4
$\begingroup$

I'm making a model of a square box where water comes in and the water level rises. I want it to be a transient, turbulent, VOF-model. The velocity of water entering changes in time ($-0.2$ to $0.2$ m/s).

For now I use a steady, laminar inflow of $0.2$ m/s, a grid of $10,000$ cells (the geometry is roughly $100\times 100\times 6$ m) and solution method PISO. To run the model I've used time step size $0.1$ s and $3600$ time steps ($10$ minutes); bigger time steps led to divergence. It took $2.5$ hours to calculate this and it wasn't even turbulent yet. In the end I've to model a period of $12$ hours, with this speed of calculation that will take more than $5$ days.

What are things I can adapt in Fluent which will (extremely) reduce this calculation time?

Thanks in advance.

Improve this question
$\endgroup$
4
  • $\begingroup$ I know that FLuent has licenses for parallel computing (distributed and shared memory). Have you tried using these? $\endgroup$
    – Paul
    May 28, 2013 at 14:11
  • $\begingroup$ Now didn't try it yet. I wanted first to try and make the model setup less complicated or time consuming to calculate. $\endgroup$
    – Elyse
    May 28, 2013 at 15:04
  • $\begingroup$ How would you quantify extremely? Transient, turbulent problems are difficult to compute, so 5 days does not strike me as dramatic for a problem like this. $\endgroup$
    – Schorsch
    May 30, 2013 at 17:47
  • $\begingroup$ Oke thank you. I don't have a lot time left to do this research (2 weeks), that's why I was wondering if there was a quicker way. $\endgroup$
    – Elyse
    Jun 7, 2013 at 9:41

1 Answer 1

Reset to default
2
$\begingroup$

The reduction of calculation time in CFD is a general problem. Although I have never worked with FLuent, I don't think that there are features that magically speed up your simulation. That's why I can only give some VERY general hints. For further insight, I suggest you browse the internet, e.g. cfd-online for the highlighted keywords.

  • Use mesh adaptation to reduce the scale of your problem. In the regions away from the walls, you maybe can reduce the number of cells. For the near-wall regions you have to take care of the boundary layer. Monitoring the y+ value helps you estimate how coarse you can go. Also the use of wall functions for your turbulence model can be considered.

  • Do some try and error with the relaxation parameters. This is eventually balancing stability and convergence speed.

  • And, as Paul suggested, you can simply use more computation power.

Improve this answer
$\endgroup$
1
  • 2
    $\begingroup$ With 10,000 cells, it seems unlikely (but possible) that parallelism will help. If there's some leeway in quality of solution, one could reduce grid resolution, reduce tolerances for iterative linear solvers, maybe reduce the complexity of the turbulence model (use LES instead of DNS, or RANS instead of LES). These steps will help to calculate results faster, but they won't be as accurate. $\endgroup$
    – Geoff Oxberry
    May 30, 2013 at 4:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.