Geometric Deep Learning for Particle-Based Computational Fluid Dynamics

Project Description

Computational Fluid Dynamics (CFD) underlies many of the workloads that we run on our supercomputers, this research will investigate how machine learning can be used to accelerate CFD computation, potentially delivering significant improvements in accuracy at reduced runtime.

Primary Supervisor: Dr Joseph O’Connor

Project Overview

Computational fluid dynamics (CFD) plays a crucial role in engineering, science and understanding the natural environment. While traditional CFD methods typically rely on the construction of a mesh on which to solve the governing equations, particle-based methods, such as smoothed particle hydrodynamics (SPH), forego this restriction in favour of freely moving particles whose dynamics are governed by the physics of the problem. This makes techniques like SPH very well suited to problems involving complex geometries and moving/deforming boundaries, such as the violent free-surface flows typically found in marine/coastal applications and fuel sloshing, for example.

One of the major challenges with CFD is that accurate simulations are usually expensive to compute and often require many days (sometimes weeks) to obtain on large-scale high-performance computing systems. Recent advances in machine learning have led to the development of data-driven surrogate models that approximate the high-fidelity CFD but are much faster to compute, enabling significant speedups over the full physics-based simulations. However, the majority of work in this area has so far been restricted to traditional mesh-based CFD methods. This is mainly due to the unique challenges posed by particle-based techniques, such as irregular (unstructured) and dynamic connectivity between particles.

Overview of the research area

Geometric deep learning (GDL) is a subfield of machine learning that aims to incorporate prior knowledge on the physical structure (e.g. symmetry, invariances) of a given problem directly into the learning process, thereby improving performance, robustness and generalisability. One particular class of GDL methods that is especially suited to particle-based CFD is graph neural networks (GNNs). GNNs are naturally capable of handling unstructured data with dynamic connectivity and pairwise interactions, such as those encountered in particle-based CFD methods, and it is this structural similarity that makes GNNs particularly attractive for developing data-driven surrogate models of particle-based CFD methods.

Although the potential advantages are clear, the application of GDL techniques to particle-based CFD is still an emerging area of research, with very little quantitative validation and performance benchmarking, especially in the context of real-world applications. With this in mind, the main aim of this project will be to investigate how GDL techniques can be adapted and applied to create low-cost data-driven surrogate models for particle-based CFD, with a focus on their performance and practical relevance for real-world engineering applications.

Student Requirements

A UK 2:1 honours degree, or its international equivalent, in a relevant subject such as computer science and informatics, physics, mathematics, engineering, biology, chemistry and geosciences.

You must be a competent programmer in at least one of C, C++, or Python and should be familiar with mathematical concepts such as algebra, linear algebra and probability and statistics. You must be willing to learn new techniques and technologies quickly, and an interest in developing new skills and expertise.

English Language requirements as set by University of Edinburgh

Recommended/Desirable Skills

  • A good understanding of machine learning techniques, especially geometric deep learning
  • Experience in working with any machine learning framework (e.g. PyTorch, TensorFlow)
  • Experience in the application and/or development of computational fluid dynamics methods, especially smoothed particle hydrodynamics
  • A good understanding of mathematical modelling in the context of fluid mechanics
  • Experience in compiling and running codes on high-performance computing systems

How to apply

Applications should be made via the University application form, available via the degree finder. Please note the proposed supervisor and project title from this page and include this in your application. You may also find this page is an useful starting point for a research proposal and we would strongly recommend discussing this further with the potential supervisor.

Further Information