A Machine Learning Based Framework for PdM

Abstract

The need for adaptation led the industry to evolve into a new revolution, where connectivity, amount of data, new devices, stock reduction, personalization and production control gave rise to Industry 4.0. Predictive maintenance is based on historical data, models and knowledge of the domain in order to predict trends, patterns of behavior and correlations by statistical models or Machine Learning to predict pending failures in advance. This paper presents a review of most applied machine learning techniques, comparing different authors’ approaches used in predictive maintenance. Also, a conceptual machine learning framework is proposed to tackle various predictive maintenance challenges such as failure forecast, anomaly detection and Remaining Useful Life prediction.