Hi All,

I'm a CFD newbie trying to learn...

I'm currently evaluating commercial electronics CFD packages. ( IcePak, Coolit, Flotherm )

To me, they all all similiar in capability, price, etc...

except that IcePak has conformal meshing ( e.g. it doesn't have to use a cartesian grid like the other two codes ).

Depending on which sales rep I speak to, this conformal meshing is advantageous or not really.

Can anyone shed light on whether or not conformal meshing really results in a more accurate solution? If it's more accurate, does it bog down the solution so much that it's not really practical?

The way that the flotherm rep made said it to me was the the accuracy gain by conformal meshing is negligible and would result in convergences problems.

Can anyone compare and contrast the difference between meshing and solvers the two codes I mentioned above?

Any help/advice would be much appreciated!

Regards, Alvin Szeto

Conformal meshing (or Body-fitted mesh) is used by all modern and popular CFD codes. There is no such problem as convergence or otherwise nobody would have bounght Fluent and STAR-CD and used them daily in thousands applications. The main advantage of body-fitted mesh is that it does not need any special treatment near boundaries.

As for the codes which have not got it, you can guess the reason.

Hi John,

Thanks for the reply. Yes I realize that most general purpose CFD codes have conformal meshes, however, I'm only interested in Electronics Cooling and there are only a few specialized pacakges for this.

One is Icepak ( Fluent ) and the other is Flotherm ( Flomerics). The Flotherm salesman keeps telling me that conformal meshing is not really an advantage, and since they've optimized they're code, they DECIDED that it's more optimal to use a cartesian grid. Further he states that conformal meshing only makes the solver less robust and adds unecessary complexity to the solution. He also said that conform meshing is in the public domain and it's not anything new or top secret...

On the other hand, the Fluent rep states that since Icepak is based on Fluent, it has the resources to implement conformal meshing in a commercial code. He said that the only reason the other packages don't support it is because they don't have enough resources to implement it. So it's a matter of resources and not really whether or not cartesian meshing is superior.

1. Who is BSing me? 2. You bring up an interesting point, neglecting the special treatment near boundary conditions, is cartesian meshing equal to conformal meshing in efficiency, accuracy, robustness, etc...? 3. How do cartesian meshers handle boundary conditions? 4. Is this really that big a deal? According to the Flotherm rep, as long as I get a solution that within 5-10% of the real solution, it shouldn't matter what the "means" are... I just care about the ends.

Anyone with insight, please feel free to comment. I need all the information I can get to help me make my decision on which CFD package to purchase.

One last thing, for all you people who think I haven't read the FAQs or did some research on the net, I have! All the info I have found is outdated, or way too technical for me to relate back to my problem...

I remember reading an article in Flomerics website which defends that their method is as good as conformal meshing. So you can conclude that AT BEST their method can match conformal meshing - conformal meshing is not top secret technique, but it could be very difficult to implememt it in an existing specific code, especially in a commercial code.

But for you application, maybe it's ok since the geometry is so simple.

John,

Just out of curiosity, you don't have any affiliation with Flomerics do you? What you just said, is exactly what the sales guy said to me!

Okay, so everything else being equal ( price, functionality, ease of use, code robustness ), I SHOULD go with Icepak because I have the option of using a conformal mesh with their solver?

I disagree with you, in that my geometries are simple. I have complicated internal flow volumes ( decreasing radii tube bends with variable diameters ) or is that still considered simple geometry?

So to reiterate, conformal meshing is only advantageous because it can handle BCs better?

The way the flomerics guy put it to me, he stated that although the conformal mesh may look good on screen, it's just a picture because it will still have to be discretized to solve the NS equations...

If you think I'm asking to many obvious questions, I apologize. I'm a CFD newbie... If you know of any links that may be helpful for me to read up on, please send them my way.

Hi,

At what size are you working with, if it is in microns then better look fome lattice boltzmann code. Though Fluent has option of grid resizing in it solver, but I don't think any Commercial solver can handle Micro-Chaneel flows accurately. I don't know much about Flowtherm but surely know very much about Fluent so I can advise you on this regards.

Also try to check with Canada based Maya Thermals. It is also quite popular for electronic cooling.

Regard

Apurva

Apurva,

Thanks for your input.

No, I am not look at the micron level. I am not modeling the flow of heat pipe internally.

I am lookin at things in the order of mm ( leads of ICs ) to inches ( internal flow passages of cast electronics chassis ).

I have checked with Maya. They use a cartesian gridding system. It seems like IcePak is the only commercial electronics cooling package to offer conformal meshing.

Can you tell me more about conformal meshing, and its real advantages? Can finite volumes really have a non-planar face? This is what conformal meshing means to me. Do I have this right? And what does this really do for me?

Alvin,

It is not clear what your concerns are wrt cartesian vs "conformal" mesh. In applications you described Cartesian mesh will be very conformal as you deal with rectilinear things. Also, as far as I know of the 3 packages you listed, Icepak uses unstructured mesh and Coolit uses hybrid meshes with cartesian as well as tetrahedral elements. Again, in you application that probably does not buy you much anyway.

Whether you are expert or not, all you need presumably is to get accurate answers in as short time as possible. So benchmark them all. Set up a problem (check the time it takes to do it), solve it (again check the time) and finally compare results with experiment. That's all. After all, you probably bought cars without really understanding how engine works. As long as it drives well and costs within your budget you buy. Use your common sense (and don't listen to salesmen).

Ted,

Thanks for your input.

Yes, electronics cooling does deal a lot with rectangular geometery.

But part of my applications involve round objects ( inlet and outlet tubes of the chassis, pin fins ).

With cartesian meshes, there is a "stair stepping" effect as the cells near the boundary of rounded are objects approximated.

My concern is, for my application ( thermal design of electronic enclosures with rounded passageways ) with a cartesian mesh (which that can only give me brick cells) be suffucient to characterize my geometry?

FYI, when before I bought my car, I did a lot of research and asked a lot of questions ( like I'm doing here ). I do this before I make ANY major purchase. Basically, I would like to know the effectiveness of each feature/option ( ie conformal meshing ) so that I may consider it's value in the overall scheme of things.

That being said, you're right, in the end, I will try each code and baseline the a sample problem and consider how well each code gets the job done.

So to satisfy my curiousity... when is conformal meshing really needed? All things being equal, does a model with curvature need less cells ( therefore more efficiently solved ) with conformal meshing than cartesian meshing to attain the same level of solution accuracy?

Alvin,

You might also look at Flow-3d, one of the sponsers of cfd-online.

That code uses regular (row-column format) meshes, Cartesian or cylindrical. But it also uses something called FAVOR, for

Fractional Area VOlume Representation.

This allows the convenience of regular meshes. Solid blockage to the flow is handled by a clever analysis for cutting corners out of cells to approximate curved non-conforming boundaries. Some formal accuracy is lost but the favorable convergence features of the regular-shaped control volumes are retained. Conjugate heat transfer, which is likely important to you, is possible.

The FAVOR concept was developed under the guidance of C. W. "Tony" Hirt while he was at Los Alamos. References will be listed at

www.flow3d.com

There are a bunch of examples on the web site showing what can be done with a regular mesh if you're really innovative.

Hirt was an innovative force in the CFD world in the 60's, 70's, and 80's. He formed Flow Science in the late 70's. The company is located in Santa Fe, New Mexico.

I have never owned any part of that company, I've never worked for the company. I did work with Hirt for a while at Los Alamos, and I used Flow-3d for a year or so while I was at the Oak Ridge National Lab. I got good personal support from the staff at Flow Science.

Q: All things being equal, does a model with curvature need less cells ( therefore more efficiently solved ) with conformal meshing than cartesian meshing to attain the same level of solution accuracy?

A: most likely it will need fewer cells. However it does not immediately follow that it will be more efficiently (faster?) solved. Besides simple cell count, that depends on a host of other parameters as well. Ditto for solution accuracy.

Hi,

In my view conformal meshing has advantage over Cartesian meshing, but its is the Solver part that is more important. I don't agree with Flowtherms view that conformal will give convergence problem. Cartesian grids do given results similar to confrontal meshing. If you have a benchmark problem you can comapre the results. As far as best conformal meshing software is concerned specially for curved boundaries, in my view StarCD. In my view it provides better solution for heat transfer predictions than other commercial softwares. I am not in field of electronic cooling so I cann't guide you much on it.

Check other vendor like Harvard Thermal, also have a visit to

www.coolingzone.com

Regards

Apurva

Hi

Have a look at C&R tech website www.crtech.com, they market SINDIA/FLUINT, which is another good software for electronic cooling.

Apurva

For electronics cooling applications, accuracy uncertainty due to different meshing schemes is not really dominate factor. There are many other uncertainties (such as power dissipation number, ambient conditions, etc) that dominate the accuracy of your simulation. However, if you really want to know if there is advantage of the un-structured grid, I would say the answer depends on your application. If the dominate heat transfer surfaces are non-planar, un-structured grid may worth the effort.

Apurva,

Thanks for your posting!

Yes, from what I've been hearing, and what I've been reading, I really do think that conformal meshing is advantageous over cartesian meshing.

Just a question though, has do cartesian meshers handle the overlapping cells at the boundary from an error standpoint?

Also, in confromal meshing, does area through which the flux is calculated the true curved area? Or only through a plane that is tangential to the body surface? If it is the latter, then it isn't truely conformal, is it?

Hi

In confrontal meshing the area is not true curved area. Actually what Fluent does is it uses NURBS to approxiamte the curved boundary. The cell boundaries are curved but no exactly to geometry that U porivde. It gives agreeable result.

I don't think any of colling software has overlapping grid.

Do check with other vendors.

Regards

Apurva

Do you know if Star-CD approximates the boundary with a NURBS surface as well?

Once you have a mesh, it is just a mesh and nothing else.Otherwise you have to store a lot of additional information which have to be available for the solver. This is not the case in all the commercial codes around.

And I also dont think that gambit supports true NURBS modelling. But this is just a completely different story.

Hi

Gambit i.e. fluents/icepack pre-processor do support NURBS, I am using an academic version of StarCD, that doesn't have a good grid generator, check for their automated grid generator (I guess known as Samm) you can enquire about it.

Also the is another good electronic cooling software by CFDesign

Regards

Apurva

No, I don't have relation with the mentioned companies. My points are purely from CFD point of view. Just a little correction from some of emails down: unstructured mesh and comforming mesh are 2 different concepts. One can have conformal mesh in a structured code.