Hello,

I'm currently using Flow-3D to model turbulent 2-fluid flow. I use air flow (100-200 m/s) onto a water surface. The problem is that the solution "blows up" after a short time with no previous warning (mu peaks and dt drops). I've run well over 400 different simulations with slightly different grids and/or different BVs and I don't see any pattern to why the solution goes crazy. I'm at a loss here and wonder if anyone has experienced this problem before.

Regards

Mikael

I don't use flow-3d but I suspect your problem has something to do with the unphysically large values of your wind velocity (in the atmosphere 10m/s is about normal for the wind at 100m above the surface and is much lower near the surface - friction velocities of around 0.5 m/s).

At the speeds you're using the shear would rip the interface between the fluids apart.

Thank you for your answer!

I realise I have misformulated the question. I use a BV of 100-200 m/s for the airflow (and only 21 dm3/min). The surface lies about 2 dm below the inlet so the air velocity is not of the same magnitude there. I've used a solid to represent water in order to get some guess of the velocity field above the surface. It shows that (using an RNG turbulence model) the air velocity lie between 3-10 m/s.

Regards

Mikael

Back in the dark ages (mid-90's!) when I was using flow-3d, their support desk was really helpful. That was one of the beauties of working with a small company - everyone had a stake in making the product useful for the customer.

Have you asked the support people for help?

Yes I have. And as you say, they are really helpful and eager to help. However, the engineer I've been in contact with doesn't have an answer to the problem. He suggests to use constant viscosity instead of a turbulence model. That might work (I haven't got a reference for some of his assumptions but I guess they are fine). But since I have worked with this problem for quite some time it pains me to just give up and use a constant viscosity model.

What is the cell/mesh size you are working with? Does Flow-3D actually complete a time step?

I have tried different cell sizes from ~5 mm down to 1mm.

Yes Flow-3D manages several timesteps before going "crazy".

Regards

Going back to my earlier post and Flow-Science's suggestion to you to "use a constant viscosity".

Did they perhaps intend that you first do a calculation with a large constant viscosity, running that to steady state to establish an initial condition for the turbulent simulation?

In my experience, time-and-space varying viscosity (i.e. turbulent models) can really complicate starting a simulation - and it appears to be complicating yours!

So, if you haven't done so already, you might try running an initial case with a high, constant viscosity, saving the resulting flow field, then turning on the turbulence model with the saved field as an initial condition. I'd expect that you'll need to cut the time step for the restart and probably do some more experimentation.

Good luck!

Thank you. Actually I did try some simulations a few months ago using an "old" flow field but didn't manage any good results. I might have given up to early though. I'll try it again.

Oh btw. Flow-3D does not use a constant time step, it is based on a few stability criterias. Only initial time step is needed.

Regards

You're right. I wasn't careful in my wording.

Often it's necessary to select an initial time step well below what the stability criteria allow when starting a simulation. The code will 'grow' the time step over many cycles to the allowable value. The reason is that the stability criteria are derived from simplifed analysis and are heavy on experience. And a startup, even from a good IC, can really push the limits.

In your case, I'd suggest a starting delta t well below the value used at the end of your constant-viscosity initial run. And well below the value that flow-3d selects to start the run.