Lattice Boltzmann Approach

philippose · September 8, 2007, 05:04

Hello and a good morning!

I recently came across a CFD simulation software known as "PowerFlow" from the company "Exa".

This software claims to perform Computational Fluid Dynamics using a very different approach from the usual discrete Navier Stokes solution. The approach they use is called the "Lattice Boltzmann Approach".

It would be interesting to know what the exact advantages / disadvantages of this approach is... does anyone have any experience with this approach?

Have a nice weekend!

Philippose

grtabor · September 8, 2007, 10:29

I've heard of the lattice boltzmann approach (and Exa as well) - basically they are solving at a discrete particle level representing the flow by a discrete set of states at each point. I think that the solution thus uses integer rather than real arithmetic and so is correspondingly faster - although this is presumably offset by the need to provide considerably more vertices to generate the same accuracy of solution.

My feeling is (and I havn't looked at it in any great detail) it would only really make sense if you were prepared to build specialist hardware for the problem - which was a possibility 10 years ago but didnt really materialise. Put another way; Exa has been talking their method up for about the last 10 years, and they still havn't made a real inroad into the CFD market, so I can't believe that there is a great benefit from the method.

Gavin

pannala · September 8, 2007, 19:18

I am no expert at this but Lattice Boltzmann Method does not operate on integers like the earlier Lattice Gas Methods. LBM is very efficient if you are looking for solutions of incompressible NS equations in the mesoscale configurations with complex geometries like flow through a porous media or flow over/through collection of particles. The main advantage is that the non-linear convection term gets absorbed into the collision term and treated using a relaxation time. Theoretically you can recover INS from current versions of LBM with certain lattice structures (D2Q9 and D3Q19). Succi's book would serve as a good introduction for LBM and there are open source solutions available like Vladymir (http://cui.unige.ch/~latt/vladymir/download.html) if you want to download and try it out for yourself.

Hope this helps,

Sreekanth

sradl · September 9, 2007, 02:14

Dear Foamers,

there is an open plattform that is currently using LBM:

http://www.openlb.org/

I've not tried this one, but i thing they are pretty at the beginning.

For those who are interested, maybe the publications of Jos Derksen, who is using a very efficient FORTRAN LBM code, give a impression on the cababilities of LBM:

http://www.tnw.tudelft.nl/live/pagina.jsp?id=0ddabf98-358a-47a0-81f8-e84856bb24f d&lang=en

(Prof. Derksen has moved to Alberta as far as i know, so maybe this page will be killed in the near future).

You will see that LBM and OpenFOAM will not fit together very well, because in LBM you have structured, cubic grids that will be incompatible with all current mesh classes in OF and totally different equations to solve (maybe that is the smallest problem, because they are quite simple (e.g. no poisson solver needed) :-) ).

I've also started some LBM work on the basis of the dersken code. What we do is LES of a whole bioreaktor with about 10-20 Mio. cells on up to 10 parallel machines. Its pretty fast...

br
Stefan Radl

msrinath80 · September 9, 2007, 02:25

Yup he is moving to my department. I had the opportunity to meet with him twice last year when he came for presentation/interviews. I doubt that he will discontinue work on LBA.

philippose · September 9, 2007, 04:17

Hello again,

Thanks a lot for the information regarding the Lattice Boltzmann approach.... looks like it is a known, and active area even though it has not hit the mainstream yet.

I tend to agree with Gavin, in that if it was a method applicable to general cases as are the current CFD techniques, it would be more popular.

On the other hand, I was reading through some papers available online, and also happened to stumble upon openLB.... it has been claimed to be faster in some cases (the comparison was with CFX).... particularly laminar incompressible flow. Also, multi-phase simulations are inherently handled within the solution, without requiring special treatment like in the normal CFD methods.

It is definitely not an "only integer" solution technique, quite contrary, it is a very memory and processor hungry method, which is why there have been lots of efforts to parallelise it (apparently this method is inherently easier to parallelise).

As for its compatibility with OpenFOAM... I am not aware of the numerical solution requirements of the Lattice Boltzmann method yet, but the mesh is definitely not the limitation. OpenFOAM can handle structured cubic grids as well as unstructured polyhedral meshes....! After all... a cubic grid is a sub-set of an unstructured polyhedral mesh.

I have a feeling, that it should be possible to implement the Lattice Boltzmann method using OpenFOAM without any change to the base code... a well thought out top-level solver should do the trick.... unless there are some very special requirements from the mathematical side, for LBM.

I wish you all a wonderful Sunday!

Philippose

vega · September 9, 2007, 22:19

I have had chance to attend training of Exa's Powerflow at their office recently. Here are my views of this software and the company
1. Powerflow's main strength, for industrial applications, is the ease of model setup. Those who
have experience of working in industry (specially automotive) will agree that CAD cleanup and meshing takes the most amount of time.
2. Powerflow runs only in the transient mode so it does take lot more resources than conventional (Navier-Stokes) NS code. This is true specially for cases where steady state simulation is sufficient.
3. Powerflow does better for low Mach number compressible simulations (10 - 40 m/sec) than the NS codes. From my hands-on experiences in the area of aeroacoustics of automotive sunroof buffeting,
I can say that two of the most popular commercial NS codes have not matched the accuracy of Powerflow.
4. One of the point mentioned in the training was that because of the Lagrangian approach, software is better suited for parallelization. Do not have any personal experience to compare its speed-up rate vs NS codes

The company seems to be doing quite well and has expanded significantly over past 2 years or so. It is being used at several automotive OEMs and NASCAR teams for external aero simulations. I think it would be 3rd most used CFD software in the US auto industry after Fluent and Star-CD

Big downside i see is as mentioned earlier that for situations where steady-state simulations are sufficient, Powerflow becomes too expensive.

--
R

sradl · September 10, 2007, 01:34

Dear Philippose,

you are wrong regarding the grid: it is not that easy as you believe, because you have additional troubles when it comes to boundary conditions. If you want to leave "Lego"-land, i.e. not only cubic cells on the boundary, you need special algorithms to get that right in your collision step of the LBM.
Of course: a cubic grid can be easily handeled by OF - the question is if this is satisfactory for your boundary conditions you want to apply.

So I disagree: it will not be possible with a top-level solver; you need much more on the grid level and for the boundary condition in addition to the current OF source code. Efficient grid handling (you need several Mio cells!!) and boundary conditions will be a lot of work to implement. Of course it will be possible; but it is not that simple as you mentioned.

br
Stefan

sradl · September 10, 2007, 01:41

Dear Srinath,

my best wishes to you and Prof. Derksen! I also met him last year in italy and his presentations are great fun. I had the change to get parts of his LBM code and its really a nice piece of code.

Once again greetings to you and Prof. Derksen from his friends in austria (Prof. Khinast group).

br
Stefan Radl

pannala · September 10, 2007, 06:31

FYI, there are extensions to LBM for unstructured grids based on finite volumes. If you are interested, please look at the following reference:

S. Ubertini, G. Bella, and S. Succi, Lattice Boltzmann method on unstructured grids: Further developments, PHYSICAL REVIEW E 68, 016701, 2003

The main limitations of LBM approach:

a) Compressible flows - need higher order terms for collisions and thus erodes into the simplicity/efficiency.
b) Strong coupling of density to temperature variations - this would also require higher order terms.

The algorithm is extremely simple so that one can easily parallelize the code. Only the propagation step needs information exchange with the neighbors and thus regular domain decomposition can be used. In complex geometries, the communication can be ordered so that one can have a well-balanced algorithm. In addition, since the propagation and relaxation steps are straight forward, one can possibly make them cache aware so that one can get very good performance.

Here is a link to a very simple LBM code in case you want to see the simplicity of the algorithm: http://www.ccrl-nece.de/lba/.

Cheers,

Sreekanth

bagogan · February 2, 2012, 09:37

dear all,
I start working with LBM to simulate couette -Taylor flow, the inner cylinder is animated with angular velocity.
I read some articles about this subject but i'm very confused.
Can anyone help me in this subject with a simple working example?
'' the above links doesn't work"
thanks

phsieh2005 · February 2, 2012, 12:13

Hi, Vega,

I read a couple of LBM paper about 5 years ago. It looks like the LBM has advanced significiantly recently. However, one of the issue I remembered from the papers 5 years ago is that it is not straight forward to relate physical units to lattice Boltzmann units. For example, inlet velocity of 1 m/s needs to be converted into LB unit. This actually is a simple one. BCs like heat flux might be tricky. Is this still true when you tried PowerFlow?

Pei-Ying

Quote:

Originally Posted by vega

I have had chance to attend training of Exa's Powerflow at their office recently. Here are my views of this software and the company
1. Powerflow's main strength, for industrial applications, is the ease of model setup. Those who
have experience of working in industry (specially automotive) will agree that CAD cleanup and meshing takes the most amount of time.
2. Powerflow runs only in the transient mode so it does take lot more resources than conventional (Navier-Stokes) NS code. This is true specially for cases where steady state simulation is sufficient.
3. Powerflow does better for low Mach number compressible simulations (10 - 40 m/sec) than the NS codes. From my hands-on experiences in the area of aeroacoustics of automotive sunroof buffeting,
I can say that two of the most popular commercial NS codes have not matched the accuracy of Powerflow.
4. One of the point mentioned in the training was that because of the Lagrangian approach, software is better suited for parallelization. Do not have any personal experience to compare its speed-up rate vs NS codes

The company seems to be doing quite well and has expanded significantly over past 2 years or so. It is being used at several automotive OEMs and NASCAR teams for external aero simulations. I think it would be 3rd most used CFD software in the US auto industry after Fluent and Star-CD

Big downside i see is as mentioned earlier that for situations where steady-state simulations are sufficient, Powerflow becomes too expensive.

--
R

anon_a · February 3, 2012, 03:27

As far as I know, OpenFOAM does not have an LBM solver (so the whole thread is in the wrong place :-D).
You might want to try OpenLB (http://optilb.org/openlb/), I would really like to check that out if I had the time.
I am by no means an expert, so I can not comment on the improvements of the method in the last years.

elvis · February 3, 2012, 04:30

Hi,

MSC Software with XFlow seems to some kind of Lattice Bolzmann as well

phsieh2005 · February 3, 2012, 05:47

Hi,

Can also look into www.palabos.org. It can import geometry in stl format.

Pei

immortality · January 6, 2013, 11:14

hi.how to start with OpenLB?
thanks.

elvis · January 7, 2013, 08:44

Quote:

Originally Posted by immortality

hi.how to start with OpenLB?
thanks.

Hi,
http://www.numhpc.org/openlb/ is your friend
read "OpenLB in ten minutes" http://www.numhpc.org/openlb/downloa...b_ug-0.5r0.pdf
=>1.3 How to compile OpenLB programs?

otherwise very easy is http://elbeem.sourceforge.net/ you should find some links to download + to manuals for El'Beem which is lattice-Boltzmann method (LBM) & an Open-Source free surface fluid simulation library.

immortality · January 7, 2013, 10:41

thank you! my professor wants to work with openLB and I want to help him.we have read its pdf but can't run and see any of examples!we unpacked it and used make command at the example cavity3D but don't know how to continue to run ans see the results.
thanks.

elvis · January 8, 2013, 02:46

Hi,
you should get in touch with some one from Engineering Mathematics and Computing Lab (EMCL)
Mr. Henn seems to be the OpenLB expert @EMCL
http://www.emcl.kit.edu/staff_thomas.henn.php

Dr. Mathias J. Krause is seems to be an Author as well
http://www.math.kit.edu/~krause/de

Do you use opengpi ?

dkbadheka · December 1, 2016, 06:54

Hello every one,
recently, i am working on lattice Boltzmann method with BGKW approximation.
i have refer many research paper but i don't know how to change grid spacing (delta x) and time step (delta t). In methodology, those values must be keep it as same or unity. But i don't know that why is it keep same or unity? So please i request to all my friends, if you have any idea or any specific method about that give me a suggestion.
yours sincerealy
DHRUV PATEL

September 8, 2007, 05:04	Hello and a good morning! I	#1
philippose Senior Member Philippose Rajan Join Date: Mar 2009 Location: Germany Posts: 552 Rep Power: 25	Hello and a good morning! I recently came across a CFD simulation software known as "PowerFlow" from the company "Exa". This software claims to perform Computational Fluid Dynamics using a very different approach from the usual discrete Navier Stokes solution. The approach they use is called the "Lattice Boltzmann Approach". It would be interesting to know what the exact advantages / disadvantages of this approach is... does anyone have any experience with this approach? Have a nice weekend! Philippose

September 8, 2007, 10:29	I've heard of the lattice bolt	#2
grtabor Senior Member Gavin Tabor Join Date: Mar 2009 Posts: 181 Rep Power: 17	I've heard of the lattice boltzmann approach (and Exa as well) - basically they are solving at a discrete particle level representing the flow by a discrete set of states at each point. I think that the solution thus uses integer rather than real arithmetic and so is correspondingly faster - although this is presumably offset by the need to provide considerably more vertices to generate the same accuracy of solution. My feeling is (and I havn't looked at it in any great detail) it would only really make sense if you were prepared to build specialist hardware for the problem - which was a possibility 10 years ago but didnt really materialise. Put another way; Exa has been talking their method up for about the last 10 years, and they still havn't made a real inroad into the CFD market, so I can't believe that there is a great benefit from the method. Gavin

September 8, 2007, 19:18	I am no expert at this but Lat	#3
pannala New Member Sreekanth Pannala Join Date: Mar 2009 Posts: 6 Rep Power: 17	I am no expert at this but Lattice Boltzmann Method does not operate on integers like the earlier Lattice Gas Methods. LBM is very efficient if you are looking for solutions of incompressible NS equations in the mesoscale configurations with complex geometries like flow through a porous media or flow over/through collection of particles. The main advantage is that the non-linear convection term gets absorbed into the collision term and treated using a relaxation time. Theoretically you can recover INS from current versions of LBM with certain lattice structures (D2Q9 and D3Q19). Succi's book would serve as a good introduction for LBM and there are open source solutions available like Vladymir (http://cui.unige.ch/~latt/vladymir/download.html) if you want to download and try it out for yourself. Hope this helps, Sreekanth jipaz likes this.

September 9, 2007, 02:14	Dear Foamers, there is an o	#4
sradl Member Stefan Radl Join Date: Mar 2009 Location: Graz, Austria Posts: 82 Rep Power: 18	Dear Foamers, there is an open plattform that is currently using LBM: http://www.openlb.org/ I've not tried this one, but i thing they are pretty at the beginning. For those who are interested, maybe the publications of Jos Derksen, who is using a very efficient FORTRAN LBM code, give a impression on the cababilities of LBM: http://www.tnw.tudelft.nl/live/pagina.jsp?id=0ddabf98-358a-47a0-81f8-e84856bb24f d&lang=en (Prof. Derksen has moved to Alberta as far as i know, so maybe this page will be killed in the near future). You will see that LBM and OpenFOAM will not fit together very well, because in LBM you have structured, cubic grids that will be incompatible with all current mesh classes in OF and totally different equations to solve (maybe that is the smallest problem, because they are quite simple (e.g. no poisson solver needed) :-) ). I've also started some LBM work on the basis of the dersken code. What we do is LES of a whole bioreaktor with about 10-20 Mio. cells on up to 10 parallel machines. Its pretty fast... br Stefan Radl

September 9, 2007, 02:25	Yup he is moving to my departm	#5
msrinath80 Senior Member Srinath Madhavan (a.k.a pUl\|) Join Date: Mar 2009 Location: Edmonton, AB, Canada Posts: 703 Rep Power: 21	Yup he is moving to my department. I had the opportunity to meet with him twice last year when he came for presentation/interviews. I doubt that he will discontinue work on LBA.

September 9, 2007, 04:17	Hello again, Thanks a lot f	#6
philippose Senior Member Philippose Rajan Join Date: Mar 2009 Location: Germany Posts: 552 Rep Power: 25	Hello again, Thanks a lot for the information regarding the Lattice Boltzmann approach.... looks like it is a known, and active area even though it has not hit the mainstream yet. I tend to agree with Gavin, in that if it was a method applicable to general cases as are the current CFD techniques, it would be more popular. On the other hand, I was reading through some papers available online, and also happened to stumble upon openLB.... it has been claimed to be faster in some cases (the comparison was with CFX).... particularly laminar incompressible flow. Also, multi-phase simulations are inherently handled within the solution, without requiring special treatment like in the normal CFD methods. It is definitely not an "only integer" solution technique, quite contrary, it is a very memory and processor hungry method, which is why there have been lots of efforts to parallelise it (apparently this method is inherently easier to parallelise). As for its compatibility with OpenFOAM... I am not aware of the numerical solution requirements of the Lattice Boltzmann method yet, but the mesh is definitely not the limitation. OpenFOAM can handle structured cubic grids as well as unstructured polyhedral meshes....! After all... a cubic grid is a sub-set of an unstructured polyhedral mesh. I have a feeling, that it should be possible to implement the Lattice Boltzmann method using OpenFOAM without any change to the base code... a well thought out top-level solver should do the trick.... unless there are some very special requirements from the mathematical side, for LBM. I wish you all a wonderful Sunday! Philippose

September 9, 2007, 22:19	I have had chance to attend t	#7
vega New Member Rajneesh Join Date: Mar 2009 Posts: 28 Rep Power: 17	I have had chance to attend training of Exa's Powerflow at their office recently. Here are my views of this software and the company 1. Powerflow's main strength, for industrial applications, is the ease of model setup. Those who have experience of working in industry (specially automotive) will agree that CAD cleanup and meshing takes the most amount of time. 2. Powerflow runs only in the transient mode so it does take lot more resources than conventional (Navier-Stokes) NS code. This is true specially for cases where steady state simulation is sufficient. 3. Powerflow does better for low Mach number compressible simulations (10 - 40 m/sec) than the NS codes. From my hands-on experiences in the area of aeroacoustics of automotive sunroof buffeting, I can say that two of the most popular commercial NS codes have not matched the accuracy of Powerflow. 4. One of the point mentioned in the training was that because of the Lagrangian approach, software is better suited for parallelization. Do not have any personal experience to compare its speed-up rate vs NS codes The company seems to be doing quite well and has expanded significantly over past 2 years or so. It is being used at several automotive OEMs and NASCAR teams for external aero simulations. I think it would be 3rd most used CFD software in the US auto industry after Fluent and Star-CD Big downside i see is as mentioned earlier that for situations where steady-state simulations are sufficient, Powerflow becomes too expensive. -- R

September 10, 2007, 01:34	Dear Philippose, you are wr	#8
sradl Member Stefan Radl Join Date: Mar 2009 Location: Graz, Austria Posts: 82 Rep Power: 18	Dear Philippose, you are wrong regarding the grid: it is not that easy as you believe, because you have additional troubles when it comes to boundary conditions. If you want to leave "Lego"-land, i.e. not only cubic cells on the boundary, you need special algorithms to get that right in your collision step of the LBM. Of course: a cubic grid can be easily handeled by OF - the question is if this is satisfactory for your boundary conditions you want to apply. So I disagree: it will not be possible with a top-level solver; you need much more on the grid level and for the boundary condition in addition to the current OF source code. Efficient grid handling (you need several Mio cells!!) and boundary conditions will be a lot of work to implement. Of course it will be possible; but it is not that simple as you mentioned. br Stefan

September 10, 2007, 01:41	Dear Srinath, my best wishe	#9
sradl Member Stefan Radl Join Date: Mar 2009 Location: Graz, Austria Posts: 82 Rep Power: 18	Dear Srinath, my best wishes to you and Prof. Derksen! I also met him last year in italy and his presentations are great fun. I had the change to get parts of his LBM code and its really a nice piece of code. Once again greetings to you and Prof. Derksen from his friends in austria (Prof. Khinast group). br Stefan Radl

September 10, 2007, 06:31	FYI, there are extensions to L	#10
pannala New Member Sreekanth Pannala Join Date: Mar 2009 Posts: 6 Rep Power: 17	FYI, there are extensions to LBM for unstructured grids based on finite volumes. If you are interested, please look at the following reference: S. Ubertini, G. Bella, and S. Succi, Lattice Boltzmann method on unstructured grids: Further developments, PHYSICAL REVIEW E 68, 016701, 2003 The main limitations of LBM approach: a) Compressible flows - need higher order terms for collisions and thus erodes into the simplicity/efficiency. b) Strong coupling of density to temperature variations - this would also require higher order terms. The algorithm is extremely simple so that one can easily parallelize the code. Only the propagation step needs information exchange with the neighbors and thus regular domain decomposition can be used. In complex geometries, the communication can be ordered so that one can have a well-balanced algorithm. In addition, since the propagation and relaxation steps are straight forward, one can possibly make them cache aware so that one can get very good performance. Here is a link to a very simple LBM code in case you want to see the simplicity of the algorithm: http://www.ccrl-nece.de/lba/. Cheers, Sreekanth jipaz likes this.

February 2, 2012, 09:37		#11
bagogan New Member mark Join Date: Jan 2012 Posts: 3 Rep Power: 14	dear all, I start working with LBM to simulate couette -Taylor flow, the inner cylinder is animated with angular velocity. I read some articles about this subject but i'm very confused. Can anyone help me in this subject with a simple working example? '' the above links doesn't work" thanks

February 3, 2012, 03:27		#13
anon_a Disabled Join Date: Mar 2011 Posts: 174 Rep Power: 15	As far as I know, OpenFOAM does not have an LBM solver (so the whole thread is in the wrong place :-D). You might want to try OpenLB (http://optilb.org/openlb/), I would really like to check that out if I had the time. I am by no means an expert, so I can not comment on the improvements of the method in the last years.

February 3, 2012, 04:30		#14
elvis Senior Member Elvis Join Date: Mar 2009 Location: Sindelfingen, Germany Posts: 620 Blog Entries: 6 Rep Power: 24	Hi, MSC Software with XFlow seems to some kind of Lattice Bolzmann as well

February 3, 2012, 05:47		#15
phsieh2005 Senior Member Pei-Ying Hsieh Join Date: Mar 2009 Posts: 334 Rep Power: 18	Hi, Can also look into www.palabos.org. It can import geometry in stl format. Pei BlnPhoenix likes this.

January 6, 2013, 11:14		#16
immortality Senior Member Ehsan Join Date: Oct 2012 Location: Iran Posts: 2,208 Rep Power: 26	hi.how to start with OpenLB? thanks.

January 7, 2013, 10:41		#18
immortality Senior Member Ehsan Join Date: Oct 2012 Location: Iran Posts: 2,208 Rep Power: 26	thank you! my professor wants to work with openLB and I want to help him.we have read its pdf but can't run and see any of examples!we unpacked it and used make command at the example cavity3D but don't know how to continue to run ans see the results. thanks.

January 8, 2013, 02:46		#19
elvis Senior Member Elvis Join Date: Mar 2009 Location: Sindelfingen, Germany Posts: 620 Blog Entries: 6 Rep Power: 24	Hi, you should get in touch with some one from Engineering Mathematics and Computing Lab (EMCL) Mr. Henn seems to be the OpenLB expert @EMCL http://www.emcl.kit.edu/staff_thomas.henn.php Dr. Mathias J. Krause is seems to be an Author as well http://www.math.kit.edu/~krause/de Do you use opengpi ? Last edited by elvis; January 8, 2013 at 03:27.

December 1, 2016, 06:54	About Lattice boltzmann method	#20
dkbadheka New Member deep badheka Join Date: Feb 2016 Posts: 1 Rep Power: 0	Hello every one, recently, i am working on lattice Boltzmann method with BGKW approximation. i have refer many research paper but i don't know how to change grid spacing (delta x) and time step (delta t). In methodology, those values must be keep it as same or unity. But i don't know that why is it keep same or unity? So please i request to all my friends, if you have any idea or any specific method about that give me a suggestion. yours sincerealy DHRUV PATEL

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
About Lattice Boltzmann Method (LBM)	imnth	Main CFD Forum	6	January 27, 2009 01:55
Lattice Boltzmann method	Chen	Main CFD Forum	2	March 12, 2003 02:19
lattice boltzmann method	yangqing	Main CFD Forum	4	August 22, 2002 10:15
lattice Boltzmann (again)	Pei-Ying Hsieh	Main CFD Forum	1	August 23, 2000 18:21
Lattice Boltzmann	Pei-Ying Hsieh	Main CFD Forum	3	August 8, 2000 11:34