In my opinion one of the biggest advantage is that PowerFLOW is really stable and it is almost impossible to get a simulation to blow up (unless you are making big setup mistakes or your geometry has serious issues).

Besides, as you already pointed out, PowerFLOW is a transient (and transient only) code. If you need transient results, a simulation performed with PowerFLOW will take less time (much much less time) than a DES or LES. On the other hand, if you can do with steady results, PowerFLOW will take more time to solve the case than a RANS simulation.

Eventually the mesh: meshing is relatively quick and you can get a lot of cells without any problem. The volume mesh is a cartesian grid made of hexahedra so no issues nor skewness nor stretching with 3D elements.

With all due respect, I would consider that a very grave flaw of Powerflow! It is - in my opinion - mandatory, that there is at least a very limited way of keeping the user from believing results based on stupid/wrong/physically meaningless setups. Any low-scale approximation of a multiscale problem (and that's what that is) has to have some form of stabilization mechanism - and I would be very very suspicious of any code, that never blows up...but that might be the standpoint of an academic, I guess :)

All LES codes are by definition transient, so I might be missing the point....
Quoting their website:
This just sounds like a standard LES approach to me. I admit I don't know anything about SGS closures for Boltzmann, but I'm guessing that it can't be less complicated than for NS.....?

As I said, it blows up if you make mistakes in your setup but it doesn't if your geometry has a small problematic detail in yoru geometry. To me that is a big advantage. It will not let you simulate a domain without any outlet without diverging, of course!


I didn't state otherwise.

Let me rephrase my statement:
If you can do with steady state results then PowerFLOW will (of course) take more time than a RANS to give you your results (it will perform a transient simulation anyway and then you can average your results in order to have something similar to a RANS).

If you need transient results (such as the ones a LES or a DES will provide you) then your case will be solved in much less time than a LES/DES.

understood, thank you. Still, the question of how inexact geometry representation biases your results remains, so it might be better to have a blow up than an inexact solution.





Do you have any comparisons / papers on this? I would be very much interested in seeing the hard facts on this, as I'm very much interested in accuracy and efficiency of numerical methods for LES. It would be nice to have a test case with comparable DOF, LES with a NS based solver (if it is a good one ;) ) and Powerflow, compared against DNS.

Is there a paper on DNS comparison of LB and DNS? I guess there must be, and for low Knudsen numbers, they should indeed give the same results....

Well, your geometry will never be perfect anyway since will still be a discretization of your CAD model. In any case, I'm not talking about a car missing a wheel here. I'm talking about small details that can cause your simulation to blow up. Again, in the case of a car, generally this is not the spoiler of your roof but possibly a small overlapping in your engine bay, ie a detail that will poorly influence your external flow anyways.


Have a look here:
http://exa.com/web_support_2012/exa_tech_pubs.html

Not sure if you can find what you are looking for though… Also on sciencedirect you can find lots of papers by looking for "powerflow".