Stratégies de contrôle avancées pour systèmes de production sous contraintes environnementales : politiques intégrées et apprentissage par renforcement

Abstract

This research develops and validates control strategies for unreliable manufacturing systems operating under environmental constraints. The proposed approach combines two distinct methodologies: dynamic stochastic programming combined with numerical approach for hybrid configurations, and artificial intelligence through reinforcement learning for simple systems. The first contribution focuses on controlling hybrid production systems integrating manufacturing and remanufacturing under environmental constraints. A dynamic environmental hedging point (DEHPP) control policy is developed to coordinate production activities while respecting imposed environmental limits. This strategy simultaneously integrates economic and environmental costs while managing prioritization and switching between machines according to system state and GHG emission constraints. Comparative study reveals substantial gains compared to relevant policies from literature, both in economic efficiency and environmental compliance. The second contribution explores the potential of reinforcement learning to control basic M1P1 systems in a stochastic environment. The Proximal Policy Optimization (PPO) algorithm is deployed with methodical tuning of learning parameters. Validation is conducted progressively, starting with a basic configuration before incorporating environmental requirements. Qualitative comparison demonstrates the relevance of AI techniques and establishes a methodological foundation for more sophisticated future applications. The synthesis of this work emphasizes the necessity of integrating environmental concerns into manufacturing optimization and reveals the synergy between classical methodologies and technological innovations. These developments offer industrial managers powerful instruments to harmonize economic and ecological objectives while mastering operational uncertainties.

Date	19 Dec 2025
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Mustapha Ouhimmou (Supervisor) & Ali Gharbi (Co-supervisor)