CFD simulation and AI
Learning objectives
By the end of the course, participants will be able to:
- Place machine learning approaches applied to CFD in context, with their limits
- Build a numerical design of experiments and prepare simulation data
- Train a surrogate model and assess its range of validity
- Implement a simple example of a physics-informed neural network
- Validate and document a model for publication or engineering use
Audience
PhD students, research engineers and R&D engineers in fluid mechanics, energy or building services.
Prerequisites
Basics of fluid mechanics and CFD simulation; programming in Python.
Day-by-day programme
| Session | Content |
|---|---|
| Day 1 morning | Overview: machine learning and CFD, surrogate models, model reduction, physics-informed networks |
| Day 1 afternoon | Data generation: design of experiments, automating simulations, extracting and storing results |
| Day 2 morning | Surrogate models: regression, Gaussian processes, neural networks; validation and uncertainty |
| Day 2 afternoon | Model reduction: proper orthogonal decomposition (POD), field reconstruction |
| Day 3 morning | Physics-informed neural networks: principle, simple example, comparison with simulation |
| Day 3 afternoon | Mini-project on a provided flow case or one brought by the participant; presentation and critical discussion |

