Contrôle des procédés verriers et conception de jumeaux numériques par une approche innovante issue de méthodes d'apprentissage profond et des données d'exploitation // AI-Driven Digital Twins for Energy-Efficient Glass Furnaces Reinforcement Learning, Pr

..
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

The PhD will focus on the development of artificial intelligence approaches coupled with physics-based models to assist the control and optimization of industrial glass furnace processes within a digital twin framework.
The research will combine simulation-based learning and industrial data-driven modeling to enable intelligent assistance for furnace operation and process optimization. The work will be structured around two complementary axes:
1. Reinforcement Learning for Process Control and Optimization
The candidate will develop reinforcement learning approaches to optimize the operation of industrial glass furnaces. These methods will rely on high-fidelity numerical simulations and/or surrogate models to train AI agents capable of learning optimal control strategies for furnace operation. The objective is to explore adaptive control policies that improve energy efficiency, stabilize furnace behavior, and maintain product quality under varying operating conditions. By interacting with virtual representations of the furnace, reinforcement learning agents will learn to manage complex thermo-fluid processes and support future autonomous or assisted process control strategies.
2. Data-Driven Modeling and Expert Systems from Industrial Production Data
The second axis will focus on the exploitation of industrial production data to develop data-driven models capable of capturing furnace behavior under real operating conditions. These models will leverage machine learning techniques to identify patterns, detect anomalies, and extract operational knowledge from historical datasets.
The combination of simulation-driven learning and industrial data analysis will enable the development of intelligent expert systems that assist operators in decision-making and process supervision.
Together, these two components will contribute to the development of AI-powered digital twins capable of supporting monitoring, prediction, and optimization of glass furnace operation in industrial environments.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Début de la thèse : 01/10/2026
WEB : https://www.cemef.minesparis.psl.eu/wp-content/uploads/2026/05/PhD_TwinHeat-en.pdf

Autre financement public

ANR Financement d'Agences de financement de la recherche

Mines Paris-PSL

364 SFA - Sciences Fondamentales et Appliquées

..
Master's degree completed or in progress in fluid mechanics, thermal engineering, applied mathematics, computational science, or artificial intelligence, with a strong academic record. Key skills: • heat transfer and fluid mechanics • numerical methods for nonlinear partial differential equations • programming in Python and/or C++ • interest in artificial intelligence and industrial process modeling • ability to work in a multidisciplinary research environment • strong English communication skills Additional assets include experience with machine learning, reinforcement learning, high-performance computing, or industrial data analysis.