[CFD Newbie]

Originally Posted by Totalsim

Interesting discussion.
I believe the main advantage is the speed of transient simulations.

I've never done a back to back but I should think there are limitations as I've not heard of F1 clients using it.

FYI - we use DES within OpenFOAM for our transient simulations. For certain bluff applications its necessary for the level of accuracy you need. http://www.youtube.com/watch?v=TXMPE5mtXcw

Thanks TotalSim

[CFD Newbie]

Originally Posted by Paulh

We use both STAR-CCM+ & PowerFlow.

Thanks Paul. Few more questions for you since you use this commercially - sorry for bothering you so much. Greatly appreciate your help.

1. How critical would you say PowerFlow is to your vehicle design process?
2. If you had to use just one software, due to say budget cuts, which one would you/your R&D team pick?
3. How difficult would the transition be from Powerflow to say another, cheaper software (say OpenFoam)?
4. Also curious which software you used before Powerflow and why the displacement. Was the transition easy?

Thanks a lot. Appreciate it.

[Paulh]

1) The Division I'm in uses the CD-adapco products. Two other Divisions use PowerFlow. They would say that PowerFlow is critical to their product development process.
2) I would choose the CD-adapco products because I know how to use them but also because they have more physics in their codes.
3 & 4) As I don't use PowerFLow, I can't make a comment on these two questions.

[CFD Newbie]

Originally Posted by Paulh

1) The Division I'm in uses the CD-adapco products. Two other Divisions use PowerFlow. They would say that PowerFlow is critical to their product development process.
2) I would choose the CD-adapco products because I know how to use them but also because they have more physics in their codes.
3 & 4) As I don't use PowerFLow, I can't make a comment on these two questions.

Thank you very much Paul. Greatly appreciate your help.

[mecobio]

RodriguezFatz, You wrote:

"Solving Boltzmann equation is (physically) only better than the fluid equations"

and later:

"Boltzmann equation is numerically i) faster to solve and ii) more stable"

Well, Boltzmann equation is for gases, and PowerFlow uses lattice Boltzmann (LB), no Boltzmann equation (BE).

Although the LB is the discretization of the BE, more than 90% of the codes use LB models, which are Galilean invariant only at low flow velocities, i.e. the energy flux u^3 is assumed to be neglected.

The solution of the BE, as far I'm aware of, is a pain in the neck.

I do agree with arjun, LB can be unstable too. However, it depends on the the boundary condition implementation, higher flow velocites so that the aforementioned assumtion of neglecting u^3 is not possible any more, and many others.

EXA is stable, I assume, since their LB relies on finite difference schemes (FDS), according to what I have read.

This has a price. Those stabilization techniques can indirectly deteriorate the correct solution.

I have read about people complaining about it elsewhere.

kyle, you are probably getting less accurate values due to those LB assumptions above and FDS implementations.

Can any of you be so kind to tell me the Re numebrs you are using?
That answer will explain why you are getting what you are getting

Thanks

[RodriguezFatz]

I am sorry for my language skills, but I think you misunderstood my posts.
It was ment like: "The only reason I can think of to solve for BE is, that it would be better in ..." (which I don't know, if that is the case)
I did not try to say that it is better in ...

sorry

[iti86]

Hi guys,

Here are some answers to your questions :

1 - Is the LBM faster than a NS scheme?
Yes, for the same order. The usual LBM codes use a second order scheme (time and space) and require 588 operations per iteration. A NS code based on a centered finite difference scheme of second order in space and third in time requires 711 operations per iteration.

2- Is the LBM more accurate than a NS scheme?
Yes, for the same order. In fact, there are two types of errors : disspation and dispersion errors. Dissipation errors reduce the amplitude of the acoustic wave. Dispersion errors change its wavelength.
For a second order LBM (space & time), the dissipation error is lower than a 6th order NS scheme. The dispersion error however is seomewhere between a 3rd and fourth order NS scheme.

3- Is the LBM more stable than NS schemes?
NO. Since it has less dissipation, the method is naturally less stable.

By the way, PowerFLow does NOT perform an LES. They sell their code as an LES or VLES (Very Large Eddy Simulation) solver, but it is not true. Powerflow uses the k-epsilon RNG model coupled with wall-law (check their manual). Their wall-law however is a little bit more sophisticated than StarCCM's since it accounts for adverse pressure gradients.

I hope my answer helps you!

[arjun]

Originally Posted by iti86

2- Is the LBM more accurate than a NS scheme?
Yes, for the same order. In fact, there are two types of errors : disspation and dispersion errors. Dissipation errors reduce the amplitude of the acoustic wave. Dispersion errors change its wavelength.
For a second order LBM (space & time), the dissipation error is lower than a 6th order NS scheme. The dispersion error however is seomewhere between a 3rd and fourth order NS scheme.

For the acoustics the problem with LBM that Powerflow implements is that it is isothermal LBM (D3Q19 or D2Q9 Lattices). So I really doubt that for acoustics it could be as accurate as some code that accounts for thermal variations too.

Originally Posted by iti86

By the way, PowerFLow does NOT perform an LES. They sell their code as an LES or VLES (Very Large Eddy Simulation) solver, but it is not true. Powerflow uses the k-epsilon RNG model coupled with wall-law (check their manual). Their wall-law however is a little bit more sophisticated than StarCCM's since it accounts for adverse pressure gradients.

This is not that straightforward. The bounceback model has one draw back that there is some slip associated with it. So even if acount for adverse pressure gradients, it loses out on no slip part (which is much more important aspect).

[iti86]

Arjun,

I agree that accounting for a non-isothermal fluid improves the accuracy of the acoustic simulation. However, what I am saying is that at the same order, the LBM is more accurate than solving the compressible NS equations.

[mecobio]

Originally Posted by iti86

Hi guys,

Here are some answers to your questions :

1 - Is the LBM faster than a NS scheme?
Yes, for the same order. The usual LBM codes use a second order scheme (time and space) and require 588 operations per iteration. A NS code based on a centered finite difference scheme of second order in space and third in time requires 711 operations per iteration.

Well, it depends on the definition of "usual LBM" codes.
There are several LB methods or approaches.
There is a comparison among two of these LBM on:

Mathematics and Computers in Simulation, Volume 84, October 2012, Pages 26–41,
http://www.sciencedirect.com/science...78475412001966

(downloadable from http://www.scribd.com/doc/132100792/...ion-approaches )

So, I suggest to specify which LBM you are refering to, and also:
Do you know any publication that reinforces your statements (the one with 588 operations per iteration while NS with 711 operations per iteration)?
Thanks.

Originally Posted by iti86

2- Is the LBM more accurate than a NS scheme?
Yes, for the same order. In fact, there are two types of errors : disspation and dispersion errors. Dissipation errors reduce the amplitude of the acoustic wave. Dispersion errors change its wavelength.
For a second order LBM (space & time), the dissipation error is lower than a 6th order NS scheme. The dispersion error however is seomewhere between a 3rd and fourth order NS scheme.

Here, it should be specified the NS equation you are refering to, as well.
The "usual" LBM is NOT complete Galilean invariant, which implies that the flow velocity has to be decreased, e.g. flow velocity (in lattice units) so that u^3 can be neglected. That is to avoid flow velocity errors.
Hence, the "usual" LBM does not get the "full" NS equation.
Let alone that 3D lattice sets used in LBM are reduced models of the D3Q3^3 (e.g. D3Q19), and thus even worse than the NS equation. Pelase, would you mind to cite some reference to confirm your above statements? Thanks.

I think those dissipation and dispersion errors should be inserted with the limitation of the "usual" LBM's context.

Originally Posted by iti86

3- Is the LBM more stable than NS schemes?
NO. Since it has less dissipation, the method is naturally less stable.

Well, instability in LBM is due to existence of negative populations.
Bottom line, instabilites are a result of that the "user" is trying to test the LBM beyond the method's own range, assuming the code is well written and provides the correct output/warnings.

Originally Posted by iti86

By the way, PowerFLow does NOT perform an LES. They sell their code as an LES or VLES (Very Large Eddy Simulation) solver, but it is not true. Powerflow uses the k-epsilon RNG model coupled with wall-law (check their manual). Their wall-law however is a little bit more sophisticated than StarCCM's since it accounts for adverse pressure gradients.

I hope my answer helps you!

I don't have a clue about PowerFlow, but this is read in their user guide 4.0 (the latest code is 4.4):

"PowerFLOWis based on the principle of performing very large eddy simulations

(VLES) that directly simulate resolvable flow scales while modelling unresolved
scales. There are two essential features of a VLES approach: time accuracy and an

appropriate wall model".

[/QUOTE]

Originally Posted by iti86

the LBM is more accurate than solving the compressible NS equations

Well, the density has to change in order to get compressible LB approach, right?. However, the "usually" LB approach is isothermal incompressible flow-velocity limited (so that u^3 can be neglected) method, so bottom line, the above statement can be questioned. Do you know any publication about it? Please, add it. Thanks.

[mecobio]

Originally Posted by iti86

Hi guys,

Here are some answers to your questions :

1 - Is the LBM faster than a NS scheme?
Yes, for the same order. The usual LBM codes use a second order scheme (time and space) and require 588 operations per iteration. A NS code based on a centered finite difference scheme of second order in space and third in time requires 711 operations per iteration.

I have noticed now that the number you mentioned of 711 for N-S and 588 for LBM comes from table 2, page 1068 from the paper:
http://www.sciencedirect.com/science/article/pii/S002199910800538X

However, that paper is solely based on computational aeroacoustic (CAA) simulations, which is a PARTICULAR model, not a general simulation without any included particular model, right?

[iti86]

Hello Mecobio,

I can see you've been busy this weekend :-)

IMHO, there is no particular model associated with CAA. As long as your numerical scheme is accurate enough to capture the acoustic sources and to propagate the acoustic waves without too much dissipation or dispersion, then it is suited for aeroacoustic as well as aerodynamic computations (provided you have a mesh fine enough for the CAA...)

Regarding the VLES, a good explanation of the modeling behing it is provided in this paper : http://www.ihs.uni-stuttgart.de/file...n/v2003_11.pdf

The VLES is based on the k-epsilon model with some added filtering techniques based on the grid spacing and time step. Although it is called VLES, it does not resemble the LES where the actual NS equations are directly filtered and the subgrid scales are modeled.
LES --> Filtered NS
VLES --> NS equations , turbulence modeling (k-Epsilon), and then filtering.

Please feel free to correct me if I’m wrong!

[arjun]

Originally Posted by iti86

Hello Mecobio,

I can see you've been busy this weekend :-)

IMHO, there is no particular model associated with CAA. As long as your numerical scheme is accurate enough to capture the acoustic sources and to propagate the acoustic waves without too much dissipation or dispersion, then it is suited for aeroacoustic as well as aerodynamic computations (provided you have a mesh fine enough for the CAA...)

Regarding the VLES, a good explanation of the modeling behing it is provided in this paper : http://www.ihs.uni-stuttgart.de/file...n/v2003_11.pdf

The VLES is based on the k-epsilon model with some added filtering techniques based on the grid spacing and time step. Although it is called VLES, it does not resemble the LES where the actual NS equations are directly filtered and the subgrid scales are modeled.
LES --> Filtered NS
VLES --> NS equations , turbulence modeling (k-Epsilon), and then filtering.

Please feel free to correct me if I’m wrong!

First , thanks for the paper.
I have small doubt, the paper you mentioned shows that you need to solve two equations for k and epsilon. I do not see powerflow solving these two equations. Or do they solve them. If they do how are they doing it, finite volume, finite element or something special? Also would solving these two quantities by means of transport equation kill the efficiency of their code because their main strength lies in not solving these two equations the way other solvers dp
If I understand correctly

[mecobio]

Originally Posted by iti86

Hello Mecobio,

I can see you've been busy this weekend :-)

IMHO, there is no particular model associated with CAA. As long as your numerical scheme is accurate enough to capture the acoustic sources and to propagate the acoustic waves without too much dissipation or dispersion, then it is suited for aeroacoustic as well as aerodynamic computations (provided you have a mesh fine enough for the CAA...)

Regarding the VLES, a good explanation of the modeling behing it is provided in this paper : http://www.ihs.uni-stuttgart.de/file...n/v2003_11.pdf

The VLES is based on the k-epsilon model with some added filtering techniques based on the grid spacing and time step. Although it is called VLES, it does not resemble the LES where the actual NS equations are directly filtered and the subgrid scales are modeled.
LES --> Filtered NS
VLES --> NS equations , turbulence modeling (k-Epsilon), and then filtering.

Please feel free to correct me if I’m wrong!

Thanks for your reply. First, it took me time (around 2 seconds) to find out the anachronism IMHO = In My Honest Opinion, right? A sign I'm getting older So, bottom line, PowerFlow 4.0 had VLES, not LES, right? No sure about 4.4, which is the latest, as I cannot find the user guide for that version.

Please, can any body put the POwerFlow user guide for 4.4 here? Thanks.

On the other hand, Xflow has "Turbulence models: state-of-the-art LES model (Smagorinsky, dynamic Smagorinsky, Wall-Adapting Local-Eddy) and Spalart-Allmaras (for consistency in 2D)".

I have worked with VLES before (not anymore), which is sometimes called as "the poor man's LES"

Thanks

[iti86]

Mecobio,

IMHO = In my humble opinion

Regarding Powerflow, there is no need to upload the user manual (although i'd appreciate it!). If you take a look at Exa's website (http://www.exa.com/powerflow.html), it says :

"State-of-the-art Very Large Eddy Simulation (VLES) turbulence model and turbulent boundary layer models inherently capture the transient nature of fluid flows."

Arjun, concerning the k-epsilon equations, i don't how Exa solves them but I heard something about adding two populations for k and epsilon... not sure though. I'll try to look deeper into it and as soon as i get an answer, i'll post it here!

[mecobio]

Originally Posted by iti86

Mecobio,

IMHO = In my humble opinion

Regarding Powerflow, there is no need to upload the user manual (although i'd appreciate it!). If you take a look at Exa's website (http://www.exa.com/powerflow.html), it says :

"State-of-the-art Very Large Eddy Simulation (VLES) turbulence model and turbulent boundary layer models inherently capture the transient nature of fluid flows."

Arjun, concerning the k-epsilon equations, i don't how Exa solves them but I heard something about adding two populations for k and epsilon... not sure though. I'll try to look deeper into it and as soon as i get an answer, i'll post it here!

Then, based on the above, the Xlfow seems to be better for turbulence simulations than PowerFlow, as LES is supposed to be better than VLES.
Interesting!!

PS: Anyone with an already Xflow project to download, so I can test it with the Xflow code I have access? Thanks

[arjun]

Originally Posted by iti86

Arjun, concerning the k-epsilon equations, i don't how Exa solves them but I heard something about adding two populations for k and epsilon... not sure though. I'll try to look deeper into it and as soon as i get an answer, i'll post it here!

Now that you made me curious i will also look into it when i get time. If I find something , will post here.

[mecobio]

I have done a little of a homework here and I found out interesting differences between PowerFLow and Xflow:

To my knowledge, PowerFlow is based on standard derivation and matching convective moments.

On the other hand, Xflow is relying on central moment space, which goes under the name "cascade LB" and it is based on the paper of
http://pre.aps.org/pdf/PRE/v73/i6/e066705
and subsequent works. That is, the Xflow is based on an AB Initio derivation, where the relaxation is of the central moments instead of convective moments, and sometimes goes under the name of Cascade Lattice Boltzmann (CLB). It is "said" that CLM is a sort of MRT construction.

The above might explain why Xflow can have a little bit of an edge to simulate turbulent flows, over PowerFlow which only has VLES (if we believe in the so called "cascade LB" method's self-claimed advantages).
Xflow has LES.

It will be interesting to see both codes tested under the same conditions.


PS: Hopefully, this information can be useful for those with money and time to decide which code to choose.

[iti86]

Arjun,

Here's a link explaining how the VLES is implemented in PowerFlow

http://exa.com/core-technology.html

Check the "Turbulence modeling using CFD section".

[mecobio]

Originally Posted by iti86

Arjun,

Here's a link explaining how the VLES is implemented in PowerFlow

http://exa.com/core-technology.html

Check the "Turbulence modeling using CFD section".

You can also see page 10 in http://hiliftpw.larc.nasa.gov/Worksh...PW2-holman.pdf where a theoretical comparison is made between VLES and LES.