Recently, the predictive maintenance (PdM) concept has received increasing attention in industrial practices and academic research. The potential opportunities for utilizing real-time data provided by the Industry 4.0 principles and technologies are often exploited for equipment health monitoring and remaining useful life (RUL) estimation. Thereby, most studies focus on the prediction of failures, which is important in itself, while the prescription of decisions for maintenance activities is much less addressed in the literature. Furthermore, it is often neglected that in the real world, all the equipment or components cannot be maintained only through the PdM strategy and planning of this type of maintenance intervention could not be separated from other maintenance strategies and activities such as preventive maintenance (PvM).
The aim of this thesis is to propose new approaches and methods for PdM and PvM through data science and mathematical optimization. The study aims to address several key questions, such as the feasibility of estimating the health state and/or RUL of equipment/components, and the methods to use this data and information in decision-making and maintenance planning, while considering their links to other activities. In this regard, a comprehensive Predictive Maintenance Management System (PdMMS) approach is proposed. This approach covers overall maintenance strategies and involves several interconnected steps, from descriptive and diagnostic analysis to prescriptive decision-making. These steps incorporate the criticality analysis of equipment and/or components, identification of appropriate maintenance strategies for each equipment and/or component, identification and acquisition of required data and information for each strategy, maintenance monitoring system for PvM and corrective strategies, equipment health monitoring and RUL estimation for PdM strategy, and finally, maintenance planning of overall maintenance interventions. Based on this approach, various appropriate methods, algorithms, and applications were developed and applied to different use cases at the Fiat Powertrain Technologies Bourbon-Lancy (FPT-BLY) plant.
Initially, a new approach for identifying maintenance strategies is proposed. This approach relies on several methods such as multi-criteria decision-making (MCDM), ABC analysis, sensitivity analysis, and optimal frequency identification. It has been applied to one of the complex Computer Numerical Control (CNC) equipment, and the proposed approach has successfully identified critical equipment and components, resulting in a significant reduction in emergency purchases. Concerning maintenance monitoring, the most relevant maintenance indicators have been identified, and appropriate visualization dashboards have been proposed to monitor maintenance performance and facilitate the identification of improvement actions. Regarding equipment health monitoring and RUL estimation, a health indicator (HI) method and a new dynamic algorithm are proposed and applied to a use case of a conveyer chain painting system. The results demonstrate the possibility of dynamically estimating the health state and RUL based on real-time data. In the same case, a problem of simultaneous PdM and PvM planning is also studied. In this context, a mathematical optimization model is proposed to minimize overall costs, including direct (intervention costs) and indirect (expected failure risk costs, unused life losses), by considering the opportunistic grouping of maintenance interventions. This study indicated that RUL information could be integrated into the comprehensive maintenance planning system to identify optimal planning. Moreover, to validate the proposed method, an inclusive sensitivity analysis is provided, and the obtained results indicate that considering the mentioned aspects could significantly impact maintenance planning and decrease overall maintenance costs in the mid/long term. Finally, the thesis conclusion highlights the main implementation challenges of the proposed PdMMS approach in other companies, along with managerial insights and research perspectives.
Key words: Predictive maintenance, RUL estimation, Maintenance planning, real-time data, data analytics, mathematical optimization, Industry 4.0.