Machine Learning-Driven Molecular Simulations of Gas Transport in Polymeric Materials

Summary:

This PhD project will focus on simulating gas transport in polymeric materials using advanced machine learning techniques. Specifically, the project will utilise the MACE machine learning interatomic potential framework to improve models of gas diffusion and solubility in polymers. This research will further our understanding of the aging process of polymeric materials, which is crucial for a wide range of industrial applications.

Background:

Polymeric materials are essential in various high-tech industries, such as packaging, membranes for gas separation, fuel cells, and batteries. However, understanding their behaviour, particularly with respect to gas transport, remains a significant challenge. In disordered polymer systems like glassy polymers, gas diffusion often deviates from ideal Fickian behaviour, following complex, system-dependent mechanisms that are poorly understood. These materials also undergo aging, which affects their structural and transport properties over time. Simulating these processes is computationally demanding, as it requires large time and length scales, and existing classical models fail to capture the intricate interactions between gases and polymers. Additionally, electronic structure calculations, though rather accurate, are too resource-intensive for large-scale simulations over extended periods. To address these challenges, this project aims to apply the innovative machine learning MACE framework. MACE allows for more efficient simulations by using machine learning to capture the underlying physics of gas transport, offering a balance between accuracy and computational efficiency, while also enabling the modelling of aging effects in polymer systems.

Objectives:

The primary objective of this project is to develop an advanced MACE potential for simulating gas transport in polymeric systems, addressing both diffusion and solubility properties. The student will fine-tune existing MACE models to improve their precision and transferability, making a new model which is suitable for a variety of polymer systems. This will involve integrating molecular dynamics simulations, electronic structure calculations, and machine learning techniques to develop accurate models that capture the complexities of aging and material degradation. Furthermore, the project will focus on incorporating uncertainty quantification into the models to ensure their reliability in predicting long-term material performance. Ultimately, the goal is to provide a computational tool capable of simulating the behaviour of polymeric materials under real-world conditions, helping to inform the development of more reliable and durable materials for various applications.