Predictive maintenance based on machine learning: a systematic literature review and perspectives in industry 4.0

Abstract

Within the framework of Industry 4.0, predictive maintenance (PdM) emerges as an essential component to maximize asset availability and ensure operational excellence. The explosion of sensor data and the increasing complexity of production systems require advanced analytical approaches. This systematic review examines literature where machine learning (ML) techniques are applied to PdM in sectors such as wind energy, railroads, manufacturing and infrastructure. Algorithms used, deployment scenarios, performance metrics and reported results are analyzed. Neural networks, support vector machines and ensemble methods dominate fault detection, remaining life estimation and intervention planning. Studies show improvements in prognostic accuracy, downtime reduction and cost reduction. It is concluded that ML acts as an enabler of data-driven PdM, enhancing reliability and sustainability, and lines of research for future intelligent maintenance strategies are identified.