Reinforcement Learning for the Inverse Design of MOF-Integrated Composite Pressure Vessels for Hydrogen Storage

Supervisors: Christoforos Rekatsinas, George Giannakopoulos, Panagiotis Krokidas

Description:

This thesis will investigate the use of multi agent reinforcement learning for the inverse design of MOF-integrated composite pressure vessels for hydrogen storage. Inspired by recent work where airfoil optimization was formulated as a Markov Decision Process, the thesis will transfer the same logic to pressure-vessel design. Instead of modifying aerodynamic shapes, the reinforcement learning agent will optimize design parameters such as vessel geometry, wall thickness, composite layup, internal volume allocation, and MOF loading configuration.

The objective is to develop a computational framework that balances hydrogen-storage efficiency, structural safety, and lightweight design. The pressure vessel will be evaluated through simplified analytical or numerical models estimating mass, stress levels, deformation, and safety factors under internal pressure. At the same time, MOF-related descriptors, such as porosity, density, surface area, and adsorption capacity, will be used to estimate the contribution of the adsorbent material to hydrogen storage.
The main outcome will be a prototype AI-driven inverse design environment where an agent learns to propose improved pressure-vessel/MOF configurations under engineering constraints. The thesis will demonstrate how reinforcement learning can support the
early-stage design of advanced hydrogen-storage systems by combining structural mechanics, materials selection, and data-driven optimizationThe main outcome will be a prototype AI-driven inverse design environment where an agent learns to propose improved pressure-vessel/MOF configurations under engineering constraints. The thesis will demonstrate how reinforcement learning can support the early-stage design of advanced hydrogen-storage systems by combining structural mechanics, materials selection, and data-driven optimization.

Qualifications required: Python, Linear Algebra