Hi Guys,

I am working on CFD tool hunting for my company. Now I have two candidates: Fluent and COMSOL. Could anyone here having experience with either of these two give me some hints? I need the tool for wind-load structural vibration simulation. Thanks a lot in advance.

I haven't used Fluent so I can't say much about it, but I have used COMSOL.

We had lots of problems initially with COMSOL. Even on the newest version I had massive problems getting it to import my CAD files from AutoCAD and Solidworks. I eventually had to go back to its very basic shapes utility to get what I could of my geometries. A colleague went to a COMSOL seminar on fluid flow and came back saying that even the instructor confessed to COMSOL's difficulties in fluid modeling (after some people had stormed out). We did a simple tube problem and the results were no where near analytic calcs (which we trusted more). COMSOL, however, did a pretty good job modeling some conductive problems we gave it. It was able to predict with accuracy the conductive properties and temperatures when our material (Tantalum) was put in the database.

I'm afraid I didn't get chance to crack open any vibration analysis, but with the problems we had, I still wouldn't recommend it.

Overall, my research group had big problems with COMSOL, and now we've gone over to CFX to take care of our fluid models. I haven't used Fluent, but I would go with it instead of COMSOL.

-M

Wooster,

Thanks a lot. Actually I have the similar feeling from my gut. I had problems of convergence in COMSOL when the size of the model goes large. But I do like their solving capacities for multiphysics. Most probably, we will go to the CFX or fluent finally.

-N.M.

I'm enjoying CFX. After I stopped using COMSOL and started using CFX, I had the vague feeling I sold a 84' Civic and bought a SR-71 Blackbird (if I remember the name right). That's about the difference.

As for convergence, I like how CFX works. It shows the residuals as it solves, thus after a while of use, I can pinpoint problems just looking at a particular residual and how it behaves over a period of time. I can then stop the solver and go back, change the problem, and use the old output to start off with (provided the changes aren't huge). With COMSOL there was no real data like that so you couldn't say until you were done if there were problems or not (unless it kicked you out). I had a problem like that were a simple change from "automatic" to a value caused my iterations to go from >500 to less than 30. I wouldn't have known had I used COMSOL. "Oh well, let 'er run till next week to get our answers." CFX does multiphysics as well, but I'd do some research first. I've got a research license so it may be different in that aspect than a commercial license.

I haven't used Fluent so maybe someone can guide you in that direction. As for me, I'd sell you CFX. It's really nice but it can get very complicated. Also, I'm sure you know this but we often had issues in COMSOL getting our models to import. Whatever you choose, getting something that can easily import pre-made models or has a usable modeling interface (ANSYS Workbench springs to mind) is a deal breaker. -M

Wooster,

You didn't specify which version of COMSOL you were using but it's good to know that the current version (3.3a) does allow the display of convergence while solving (cf User's Guide, page 458 and 489), as did the previous version.

Regarding the inaccuracy you report in your attempts to model of flow in a tube, I'd be curious to take a look at them for you - just send me your .mph file.

COMSOL Multiphysics has done very well in independent benchmarks and refereed papers regarding getting close to analytical solutions or available experimental data. The Model Libraries contain several such tutorials in the product (backstep FLOW, developing flow, flow over a cylinder, etc) where results are compared to published data.

Jeff

[ivan-s]

Regarding COMSOL, it actually has a fairly steep learning curve with regards to getting CFD to work. I now have a very complex model running in it, something with 13e6 degrees of freedom, a network of hundreds of tubes, and it runs and does converge in reasonable time. My total time invested to learn it and get it this far was around six months, while working on other projects not related to CFD.

I am currently waiting for 5.0 which should contain a better set of equations for CFD turbulent flow.

I do, however, understand the frustration of getting it to run. It's almost like it was written by someone in their basement, without much thought of what a user will do when he gets a random "Illegal input vector" error message, for example. I mean, which was was illegal? Where do I find it?

But it is inexpensive and it does work, and it is fully capable with multiphysics.