CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   OpenFOAM Meshing & Mesh Conversion (https://www.cfd-online.com/Forums/openfoam-meshing/)
-   -   [Other] OF meshless (https://www.cfd-online.com/Forums/openfoam-meshing/123511-meshless.html)

cfdivan September 14, 2013 11:43

OF meshless
 
Dear all,

Considering how time consuming is the mesh process in any of our simulations, is there anybody aware of the implementation of "meshless methodoly" in OF? I am expecting with this approach reduce significantly the hours spent in mesh debug and improve the solution performance by reducing the non-convergence solutions. There are any validation tests together with the code ready to run?

It would be quite interesting try out some performance tests of OF with this mesh methodology.

Regards,

wyldckat September 15, 2013 07:26

Greetings cfdivan,

It took me a while to remember about this... but I remembered just now. There is a non-official implementation (aka it's not in the official and public OpenFOAM version) of the "Immersed Boundary Method" developed by Wikki: http://wikki.gridcore.se/application...BoundaryMethod

AFAIK, always keep in mind that there is no such thing as a silver bullet in CFD. One meshing strategy might be excellent for one thing, but terrible for another. And the "Immersed Boundary Method" and/or meshless are always implied meshing strategies, namely that even though you do not have to bother to generate a mesh, a mesh does still exist, since the geometry is still analysed in a discrete way, not in a continuous way!

Best regards,
Bruno

elvis September 16, 2013 08:24

Hello,

just my 2 cent

you can dig into Immersed Boundary Meth. (IBM) =>Z. Tuković (source code 4 OF-extend) by studing OF-workshop material (OFW7_2012_Darmstadt)
http://sourceforge.net/projects/open...darySlides.tgz
http://sourceforge.net/projects/open...lmMaterial.tgz

I remember two guys that implemented Smoothed Particle Hydrodynamics (SPH)-method with OpenFOAM one presentation at OFW6 @ Penn State
http://sourceforge.net/projects/open...mar_slides.pdf
the second guy Basti Werdelmann was giving a presentation at OF-Stammtisch @ Karlsruhe
http://www.rz.uni-karlsruhe.de/~hj16...Stammtisch.pdf.
You should contact Hendrik Rusche (Wikki) or Ivana Buntic-Ogor (on the OF-Stammtisch List) if you want the Stammtisch-presentation they might have it

But I am not aware of any SPH-code for OF out in the wilderness to the public.

But these examples show it can be done with enough OF-expertise, s.th I do not have
:-(

cfdivan September 17, 2013 18:28

Hi all,
Thank you very much for both posts...they are quite usefull.

Bruno,
Yes, even for "meshless" approach the domain/flow is solved in discrete way but, I meant the immersed BL approach . It is tipically quite easy do a pretty decent surface meshes. Even with the decent surface mesh, problems in "BL zone" (the prism layer extrusion process) are quite often. Those areas are usually the origin for some solution divergence. I knew that wikki already tried that approach. I guess that karalit also implemented that approach. I was just wondering if this approach was already implemented by someone in the forum who can share the code/expereciences.

"AFAIK, always keep in mind that there is no such thing as a silver bullet in CFD. One meshing strategy might be excellent for one thing, but terrible for another". Do you know what kind of applications that immersed BL approach could be terrible?

Elvis,
Thank you very much for all the links. I'll try to dig in a bit more and share the findings/ideas here.

Regards,

wyldckat September 17, 2013 19:07

Hi cfdivan,

Quote:

Originally Posted by cfdivan (Post 452285)
"AFAIK, always keep in mind that there is no such thing as a silver bullet in CFD. One meshing strategy might be excellent for one thing, but terrible for another". Do you know what kind of applications that immersed BL approach could be terrible?

Sorry, I vaguely remember reading somewhere about one or two situations in which it didn't work very well, but I can't remember where I read about it.
Which is why I ended writing about what has been found several times in CFD for many other situations: no matter how many new methods we find for simulating or meshing, they always have some flaw that leads us to go back to the good old standard procedures/methods/models.

Best regards,
Bruno

elvis September 30, 2013 04:13

Hi,

I saw http://www.wire.tu-bs.de/workshops/nofun13/d_nofun.html
=>"LBM mesh generation in OpenFoam" a presentation by Andrea Pasquali (TU Braunschweig)
so some kind of Lattice Bolzmann Method might be possible to be handled with as well

cfdivan October 2, 2013 18:25

Hi Elvis,

Thk u very much for your link. "LBM mesh generation", means a meshless process? What is the main difference between "LBM mesh generation" and "Immersed Boundary Method"? I checked on the website but I did not find any additional information about the topic. Do you know where I can find more documentation about the topic LBM mesh generation in OpenFoam and/or comparision with classic methods or Immersed Boundary Method?

Many thank

wyldckat October 5, 2013 02:51

Greetings to all!

@cfdivan:
Quote:

Originally Posted by cfdivan (Post 454762)
"LBM mesh generation", means a meshless process?

Nope. LBM is this: http://en.wikipedia.org/wiki/Lattice_Boltzmann_methods
In a nutshell, based on what I studied a few years ago:
  1. It doesn't use FVM (Finite Volume Method), while both OpenFOAM and IBM do use it.
  2. You still have a mesh, usually using uniformly cube-shaped cells.
  3. The data is passed between the cells as if using a particle based mechanism, i.e., one particle can only move from one cell to another. It's not exactly this, but its the mnemonic I use to understand this model.
  4. The advantage of this over FVM CFD is that it's easily parallelizable, namely for using GPGPU hardware.
I think LBM has improved considerably in the last 3-4 years, but it probably still has several limitations.

Best regards,
Bruno

elvis October 5, 2013 06:43

Hi,

basicly I wanted to say that OF is a Computational Continuum Mechanics (CCM) toolkit
which can be used for different methods like FVM and FEM which are already implemented.
But other people used it for different methods like SPH

cfdivan October 5, 2013 08:12

Hi wyldckat,
Thk u very much for your link. Pretty usefull for a brief review.

Regarding FVM - I already tried to investigate this point some time ago and I found that they likely use FD...is it true? I never was able to confirm that :(

With FD we have several advantages when we are working with structured meshes....it is very simple and effective and quite easy to obtain higher-order schemes (High-Mach numbers applications for example).

"The data is passed between the cells as if using a particle based mechanism" - This really sounds a restriction for high mach-number flows where you need higher-order schemes .
Considering that you have different sizes of the cells in your domain and the equations are solved in "Lattice world", in one time step you can have one particle go to the neighbor cell while in the other zone of your domain the particle can travel 2 or more cells. Based on this assumption, we will not be able to track the particle distribution. They need to work around with field update, is it right?

The paralelization is in fact an significant advantage of this code. The algorithm is quite simple and only the stream step needs exchange of information with neighbors which does not happen in FVM.

It looks that the efficieny of this code comes from the implementation of simple algorithms where non higher-order schemes are needed. Though, when it comes to high-mach numbers and coupling of density to temperature variations, higher-order schemes are nedded and the simplycity/efficiency of the code decreases.

There is also a point regarding the full unsteadiness of this approach. When compared with DES-SA of NS for example, it might be more fast/easy to obtain a unsteady converged solution. Nevertheless there are an wide range of engineering applications where the RANS/steady state provide us enough accuracy for very low prices in terms of CPh which migh not be possible with LBM. It looks like LBM fits better for applications where the unsteadiness is the king.

In any case, what looks really interesting for me is the "meshless concept" that is very painfull process on my simulations that I am expecting that we do not have neither in IBM nor LBM. From time to time I spend a lot of hours to check why there is no convergence on simulation, ending up play around with disc schemes and mesh.

Regards,

wyldckat October 5, 2013 08:22

Hi cfdivan,

As far as I can remember, LBM works in a similar way to the Courant number restriction that we're taught in the very first tutorial OpenFOAM User Guide: http://www.openfoam.org/docs/user/cavity.php
In essence, the size of each cube must respect the flow speed limit, otherwise it simply cannot present properly your problem. In addition, the cube sizes also had to take into account the properties of the fluid, mainly the viscosity.
In other words, LBM is almost a "Similitude" modelling strategy: http://en.wikipedia.org/wiki/Similitude_%28model%29

Therefore, with any luck, for supersonic flow, you only need several billion LBM cells, working on a cluster with 8 GPUs per node... :rolleyes:


As for "FD": which "FD" are you talking about?

Best regards,
Bruno

cfdivan October 9, 2013 21:10

Hi Bruno,

Sorry for the delay on reply
FD, I meant Finite diffrences.

I also agree that the LBM might works in similiar way of Courant number. To me it sounds that the scalling of the flow properties (like velocity for example) to the "Lattice world" take in consideration the cell size and the speed of the sound, is it right? The "Courant number" in LBM is ajusted at each cell in order to keep the tracking of the particles at each timestep.

Regards,

julien.decharentenay October 9, 2013 22:47

Hi cfdivan/Bruno,

Great discussion. The openFoam solvers (simpleFoam, etc) are based on FVM discretisation of the NS equations.

Wouldn't it be great to have access to various methods (FVM, IBM, LBM, SPH) using a unique and compatible set of input data rather than juggling between solvers, pre-processors, etc.... This would make it easier to use the right tool for the right job.

ali_atrian November 9, 2014 07:49

"Basti Werdelmann" presentation at OF-Stammtisch
 
Quote:

Originally Posted by elvis (Post 451989)
the second guy Basti Werdelmann was giving a presentation at OF-Stammtisch @ Karlsruhe
http://www.rz.uni-karlsruhe.de/~hj16...Stammtisch.pdf

Hello elvis
could you please give me a more accurate address of "Basti Werdelmann"?
I'm going to know how to implement SPH in openFoam so I googled the name for hours but noo cunclusion!
is the presentation on 6th openfoam workshop (13-16 June 2011)? or ...?
the link you have inserted does not work any more!
do you know any other presentation, work, coding, tutorials on the issue?
Best Regards

elvis November 10, 2014 03:33

Hi Alireza,

the implementation of the SPH presented at Karlsruhe OF-Stammtisch is a different one to that presented at Penn State OF-workshop, but both are closed source to my knowledge.

You should find his address in https://wiki.manchester.ac.uk/spheri..._Abstracts.pdf
Hope you are not a headhunter ;-)


All times are GMT -4. The time now is 03:31.