Welcome to the OpenFOAM machine learning hackathon repository! The hackathon is a community event organized by the data-driven modeling special interest group. If you are an OpenFOAM user excited about combining OpenFOAM and machine learning, this event is for you!
A hackathon is an intensive get-together for creative problem solving in groups. The corner stones of the OpenFOAM-ML hackathon are as follows:
- objective: we prepare 2-3 exciting projects combining recent ML techniques and OpenFOAM; the topics are diverse and change from event to event; for each project, a starter code is provided; your task is to advance the starter code in a self-chosen direction; we provide a couple of ideas to get you started
- time limit: the hackathon consists of three full days of intense hacking
- team work: each participant chooses the preferred project/starter code; within each project, the participants are split up into groups of 2-5 people; we aim for a minimum of one advanced hackathon participant per group to provide some guidance
- workshops: for each project, a workshop introduces the starter code and a necessary minimum of theory
- hacking sessions: the groups advance their projects; we aim to provide close mentor support for all groups via gather.town and slack
- final presentation: each team presents their final results and receives feedback from the other participants and mentors
The workshop is fully virtual. There is no geographical restriction for participants, but keep in mind that we cannot accommodate all time zones. The organizers' time zone is CET.
A detailed schedule will be provided. Note that you should reserve three full days for the hackathon.
Since we aim to provide all participants with close support during the hackathon, the number of participants is limited to 20. There are no registration fees or other costs. We can also provide compute resources thanks to AWS, so you do not need any specialized hardware. Admission is not guaranteed. Based on all applications, we will select the most suitable candidates.
-
Building scalable Computational Fluid Dynamics + Machine Learning Workflows using OpenFOAM and SmartSim
-
Integrating Physics-Informed Machine Learning Models into OpenFOAM.
-
Streaming Singular Value Decomposition and Dynamic Mode Decomposition for Computational Fluid Dynamics using OpenFOAM and SmartSim
-
Bayesian Optimization in Computational Fluid Dynamics using Ax and OpenFOAM. The goal of this example is to find optimal parameters of an OpenFOAM simulation that minimize some target function using Bayesian Optimization algorithms from Ax - Adaptive experimentation platform.
-
Learning and monitoring closed-loop flow control strategies with drlFoam and Gym-PreCICE; we'll apply deep reinforcement learning to control the flow past a cylinder using jet actuation and Rayleigh-Bérnard convection by heating; to learn about closed-loop control with DRL, refer to this article; this preprint introduces DRL applied to Rayleigh-Bérnard convection; for an introduction to DRL for flow control, you may also find this video helpful
Questions about the event? Get in touch by opening a new issue in this repository or contact the chairs of the data-driven modeling SIG.
This event is generously supported by our sponsors.
| AWS | ESI | Nvidia |
|---|---|---|
 |
 |
 |