Optimisation des politiques de contrôle pour les systèmes de production non fiables par l'analyse des données et le machine learning

Abstract

This research focuses on the development of an approach combining simulation, machine learning, and optimization for determining the control parameters of manufacturing and manufacturing–remanufacturing systems subject to various sources of uncertainty, including machine failures, demand variability, and product return variability. The proposed approach relies on the use of simulation-generated data to train machine learning models integrated into the optimization process. First, the approach is developed for an unreliable M1P1 manufacturing system (one machine, one product) operating under constant demand. A database construction strategy is proposed and used for training the machine learning models. The approach is then extended to environments characterized by variable demand. The results show that the developed model is able to accurately predict both the total system cost and the associated control parameters, providing results comparable to those obtained through simulation-based optimization. A further extension is carried out in the context of seasonal demand variability in order to evaluate the benefits of dynamically adapting the control parameters according to demand evolution. The results show that this dynamic approach yields lower costs than a static policy based on a fixed hedging point, with performance improvements becoming more significant as demand variability increases. In a second stage, the approach is applied to a hybrid manufacturing–remanufacturing system subject to variable demand, variable product returns, and random machine failures and repairs. The developed models are used to predict the total system cost and determine the control parameters associated with manufacturing and remanufacturing activities. The results show that the proposed approach consistently outperforms simulation-based optimization relying on fixed control thresholds and leads to improved economic performance of the system. Overall, the results demonstrate that machine learning constitutes a relevant tool for determining control parameters in production systems operating under uncertainty. The findings confirm the ability of the proposed approach to efficiently address manufacturing and manufacturing–remanufacturing systems characterized by increasing levels of complexity and uncertainty.

Date	22 Jul 2026
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Ali Gharbi (Supervisor)