Toward a Predictive Maintenance: Implementing an Innovative Maintenance System for Manufacturing Production Lines Based on New Technologies

The rise of “big data” has been profoundly transformed the landscape of the manufacturing sector, putting pressure
on companies’ competitiveness and making the effective utilization of this vast data a critical imperative for
efficient maintenance. Therefore, predictive maintenance plays an essential role in meeting these challenges,
anticipating maintenance needs and enhancing equipment durability. This paper introduces a new dimension that
considers not only the monitoring and prognosis of component failures but also the formulation of optimal and
efficient decisions for maintenance planning and execution. The objective of the developed system is to include
real-time data acquisition from various sensors through an IoT-based system, proactively detecting potential
failures before they escalate into major issues, estimating the remaining useful life (RUL) of equipment using Deep
Learning models, and ultimately making optimal and efficient decisions for maintenance planning and execution.
This minimizes intervention costs and increases the productivity and efficiency of machines and systems,
enhancing the company’s competitiveness, among other benefits. In this study, a real-world application was
conducted to evaluate the proposed methodology, and several algorithms and AI techniques were introduced to
suggest economical replacement methods. Based on the collected data, a new solution for prognosis is proposed
using a regression model with the LSTM algorithm. The accuracy achieved by this model is 88.68%. The outdoor
validation results obtained so far suggest that the developed method could be an efficient solution for industrial
companies. As the proposed system performs prognosis data for each component, it develops an optimal
maintenance scenario based on the ANN algorithm to optimize the maintenance plan and devise the most costeffective and reliable strategies for equipment upkeep. The test loss and accuracy obtained by the decision model
are respectively 0.0347 and 0.98. This approach opens up new perspectives for more efficient industrial
maintenance management in the era of industry 4.0.