[michael]

hello,

i have computed a simple model, which is called elbow in the star tutorial guide, with star-cd. Then i have converted the model to openfoam and tried to transfer the settings manually. I've normalized the results from foam with the pressure (multiplied with the pressure).

In the results there is a pressure difference between starcd and foam about 10 percent or more. The settings:

OpenFOAM:
Turbulenz model: Standard k - epsilon
Difference Scheme: upwind
Solver: simplefoam
pressureOutlet, 0 Pa

STAR:
air, density 1.205
Turbulenz model: k epsilon high reynolds number
Difference Scheme: UD
double precision
pressure outlet, 0 Pa

Then i've switched the solver to laminar, because of the influence of the turbulence model, but afterwards the difference is higher. Why is there such a big difference?

[gschaider]

And you also normalized the viscosity? That's what I usually forget to do when I compare with the commercial solver we have here.

[michael]

yes, in star molecular viscosity and in foam the dynamic viscosity (mu/rho). what have been your results?

[gschaider]

OK. Sorry for asking.

My results (for a more complicated "real life" geometry) with the commercial solver starting with an F using k-eps/inkompressible (==simpleFoam) showed 2.5% difference for the pressure drop (mass flows at selected control surfaces were in the order of 0% to 6%)

[michael]

thanks for your replies.

my results are:

k - epsilon model:
1584 cells : 3.58 % difference
12672 cells: 1.89 % difference

1.89 % pressure loss is acceptable, but there is also a higher difference between inlet and outlet.

laminar:
1584 cells : 8.23 % difference
12672 cells: 6.68 % difference

8.23 % is very high

What commercial solver you've used? Are your results acceptable for a "real life" geometry? Perhaps there is some difference in the differencing schemes. What are your settings in both codes?

thanks!

[gschaider]

"commercial solver starting with an F": Fluent

TBH I don't know which discretization schemes. (It was a quick shot with the defaults of both solvers)

I was quite satisfied with the 2.5% (although it leaves room for improvement). It was in the range of what the partner could supply us data with. You can get bigger differences if you change the convergence criteria on the same solver.

[braennstroem]

Hi to both,

do you have some more information about your 'benchmarks'. It would be
interesting to know the costs for each simulation using OpenFoam
compared to StarCD and Fluent!?
Are there any other comparisons between OpenFoam and other packages out
there?

Greetings!
Fabian

[vtk_fan]

Hello,

I am very interested in this question as well. I am consulting for a major company here, and am wondering if the performance of OpenFOAM vs. Fluent would be comparable. For example, for an average external aerodynamic simulation, 3D, k-e, 1 million cells, hex-dominant grid, what would be the approximate time taken per iteration by OpenFOAM vs. Fluent? Also, would OpenFOAM take longer to converge, so that the total cost can be ascertained?

Even if OpenFOAM is free, the cost of a Fluent License (~20,000 US$) might be worth it if, let's say, it is 3-4 times faster than OpenFOAM at solving the above simulation. For time-bound projects, this is an important consideration. Any responses would be much appreciated.

[fra76]

From my experience, for an incompressible, turbulent external flow simulation, each fluent iteration is faster than an OpenFOAM one, but OF requires less iteration to converge.
The overall time is shorter for OF. Please, notice that I run my test case in parallel on a case bigger than 1 million cells, and OF scales really very well, over a fast network.
However, setting up an OF case requires more experience for some numerical settings that can speed up or completly fail your simulation.
So, as always, it's not an easy decision!

[vtk_fan]

Is there a place where I can get a roadmap on OpenFOAM development plans? I am also interested in dynamic mesh motion (rigid body only) for aerospace applications. The range of motion can be quite large, a turn through 60 degrees for flap movement. Has anyone done any benchmark comparisons on this kind of problem anywhere with OpenFOAM and Fluent 6.3's dynamic mesh capability?

I had a huge bunch of trouble with Fluent 6.1's dynamic mesh capability because it was very buggy and limited to rectilinear motion of the moving boundary.

[vtk_fan]

By the way, thanks Francesco for your information. Can you please provide some actual numbers and a more detailed description of your case? Many thanks.

[hjasak]

You seem to be asking for very specific performance numbers for cases of your interest - why don't you just run the test yourself and report the result?

Hrv

[fra76]

I cannot add more details on my tests, I'm sorry...
You can find some more hint here: http://www.cfd-online.com/OpenFOAM_D...es/1/3503.html
I've no experience with OF mesh movement, unfortunatly. However Fluent 6.3 should be much better of 6.1 on this side. Actually, it works quite well, but forget about hexcore meshing for that range of movement!

[vtk_fan]

>By Hrvoje Jasak on Monday, April 02, 2007 - 09:15 >pm: Edit Post

>You seem to be asking for very specific performance >numbers for cases of your interest - why don't you >just run the test yourself and report the result?

>Hrv

I plan to run a reduced size test case (200,000 cells) this weekend, and will report the results. My Fluent license expired last year, and I will not be able to do a one-to-one comparison.

Please do not get me wrong. I think OpenFOAM is a great alternative, and since it is open-source, long term it has very bright prospects of being more widely used than Fluent after another few years. These questions are a reflection of the interest in the software.

[jens_klostermann]

Hi Mark,

I can offer to run your testcase with Fluent, if you are willing to share the case (OpenFOAM case, incl. setup).

Jens

[vtk_fan]

Thanks Jens, I will definitely take you up on your offer early next week.