Scientific context and challengesIndustry 4.0 relies on increasingly reactive and flexible control of production systems, made possible by the availabilityof real-time data and by the development of new decision-support tools for planning, scheduling, and maintenance.Despite these advances, the effective integration of such tools into traditional production management approachesremains a major scientific challenge.The ANR POMPIER project aims to address these challenges by proposing a methodological framework to improve theconsistency and robustness of production plans, in close connection with predictive and prescriptive maintenanceissues. The objective is to better account for the robustness of planning and scheduling decisions with respect to:• discrepancies between tactical plans and operational reality,• strong interdependencies between production and maintenance,• uncertainties affecting processes, whether random or epistemic in nature.In particular, the project relies on the identification and modelling of prescriptive data derived from the relationshipbetween system usage and degradation, in order to promote better synchronization across decision-making levels.Scientific problem statementProduction planning and scheduling are traditionally addressed as optimization problems. The consideration ofuncertainty has been studied for several decades through approaches such as stochastic optimization and robustoptimization. However, these approaches largely rely on the assumption that uncertainty models and theirparameters are known with sufficient accuracy.In practice, two major categories of uncertainty must be distinguished:• Aleatory (or stochastic) uncertainties, related to the intrinsic variability of processes;• Epistemic uncertainties, resulting from lack of knowledge, incomplete data, or imperfect models.While aleatory uncertainties have been widely studied in scheduling, epistemic uncertainties remain comparativelyunderexplored, particularly in models coupling scheduling and maintenance. This postdoctoral project is situatedwithin this perspective and aims to contribute to the evaluation of scheduling robustness with respect to epistemicuncertainties, by relaxing the assumption of perfectly known uncertainty models.