Data-driven modeling

OpenFOAM special interest group

1st meeting

Plan for today

  • introduction
  • committee overview
  • ongoing/upcoming activities
    • Github repository
    • OpenFOAM workshop
    • OpenFOAM+ML hackathon

Hi, my name is ...

  • affiliation
  • experience with OpenFOAM
  • experience with ML (and OF/CFD)
  • ...

We have scheduled only 1h for the meeting :)

Committee overview

Current members, alphabetical

  • Andre Weiner, TU Braunschweig, chair
  • Dirk Gründing
  • Holger Marschall, TU Darmstadt
  • Karthik Kashinath, Nvidia
  • Neil Ashton, AWS
  • Tomislav Marić, TU Darmstadt, co-chair

Want to get involved?

  • a.weiner@tu-braunschweig.de
  • maric@mma.tu-darmstadt.de

Short term goals

  • lower technical barrier to get started with data-driven modeling and OF
  • promote data-driven modeling available in OF (e.g., dynamic mode decomposition)

More details available in the write-up.

Long term goals

  • when/how to combine ML and CFD?
  • accelerate developments (best practices)
  • ML as natural element of CFD workflow

More details available in the write-up.

Roadmap (excerpt)

  • automated integration of libTorch in OF
  • OF-Python interface
  • tutorials and lecture material
  • ML+OF hackathon

More details available in the write-up.

ongoing/upcoming activities

Github repository

https://github.com/AndreWeiner/mlfoam

  • point of contact
  • committee overview
  • aggregation of links

For access to the repo, send your Github username to Tomsilav or Andre (only committee members).

OpenFOAM workshop

  • https://openfoamworkshop.org/
  • 11 to 14 July 2022, Cambridge
  • submission deadline 14 Feb 2022
  • abstract training ~1 page
  • hands-on experience with technology/research
  • ~90min
  • dynamic mode decomposition
  • physics-based DL with OpenFOAM
  • ?

Any planned contributions on data-driven methods and OpenFOAM?

OF-ML-Hackathon

Hackathon: intensive get-together for creative problem solving (hacking) in groups.

Elements of a hackathon:

  1. central objective/problem/theme
  2. limited time, less than 3 days
  3. team work, 2-5 people
  4. parallel workshops, mentor support
  5. final project presentation
  6. evaluation by jury
  7. prices for winner(s)

Why this format?

  • kick-off (research) projects
  • initiate collaborations
  • get hands-on experience
  • bring together divers backgrounds
  • create demonstrators for community
  • ...

Some ideas:

  • 2-3 days, July, after workshop
  • online, Gather
  • workshops, mentorship by TC
  • renowned community members for jury

Formulating a good objective is hard ...

Maybe too fuzzy:

"Combine ML and OpenFOAM"

Maybe too specialized:

"Implement a data-driven turbulence model"

Maybe too individual:

"Bring your research project and enhance it with ML"

Closing words

  • next meeting: April, Doodle follows
  • protocol will be passed around
  • get people involved!