Overview
Why read this?
Computational Fluid Dynamics (CFD) is a great way to visualise and understand flows… until you need weeks (or months) of solver time to capture every eddy of a wind-farm wake, or the swirling chaos inside an industrial mixing vessel. FluidGPT-1—our proof-of-concept Artificial-Intelligence (AI) surrogate—already proves a different future is possible, but right now it’s limited to small example cases. Your PhD will transform that prototype into FluidGPT-2D, a generative transformer that can take a handful of CFD seed timesteps you run yourself and instantly spin them into the full unsteady flow field.
About this opportunity
What you’ll actually do
- Build the dataset—Use the University of Liverpool’s Barkla supercomputer to generate a vast library of training flows with the Lattice-Boltzmann Method (LBM). Instead of solving for velocity and pressure directly, LBM tracks particle distribution functions on a lattice which makes it easy to parallelise and perfect for GPUs, so you can quickly produce high-resolution, vortex-rich datasets.
- Harness OpenLB—We collaborate with Karlsruhe Institute of Technology (Germany), home of OpenLB, the open-source C++ framework that lets you prototype complex LBM setups. During an optional secondment, you can work alongside the core OpenLB team, learn their codebase, and contribute new modules that ship back to the community.
- Design the AI—Experiment with transformers (as in ChatGPT) and convolutional encoders/decoders (as in computer vision). Integrate physics-informed losses so the network learns to predict truth to nature.
- Validate in the real world – Compare AI-generated velocity fields with laser-Doppler anemometry and particle-image velocimetry measurements in our fluids lab or from our partners. When the predicted flow matches the experimental one, you’ll know you’re ready for production.
- Release and publish – Open-source FluidGPT-2D under the GNU-GPL, write high-impact journal papers, and show your work at international conferences.
Training, people, and support
- Supervisors—Dr Davide Dapelo (Lattice-Boltzmann expert, AI-CFD hybrid specialist) and Professor John Bridgeman (CFD and fluid mechanics expert).
- Skills menu—C++, Python, TensorFlow, HPC optimisation, experimental diagnostics, and public-speaking.
- Optional adventure—One months at KIT to dive deep into OpenLB’s code and German research culture.
Life in Liverpool
Student Crowd named Liverpool the best student city, while Time Out voted it the 7th best city in the world. We are spoilt with beautiful parks, as well as our famous waterfront with the iconic Royal Albert Dock, and nearby beaches. Liverpool is home to the largest number of museums and galleries outside of London. We also proudly lay claim to Time Out’s coolest neighbourhood in the UK – The Baltic Triangle is a buzzy creative hub that has sprung up in old shipping warehouses. And you can’t mention Liverpool, without talking about the people – voted the 4th friendliest city in the world by Rough Guide, the city is renowned for its warmth and humour.
Want to know more?
Feel free to email either of the two supervisors with any questions.
TL;DR (skim-friendly)
CFD runs can take months—FluidGPT-2D aims to shrink them to minutes. You will:
- Generate massive Lattice-Boltzmann datasets with OpenLB.
- Train a physics-aware transformer that turns a few seed timesteps into full unsteady flow predictions.
- Validate with lasers, collaborate with international partners, publish, and open-source your code.
- Get fully funded travel to Germany if you want.