Project: Scaling granular material simulations with deep geometric generative models (TAKEN)
Description
Granular materials are one of the world’s most widely used and manipulated materials, only behind water. The modelling of these materials is relevant to various sectors, including energy production, agriculture, cosmetics, construction, and the pharmaceutical industry. Granular materials are collections of discrete particles, which results in complex mechanical behaviour, featuring anisotropy, non-linearity, and path dependence. This combination of complexity and societal relevance means that modelling granular materials has remained a very active field of research for decades.
This MSc project aims to advance the state-of-the-art in simulating large-scale granular materials by leveraging advanced machine learning (ML) methodologies. Traditional simulation techniques, such as Discrete Element Method (DEM), provide high-fidelity insights into granular behavior but suffer from prohibitive computational costs, especially when scaling to industrially relevant system sizes.
The core of the research will involve developing a deep generative model that respects the symmetries present in the representation of the granular material systems. Various possibilities based on Graph Neural Networks (GNNs) and Transformer-based models exist that are suitable for such dynamic particle systems. The goal is to use such architecture to develop a generative model (based on flow matching or other diffusion-based approach) to reproduce the high-fidelity DEM simulations and act as a computationally efficient surrogate model.
Details
- Student
-
TVThanos Vaios
- Supervisor
-
Vlado Menkovski
- Interested?
- Get in contact