There are a lot of postings on this forum in the style of - 'Which CFD code is best for this, that or the other application?'. It is my impression that the one code least likely to get an endorsement from the learned readership of this forum is Fluent (CFDRC & CFX seem to generate a lot more positive responses). Is there some sort of split appearing here between those who know (who wouldn't touch Fluent with a barge pole) and those who don't (who are suckered in by slick marketing and extravagant claims of success). Is Fluent the MS Windows of the CFD world with everyone else playing the part of various flavours of UNIX ?

Nah. Half the 'recommendations' are from the vendors themselves. The rest of the people that seem to know which code supposedly is better than another for a specific purpose never substantiate this. WHY would CFX be so much better than Fluent for multi-phase flow? Besides, I think there will be plenty of areas other than multiphase flow (combustion, grid creation, etc) where Fluent beats the other packages. Maybe Fluent advocates are more modest.

Nah. Good marketing.

If you want to know why cfx is better than Fluent at multi-phase then try applying the models.

It converges well, fast and easily with cfx.

It doesn't with Fluent.

This is well know to the cfd community, but as fred said, Fluent marketing machine is very strong. they claim to have all the models, but rarely do they work (well).

The proof is in the eating. All I can suggest is that ,if you're having trouble converging the models in fluent then try another package.

If not then, stay where you are.

Fair?

I personally had no intention to discuss this matter, however, this might be a good example that most cfd-online people are not aware of.

I was one of the participants at the 1st MIT conference on Computational Solid and Fluid Mechanics this summer. Frankly, it was one of the best I ever been. And, there was a mini-symposium on convective flows with 4 or 5 sessions, if I recall it correctly. The guys from Los Alamos and Sandia who were in charge for the symposium decided to pick a problem of transient natural convection in a rectangular cavity with aspect ratio 1:8 for Ra=340,000. The idea was to compare numerical solutions as no solutions were available to the problem. None published, none studied. This was quite a fair exercise, as everyone had no idea initially in which direction to "adjust" the solutions.

Having said this, most of the participants from different schools/labs were able to get numerical solutions quite close to the "benchmarked" solution by Le Quere who used a spectral method. Only a guy from Fluent presented solutions (using Fluent, of course) that were incredibly off, say 10 times off for some parameters/values compared. He had a good reason to feel embarrassed, I assume. And, I think he knew he would be fired right after the conference is over. I believe these Fluent's solutions were not only because Fluent is so bad ( I personally believe it's as good/bad as other packages) -- but rather the quality of Fluent personnel. And note that Fluent makes a lot of money by consulting. Imagine these people doing consulting for you now... This is what this forum should know, I believe.

See, this is where are try to learn from other people that already have eaten the pudding, but the answer to the big questions remain vague.

"It converges well, fast and easily with cfx. It doesn't with Fluent."

Now, does that mean that the results you got from both pakages were the same, but you had troubles getting there with Fluent and not with CFX? So it's a stability problem and not the accuracy of models & solvers?

"they claim to have all the models, but rarely do they work (well)."

Now, does this mean that both the structured (4.5) and unstructured (5.x) versions of Fluent gave troubles? Eulerian or Lagrangian models? DPM, VOF etc? Or just plain troubles in every model, every solver?

Not everyone has the freedom to switch from code to code to try and speed up convergence. A change from one package to the other has to be based on facts and it would be nice if this forum could provide a bit more solid feedback backed by actual comparison results.

Erwin,

What you can do in case of any doubt is to approach the different commercial companies (say FLUENT, CFX and STAR) and give them a typical test case scenario to run as a prospective client. You will then be able to assess how performant the codes are.

You can also rely on the discussions here or/and on the list of commercial companies using one particular code for a given type of application as a suggestion for the code being acceptable in their field maybe (?). Here again talk to the commercial providers. they have references that you can check.

But I think it is difficult for most people in the forum to be fully objective because they mostly use one code, and they don't have that much time to run generic test cases unfortunately. During my PhD, when I still had that time, I did attempt to reproduce a general test case for which I had detailed data with the softwares we had, before embarking onto the project with CFX.

Herve

COBOK,

Was this exercise meant as a test for the solvers or the physical models (LES, k-e, second order model etc.) or the combination. If it is intended to test the solver, I am quite surprised any solver would be so bad since the problem seems to involve a simple cartesian geometry.

If the solution depends on the physical model, then it is altogether a different discussion. Commercial companies do not research with physical models. They just take what is published and "well accepted" and code it in and couple it to their solver. Some times, the coupling isn't very good. The numerics and the physical model can interfere with one another and have adverse effects on accuracy, efficiency (or even stability) or both. A good example is LES modeling. The numerical damping in most commercial codes would make an LES model redundant. I have actually seen LES spectra generated using one of the commercial codes and it was very evident that the numerical damping was unacceptable (atleast from the LES point of view). Based on what I have seen or read of the commercial code formulations (and a few results I have seen), I have reason to believe that the results would have been similar if another commercial code were used.

As far as physical modeling is concerned, some people seem to have a intuition about how to tune them for a particular problem even without experimental data (I would call it wisdom based on experience). A good example here is the K-e model. The model constants, curvature correction (and so many other specific things) are things that many people need to play with to match the experiments. But, some I am sure can do it much faster or even in a couple of iterations.

That said, could you provide us with some more details of this conference (like the slides or papers).

The reference is "Computational Fluid and Solid Mechanics", (Ed. K.J.Bathe), Proc. 1st MIT Conference on Computational Fluid and Solid Mechanics, Vol. 2, Elsevier, 2001. Pages 1446-1509 (topics on CFD for the Natural Convection Problem, 16 contributions). The session chairs talked about publishing summary in the IJNMF, but I have not seen any yet. Also, I have a tables with comparisons of the numerical solutions distributed during the conference, and I assume it has to be somewhere on the net (Sandia/Los Alamos).

The flow is laminar, 2d, but time-dependent. It's about a transition to a quasiperiodic flow. So, no turbulence modeling is involved. The geometry is extremely simple, as you have already noted.

Well, I would bet my money that even 1D Burger's equation solution (for a reasonably low value of viscosity) obtained by commercial software would be greatly overdiffused or oscillatory. Remember looking at 1D converging-diverging nozzle benchmarking a couple of years ago by a commercial vendor. The statement was: solution converged, the shock wave was non-existent (just a hill-looking thing) because of excessive diffusion, but it converged. That's why you get a PhD in developing more accurate numerics for a particular problem. Now, try to use that very accurate University code to mesh any kind of 3D geometry, try to get that shock wave standing on a "single node" for any kind of real world problem, try to get those nice color printouts to sell to a customer. Does not work. My point is, it's important to know exactly what you are getting in both cases, how accurate you want your solution to be, what you are willing to pay. Seems like for an accurate solution with all bells and whistles a company would need to have its own in-house code, written/run by its own specialists and backed up by years of experience. But then no one else gets it, because it's highly secret, or if one gets it (Nat'l Labs), then it doesn't run on his machine/his physics/his graphics card/his county.....

COBOK,

In that case, it is good to know that those of us developing research codes for specific problems are still needed in an environment that is increasingly being dominated by commercial codes.

Do you happen to know if any other commercial company participated in this exercise. I would like to know how good the best commercial solution was. I will be on the look out for the publications. Thanks.

I couldn't agree more.

Do you think Fluent would be the world leader in CFD if it was a crap code? Fluent is 3 times bigger than its nearest rival Star-CD. In my humble opinion, Fluent is the most user friendly, robust, unstructured CFD code available and this is why Fluent is so succesfull. I am sure one can do the same type of simulations using Star-CD or any other code but Fluent is by far the easiest to use. Fluent pioneered unstructured mesh technology and this allowed people to mesh and analyze complex geometries in a very short period of time. Fluent's technical support is also the best available anywhere. I am sure there are many excellent codes for pure academic research but for real-world industrial problems Fluent blows the competition away.

Thank you for your patience,

shiva-G

Hi COBOK,

Fluent makes most of its money from sales of licenses. Consulting is a very small percentage of its business. There are a lot of other CFD codes who make most of their money from consulting. They do very good work, Star-CD for example, but the reason they get so much consulting work is because their codes are not easy to use. I feel it is very important to put CFD in the hands of the engineer. this is where it will be the most usefull and this is where Fluent succeeds the most.

chathrapati shiva-G

Kalyan,

I am not aware of the reasons, but except for Fluent (which was one of the sponsors for the conference) no other CFD company did participate. I assume most of the companies simply were not informed and missed a good opportunity to demonstrate attractiveness of their product. Well, I myself was not aware of the symposium either, even though I keep track on these happenings.

Two commercial codes were utilized, namely LS-DYNA and FIDAP. The former was initially developed for solids/structures and recently extended for fluids at Livermore (I might be wrong here -- please correct me if I am). The latter is now marketed by Fluent, however, the people who did run analyses were not affiliated with FIDAP, rather, with the "organizing committee". That could be a reason that these codes were not too off. Tuning sometimes might help a lot.

What I would want to see is a comparison of the numerical solutions produced by different commercial codes/vendors for a set of problems that have not been solved before. Every company gets an invitation to do so, and gets, say, one month to complete all the job and submit results to an unbiased committee. And everyone gets a chance to see and compare results. Sweet dreams...

Do you think Microsoft would be the world leader if its OS was crap?

There are many who think it is.

Which brings us right back to the beginning : Good marketing.

Amita

Good point.

Engineer? I couldn't agree more. I hope John Chien is reading this.

Fluent lead the way with unstructuring, but CFX5.5 has caught up and perhaps overtaken it.

To say Fluent is the most robust is nonsense. Many models are difficult to converge and as you see above, often wrong.

The CFX5 solver is extremely strong, robust and fast (see comments on comparison between cfx and star-cd, by USERS not cfx employees).

The multi-phase models have always been very strong, well validated and robust in cfx. Coupling this with the strong solver make it the easiest and post powerful tool for multi-phase flows.

This includes a very good new free-surface model.

All I said before was, if you having trouble coverging your models in Fluent and are disillusioned with the promising of this code does everything, try another on trial.

Fair?