Hi,

I need to perform a 3D turbulence calculation with a complicated irregular geometry. To understand the complexity involved, think of Couette-Taylor flow with a complicated obstruction at some azimuthal position, and boundary conditions on the top and bottom lids. I have 2 years of CFD experience with Spectral methods, but don't have too much knowledge of turbulence modeling by finite volume/elements.

What would you recommend?
1. Should I start from scratch and develop my own solver? I am well versed with F90 programming, and have access to fast Linux Machines.
2. Should I procure something like FEMLAB (<a href=http://www.femlab.com/>link here</a>) which will make the exact flow modeling easier,
3. Or should I think of procuring a packaged solver like FLUENT, which will reduce the time I spend on it drastically. If not a commercial package, does anyone know of a realiable finite element or finite volume solver that is freely available?


I have about 6 months to execute the project. The main constraint is time. An ANSYS model (in IGES format) of the flow domain is ready; boundary conditions are to be specified.

One opinion - conventional wisdom is that to develop a CFD code for personal use requires at least a year of hard work. This time could be reduced if you have source code for a solver that you understand and can modify extensively. Remember you would need to deal with algorithm details as well as the physics to be included.

Understanding enough about turbulence modeling to make an informed selection from models available in the literature, and then to actually install one or more of those models into your code, test it (and perhaps rip it out and replace it with your second choice!) could well require another year.

This estimate already exceeds your time limit by 1.5 years.

If you have the budget, take some time to determine what turbulence model will do the job for you, then have some serious discussions with the vendors who have that model in their product. Their code is developed already. You need to satisfy yourself that it will do the job you need. And the vendor can give you a great deal of help in getting the solutions you need.

At best, you're looking at 6 months of hard work.

Good Luck!

Well, one thing you probably need to decide is the turbulence model you would want to use which you find suits your very problem, I it is worth it to spend a few weeks to have a good literature review as you will anyway need to almost know different aspects and the advantages and disadvantages of the different turbulece models. A motivated person could do this in about a month. Secondly, I personally believe in order to fuly comprehend the phenomenon, it is worth to develop a model by your own, especially when you are good in coding from what I understand, this could be done for convenient turbulence models in not a very long time, and maybe you would find that model not far from your expectations for your very problem regarding the fact that sophisticated turbulence models like Reynolds stress model etc, may lose some of their attractiveness when they are applied to three-dimensiuonal problems due to the computational time. I think you would probably need about 1.5 months to develop a k-e model and probably about 2 weeks to include it in your model. This, however, may change dramatically if you want to use Reynolds stress or some flux models. Good luck with the hard work.

You have not sufficiently defined your problem. We do not know if the turbulent flow is high or low Reynolds number and, perhaps more importantly, we do not know how accurate your answers need to be.

>>> 1. Should I start from scratch and develop my own solver? I am well versed with F90 programming, and have access to fast Linux Machines.

I would agree with Jim's answer here. Unless there are other factors we do not know about this is not a sensible option.

>>> 2. Should I procure something like FEMLAB (link here) which will make the exact flow modeling easier,

If accuracy and/or flexibility is a concern I would be very wary of products that promote ease-of-use particularly if they are not widely used in the CFD community. (Perhaps I should declare some bias here since the linked web page crashed my browser after I had started replying to your message!)

>>> Or should I think of procuring a packaged solver like FLUENT, which will reduce the time I spend on it drastically.

If the significant upfront costs are not a problem then this is the fastest route to an answer - perhaps a week or so after installation. This should give you plenty of time to assess the quality and possibly have another go using a different set of modelling assumptions. The only signficant risk I can see is if the accuracy requirement is high and the package has not implemented the approach you require.

After seeing the other replies, I'll add to mine:

My answer should be read as 'time to get the first calculation done.' Then you decide what you should have put in the model the first time!

There are questions of resolution - in my experience (and this includes observing some of the best at work!) you never have enough resolution the first shot. Or you've put the finest mesh in the wrong places. So plan on reworking after you get first results.

Also overlooked. The ability to look at your results graphically is necessary if you're to interpret all of those tables of numbers. If you write your own code, you'll need to provide for the graphics as well. That's a separate purchase and learning curve. But I think you'll find graphics either integrated or added on to the commercial packages.

Writing one's own code IS FUN and is absolutely the best way to learn the whole game - but it's not something that can be done quickly.

Hi,

Six months is not enough to develop your own 3-d code and procuring a commercial code would be your best bet. FEMLAB does not have any in-built turbulence models and does not have many fluid-dynamic options(if you look at their manual they even attempt to solve the shock-tube problem by expanding out the 1-d conservative Euler equations into a non-conservative form !!!). You may want to look at commercial codes such as Fluent among others (depending upon your budget) which are specifically designed for fluid flow.

Why not let somebody let do it for you? If this is a one-of project then paying a CFD-consultant (some CFD package vendors will do it for you including FLUENT, STAR-CD and CFX) might be quicker and cheaper. You can skip the learning curve then. Only drawback is not using your hardware.