The model employs supervised learning, utilizing the sliding window method and the Support Vector Machine algorithm (SVM), reaching an accuracy of 99.7%. By leveraging artificial intelligence for predictive maintenance, our algorithm enables the monitoring of a centrifugal pump using acoustic and vibration parameters. This enables the identification and prediction of five distinct operating conditions of the pump.