CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > Main CFD Forum

XFlow

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   July 1, 2013, 13:08
Default
  #21
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 369
Rep Power: 10
arjun is on a distinguished road
Quote:
Originally Posted by mecobio View Post
Excuse me, with all respect, but LB need indeed mesh.
Some people use LB models based Cartesian, orther on spherical, etc.
Some people use refinned meshes in LB too.
It is at those mesh-nodes where collisions happen, usually modelled by the SRT BGK/Wallander model or MRT.
So, where the meshless-myth come from?!
You are correct. In fact the dependency of LBMs on mesh is one big obstacle in path of LB Methods. Also trying to get it to work on general polyhedrals is not easy, specially when viscosity and delta T depend on type of streaming which in turn depend on mesh type. (like classical streaming can not be applied to general meshes directly).
arjun is offline   Reply With Quote

Old   July 1, 2013, 13:30
Default
  #22
New Member
 
An
Join Date: Feb 2011
Posts: 27
Rep Power: 6
mecobio is on a distinguished road
Quote:
Originally Posted by arjun View Post
You are correct. In fact the dependency of LBMs on mesh is one big obstacle in path of LB Methods. Also trying to get it to work on general polyhedrals is not easy, specially when viscosity and delta T depend on type of streaming which in turn depend on mesh type. (like classical streaming can not be applied to general meshes directly).
Why do you want to work on (general) polyhedrals when dealing with LB?

You wrote: "when viscosity and delta T depend on type of streaming which in turn depend on mesh type"

You have flow velocity u, in lattice units, which is limited, up to a value depending on the model. Then, you have a characteristic length L, so the only way to increase Reynolds number is to decrease the kinematic viscosity nu=(tau-1/2)*R*T*(delta T). With Delta T =1 fixed, decrease the relaxation time tau, which might lead to instability.

What do you mean with "streaming which in turn depend on mesh type"?

If you meant the stream step, it is supposed to be from one one to another, and that step is supposed to be fixed. If the boundary is located between nodes (due to the geometry your gas is inmersed) you use interpolation (there is a paper from the people from Exa) and it guarantees conservation laws.

There are LB models relying on (around) Cartesian, Spherical, etc, but there is a price to pay when dealing with such non-Cartesian grid.

Again, what is the point to use (general) polyhedrals when dealing with LB?
mecobio is offline   Reply With Quote

Old   July 1, 2013, 16:04
Default
  #23
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 369
Rep Power: 10
arjun is on a distinguished road
As to why I want to work with LBM's with general polyhedrals is for the reasons that they represent the shape much better compared to the approach that involves cube. Another approach is the surfels that exa use to represent bodies.
Quote:
Originally Posted by mecobio View Post
You have flow velocity u, in lattice units, which is limited, up to a value depending on the model. Then, you have a characteristic length L, so the only way to increase Reynolds number is to decrease the kinematic viscosity nu=(tau-1/2)*R*T*(delta T). With Delta T =1 fixed, decrease the relaxation time tau, which might lead to instability.

What do you mean with "streaming which in turn depend on mesh type"?

If you meant the stream step, it is supposed to be from one one to another, and that step is supposed to be fixed. If the boundary is located between nodes (due to the geometry your gas is inmersed) you use interpolation (there is a paper from the people from Exa) and it guarantees conservation laws.
Well the factor of 1/2 in the viscosity formula you have is due to the streaming step of classic lattice boltzmann (which you said is fixed). But as soon as you move into finite difference and finite volume that term (1/2) disapears. This is why I said that streaming depends on the mesh, on certain meshes you could apply the streaming that you mentioned. On others you do it by using finite difference method or finite volume methods or using interpolations etc. There is a good paper explaining viscosites in each approach, give me time I will find the title of it. I have a hard copy of it.
arjun is offline   Reply With Quote

Old   July 1, 2013, 16:20
Default
  #24
New Member
 
An
Join Date: Feb 2011
Posts: 27
Rep Power: 6
mecobio is on a distinguished road
Quote:
Originally Posted by arjun View Post
As to why I want to work with LBM's with general polyhedrals is for the reasons that they represent the shape much better compared to the approach that involves cube. Another approach is the surfels that exa use to represent bodies. Well the factor of 1/2 in the viscosity formula you have is due to the streaming step of classic lattice boltzmann (which you said is fixed). But as soon as you move into finite difference and finite volume that term (1/2) disapears. This is why I said that streaming depends on the mesh, on certain meshes you could apply the streaming that you mentioned. On others you do it by using finite difference method or finite volume methods or using interpolations etc. There is a good paper explaining viscosites in each approach, give me time I will find the title of it. I have a hard copy of it.
Yes, but by using finite difference schemes (FDS) and/or finite volume methods (FVM) you are violating the main LB idea. N-S based CFD and LB based CFD have difference physolophies. Any way, please, give me the paper title and journal number, page, etc so I can download it myself (when it suits you ). Thanks.

PS: BY the way, are you working at X-Flow, Exa or? Just curious
mecobio is offline   Reply With Quote

Old   July 1, 2013, 16:26
Default
  #25
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 369
Rep Power: 10
arjun is on a distinguished road
Quote:
Originally Posted by mecobio View Post
Yes, but by using finite difference schemes (FDS) and/or finite volume methods (FVM) you are violating the main LB idea. N-S based CFD and LB based CFD have difference physolophies. Any way, please, give me the paper title and journal number, page, etc so I can download it myself (when it suits you ). Thanks.

PS: BY the way, are you working at X-Flow, Exa or? Just curious
I am working at CD - Adapco. I am just interested in LBM for fun of it. I usually take up something that is not related to my work and learn it. Spent time with meshless methods but LBM I think have more appeal. Tomorrow I will give you the paper's name. About stablity if you are interested, I find his work very interesting. http://www.rbrownlee.org.uk/index.php
arjun is offline   Reply With Quote

Old   July 2, 2013, 00:35
Default
  #26
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 369
Rep Power: 10
arjun is on a distinguished road
Quote:
Originally Posted by mecobio View Post
Yes, but by using finite difference schemes (FDS) and/or finite volume methods (FVM) you are violating the main LB idea. N-S based CFD and LB based CFD have difference physolophies. Any way, please, give me the paper title and journal number, page, etc so I can download it myself (when it suits you ). Thanks.

PS: BY the way, are you working at X-Flow, Exa or? Just curious
I found the name of the paper "Viscosity of finite difference lattice Boltzmann models" by Victor Sofoneaa, Robert F. Sekerkab, I have a pdf of it, if you want pm me your email i will mail it to you. Its a good read.
arjun is offline   Reply With Quote

Old   July 2, 2013, 05:45
Default
  #27
New Member
 
An
Join Date: Feb 2011
Posts: 27
Rep Power: 6
mecobio is on a distinguished road
Quote:
Originally Posted by arjun View Post
I found the name of the paper "Viscosity of finite difference lattice Boltzmann models" by Victor Sofoneaa, Robert F. Sekerkab, I have a pdf of it, if you want pm me your email i will mail it to you. Its a good read.

Thanks, I know that paper, and I have it. Thanks. The issue is that the use of FDS and/or FVM insert numerical dissipation, which is not desirable. Im aware of that many people are using so in LBM. I don't follow that path.


On another matter:
According to Exa, their constraints is Re < 10,000
SOURCE: http://exa.com/core-technology.html

1000 < Re < 10 000 can be reached for grid resolution far less than 10^3, as seen in fig 6 in this paper
Mathematics and Computers in Simulation 84 (2012) 26–41 (http://www.sciencedirect.com/science...78475412001966).
Interesting that paper compares two different LB models and they give similar results.

However, no information if provided about their implementation in Exa or X-flow.

So my main question are:
1-Is Exa using finite difference schemes, volume method, or in the implementation of the LBM?
2-What about X-flow?

The only thing I know is that X-flow uses "MRT and refined-schemes". Beyond that, not sure about their implementations.

Thanks in advance
mecobio is offline   Reply With Quote

Old   July 2, 2013, 06:20
Default
  #28
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 369
Rep Power: 10
arjun is on a distinguished road
Thanks for the link. I am not well aware with x-flow but with PowerFlow i have fairly good idea. They are using single relaxation model, with the classical streaming ( fbar = f) so they are not using finite difference though it seems have published something separately (not in powerflow). I have some papers they published that outlines what they do. I can try to scan them into pdfs if you are really interested. They have this series of voxels of different sizes, typically ratio is 2 in length dimention. Number of solves vary on each size, so if coarse cell get N updates, finer children of it would get 2N updates (or delta T is half of coarser one). To join these fine and coarse cells they use liner interpolations. Currently their mesh is once generated then fixed during the iterations but they are looking into dynamic meshing and mesh movements (inferred from their job postings). Also about their limit of Re=10000, I think they are saying that uptil this Re they could run it without any turbulence model, after that if mesh is not fine enough they would need turbulence model. which actually should bail them out of stability issues too. My very personal opinion (which I can not verify or prove) is that Powerflow uses some kind of entropic update to keep the solver stable even on higher Reynolds numbers. Also I believe that speed of classic lattice boltzmann is main reason they are not using other LBMs, even though they are fully aware of developments of various other types of LBMs. (again this is an opinion).
Quote:
Originally Posted by mecobio View Post
Thanks, I know that paper, and I have it. Thanks. The issue is that the use of FDS and/or FVM insert numerical dissipation, which is not desirable. Im aware of that many people are using so in LBM. I don't follow that path.


On another matter:
According to Exa, their constraints is Re < 10,000
SOURCE: http://exa.com/core-technology.html

1000 < Re < 10 000 can be reached for grid resolution far less than 10^3, as seen in fig 6 in this paper
Mathematics and Computers in Simulation 84 (2012) 26–41 (http://www.sciencedirect.com/science...78475412001966).
Interesting that paper compares two different LB models and they give similar results.

However, no information if provided about their implementation in Exa or X-flow.

So my main question are:
1-Is Exa using finite difference schemes, volume method, or in the implementation of the LBM?
2-What about X-flow?

The only thing I know is that X-flow uses &quot;MRT and refined-schemes&quot;. Beyond that, not sure about their implementations.

Thanks in advance
arjun is offline   Reply With Quote

Old   July 2, 2013, 08:30
Default
  #29
New Member
 
An
Join Date: Feb 2011
Posts: 27
Rep Power: 6
mecobio is on a distinguished road
I'm aware of the paper you are refering to by people of Exa.
But that construction is not local, unfortunatelly.

If you are refering to Entropic LB (ELB) model, well, the paper
Mathematics and Computers in Simulation 84 (2012) 26–41 (http://www.sciencedirect.com/science...78475412001966).
shows that in 2D, the ELB is no better than the other one.
There are other papers where 1D comparisons are made, and still the ELB is no better thatn the other one in 1D shocktube.
Hence, there is no evidence that in 3D the ELB would be better, at least no independent people have tested anyway.
These comparison are made by independent people, while the ELB creators still claim otherwise.
There is an ongoing controversy seen on http://pre.aps.org/abstract/PRE/v84/i6/e068701, but that is another story.
Maybe EXA is using positivity rule, to maintain positive populations, to avoid instabilities.
Not sure. There is always a price, e.g. in the accuracy, since the issue is the boundary conditions, as shown in the aforementioned paper.

There are hidding tricks at the boundaries, as usual.

The question is: If PoweFlow is under/over predicting turbulent flows (according to some complains, never sure,
but it what it is said in some forums) then:

Is that due to their turbulent models?
(It is well known than RANS can even given the wrong direction of the swirl, due to the modeling).

What about the performance of PowerFlow and X-flow compared to other NS CFD codes?

Are LB CFD codes faster? How faster? Two fold?, Three fold?, etc

One thing is for sure: The car industry has NOT rejected the use of LB in CFD, as seen here: http://jobs.gm.com/job/Warren-Vehicl...48088/2687218/



Quote:
Originally Posted by arjun View Post
Thanks for the link. I am not well aware with x-flow but with PowerFlow i have fairly good idea. They are using single relaxation model, with the classical streaming ( fbar = f) so they are not using finite difference though it seems have published something separately (not in powerflow). I have some papers they published that outlines what they do. I can try to scan them into pdfs if you are really interested. They have this series of voxels of different sizes, typically ratio is 2 in length dimention. Number of solves vary on each size, so if coarse cell get N updates, finer children of it would get 2N updates (or delta T is half of coarser one). To join these fine and coarse cells they use liner interpolations. Currently their mesh is once generated then fixed during the iterations but they are looking into dynamic meshing and mesh movements (inferred from their job postings). Also about their limit of Re=10000, I think they are saying that uptil this Re they could run it without any turbulence model, after that if mesh is not fine enough they would need turbulence model. which actually should bail them out of stability issues too. My very personal opinion (which I can not verify or prove) is that Powerflow uses some kind of entropic update to keep the solver stable even on higher Reynolds numbers. Also I believe that speed of classic lattice boltzmann is main reason they are not using other LBMs, even though they are fully aware of developments of various other types of LBMs. (again this is an opinion).
mecobio is offline   Reply With Quote

Old   July 2, 2013, 15:36
Default
  #30
Senior Member
 
Arjun
Join Date: Mar 2009
Location: Nurenberg, Germany
Posts: 369
Rep Power: 10
arjun is on a distinguished road
I mostly agree with your assesment about stablity. I do not have much opinion of it at the moment because though i have seen papers showing that it is stable and all , but I do not have first hand experience of it. By this I mean, like for example take starccm type solver, I have lots of experience with it and can say lots of things with confidence about its behaviour. But since I have never implemented entropic version myself and have not tried it on complicated cases, really can not conclude anything. Also about PowerFlow over predicting turbulence etc, it is really difficult thing to comment on, as it depends on lots of factors, their turbulence model being cardinal of them. About the performance, which is again tricky, for the same mesh sizes one it seems Powerflow would be faster than codes like fluent and starccm+ . It could be same speed or less efficient (not the word efficient, some solvers might take larger time step) than well implemented direct solvers based codes (that use FFT etc for pressure). On the other hand, one could run calculations with much smaller mesh sizes and also run steady runs to get results much faster than powerflow with say starccm+ . For transient problems related to external aerodynamics , it sounds most attractive.
Quote:
Originally Posted by mecobio View Post
I'm aware of the paper you are refering to by people of Exa.
But that construction is not local, unfortunatelly.

If you are refering to Entropic LB (ELB) model, well, the paper
Mathematics and Computers in Simulation 84 (2012) 26–41 (http://www.sciencedirect.com/science...78475412001966).
shows that in 2D, the ELB is no better than the other one.
There are other papers where 1D comparisons are made, and still the ELB is no better thatn the other one in 1D shocktube.
Hence, there is no evidence that in 3D the ELB would be better, at least no independent people have tested anyway.
These comparison are made by independent people, while the ELB creators still claim otherwise.
There is an ongoing controversy seen on http://pre.aps.org/abstract/PRE/v84/i6/e068701, but that is another story.
Maybe EXA is using positivity rule, to maintain positive populations, to avoid instabilities.
Not sure. There is always a price, e.g. in the accuracy, since the issue is the boundary conditions, as shown in the aforementioned paper.

There are hidding tricks at the boundaries, as usual.

The question is: If PoweFlow is under/over predicting turbulent flows (according to some complains, never sure,
but it what it is said in some forums) then:

Is that due to their turbulent models?
(It is well known than RANS can even given the wrong direction of the swirl, due to the modeling).

What about the performance of PowerFlow and X-flow compared to other NS CFD codes?

Are LB CFD codes faster? How faster? Two fold?, Three fold?, etc

One thing is for sure: The car industry has NOT rejected the use of LB in CFD, as seen here: http://jobs.gm.com/job/Warren-Vehicl...48088/2687218/
arjun is offline   Reply With Quote

Old   August 14, 2013, 04:18
Default Anybody has any experience with XFlow in FSI
  #31
New Member
 
王国斌
Join Date: Jun 2010
Location: beij,chongqing,Chengdu
Posts: 22
Rep Power: 7
wangguobin14 is on a distinguished road
Send a message via ICQ to wangguobin14 Send a message via MSN to wangguobin14
Anybody has any experience with XFlow in FSI
nect to solid solver....
Trying to connect to solid solver.... ,what's mean ,searching for nastran solid.exe

Quote:
Originally Posted by Yaping View Post
Hello,

Anybody has any experience with XFlow?

Thanks,

Yaping
wangguobin14 is offline   Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
XFLOW Does anyone know something about that? eleazar Main CFD Forum 7 August 12, 2013 11:33
X Flow CFD? Will you use it? HarshThapar Main CFD Forum 9 August 12, 2013 11:22
Validations? -- Heat X-fer -- Cyl in Xflow Glenn Main CFD Forum 1 March 5, 2013 05:06
NextLimit Xflow flömaster B Main CFD Forum 13 August 26, 2011 14:42


All times are GMT -4. The time now is 02:53.