Adventures In CFD

n72.75

Move slow and try not to break too much.
Orbiter Contributor
Addon Developer
Tutorial Publisher
Donator
Joined
Mar 21, 2008
Messages
2,687
Reaction score
1,337
Points
128
Location
Saco, ME
Website
mwhume.space
Preferred Pronouns
he/him
Logo%20R&D%20White.png



So I got bored and started playing around with lifting body shapes and the trial version of MicroCFD. It produced some interesting results. As you can see, I am able to get force and moment calculations out of this software (which is not free, but the way, but is affordable bo mere mortals, unlike ANSYS or the like) however, running on my 8 year old Q6600 took over 9 hours to solve.

If you wanted to build a lookup table with a resolution of:

α 5° increments up to 90°
β 5° increments up to 90°
δe in 5° increments from -40° to 40°
δa in 5° increments from 0° to 40°
δr in 5° increments from 0° to 30°

α -- 18 points
β -- 18 points
δe -- 16 data points
δa -- 8 data points
δr -- 8 data points

With my setup, you'd be looking at roughly 340 years of computation time. There are of course a number of ways to shorten that. My GPU only has 1.2 GB of ram. If it had 6GB or so, I'd be able to run this simulation of the GPU, shortening the simulation time by an order of magnitude or so.

http://microcfd.com/runtime.htm

If you were smart about what you treated as a superposition, particularly control surface deflections, you could probably get your comput times down to a year or two.

Then you'd be able produce a lookup table, assuming you trust the mesh and solver to be accurate.

With adaptive meshing and GPU computing you could probably cut two orders of magnitude off of the compute time. Then you're only 10000 times too slow to do this in real time. But Moore's law says we're only 20 years away from being able to do that...

Enjoy the pictures:

01.png


02.png


03.png


04.png


05.png


06.png


07.png


08.png


09.png


10.png


11.png


12.png


13.png


14.png


15.png


I'm calling the space plane the Moth

Moth.png




I might develop it into something, but no promises.
 
Last edited:

Urwumpe

Not funny anymore
Addon Developer
Donator
Joined
Feb 6, 2008
Messages
37,588
Reaction score
2,312
Points
203
Location
Wolfsburg
Preferred Pronouns
Sire
I wanted to do something similar with OpenFOAM, but I lack Linux box with a better CPU right now. :(
 

SolarLiner

It's necessary, TARS.
Addon Developer
Joined
Jun 14, 2010
Messages
1,847
Reaction score
2
Points
0
Location
404 ROAD NOT FOUND
I too wanted to try OpenFOAM when I discovered there were Windows binaries available, but libraries were too scattered out, and trying to find them ended up in a nightmare. I do have a dual boot with Ubuntu on it but it's been quite some time I've actually used it. That is giving me ideas though, so I'll try and see if I can get it to run on my machine.
 

Urwumpe

Not funny anymore
Addon Developer
Donator
Joined
Feb 6, 2008
Messages
37,588
Reaction score
2,312
Points
203
Location
Wolfsburg
Preferred Pronouns
Sire
I too wanted to try OpenFOAM when I discovered there were Windows binaries available, but libraries were too scattered out, and trying to find them ended up in a nightmare. I do have a dual boot with Ubuntu on it but it's been quite some time I've actually used it. That is giving me ideas though, so I'll try and see if I can get it to run on my machine.

We also tried to get OpenFOAM running on Windows machines at work, but all concentrated expertise of computer experts and engineers failed there. :lol: Its way easier and better documented on Linux.
 

n72.75

Move slow and try not to break too much.
Orbiter Contributor
Addon Developer
Tutorial Publisher
Donator
Joined
Mar 21, 2008
Messages
2,687
Reaction score
1,337
Points
128
Location
Saco, ME
Website
mwhume.space
Preferred Pronouns
he/him
We also tried to get OpenFOAM running on Windows machines at work, but all concentrated expertise of computer experts and engineers failed there. :lol: Its way easier and better documented on Linux.


Same experience here.
 

Thunder Chicken

Fine Threads since 2008
Donator
Joined
Mar 22, 2008
Messages
4,327
Reaction score
3,248
Points
138
Location
Massachusetts
I got OpenFOAM to run on a WindowsXP machine using Cygwin. Other than being on a desperately underpowered box it worked fine.

This was a few years ago, but I don't recall any major headaches getting it done.
 

boogabooga

Bug Crusher
Joined
Apr 16, 2011
Messages
2,999
Reaction score
1
Points
0
I've always wanted to see a CFD simulation of the Delta Glider. :hmm:

Questions:

1) I assume that since you are resource limited, you are doing steady-state RANS?

2) What numerical scheme are you using?
In case you don't know, you have to be careful with solvers that are not designed to handle supersonic flow. Really, you should be using upwinding schemes and be aware of numerical dissipation/dispersion issues, especially around your shocks.

3) If RANS, what, if any, turbulence model are you using?
 
Last edited:

n72.75

Move slow and try not to break too much.
Orbiter Contributor
Addon Developer
Tutorial Publisher
Donator
Joined
Mar 21, 2008
Messages
2,687
Reaction score
1,337
Points
128
Location
Saco, ME
Website
mwhume.space
Preferred Pronouns
he/him
I've always wanted to see a CFD simulation of the Delta Glider. :hmm:

Questions:

1) I assume that since you are resource limited, you are doing steady-state RANS?

2) What numerical scheme are you using?
In case you don't know, you have to be careful with solvers that are not designed to handle supersonic flow. Really, you should be using unwinding schemes and be aware of numerical dissipation/dispersion issues, especially around your shocks.

3) If RANS, what, if any, turbulence model are you using?

These are all very good questions, regrettably, to which I have no answers. This software is MicroCFD, written by Axel Rohde of FIT. He's published a number of papers with the AIAA (I dropped mu membership last year so I can't download them unfortunately).

The great word of warning about FEM/CFD is: understand your meshing method, and understand your solver.

I assume it's RANS. I know with Eüler solvers you can a gibbs phenomenon like behavior with shocks. One of things that's turned me off from this software package is the lack of published material on the methods it uses.

I've done a bit of research on: http://www.cfd-online.com/ but found nothing but heresay and opinion. I'm sure if you emailed the guy he'd tell you.

It's an interesting bit of software, and it (at the very least) goes to show that this sort of thing is not impossible (different than "possible").

Maybe I'll email him tomorrow.
 

boogabooga

Bug Crusher
Joined
Apr 16, 2011
Messages
2,999
Reaction score
1
Points
0
Are you using a structured or unstructured grid?
 

n72.75

Move slow and try not to break too much.
Orbiter Contributor
Addon Developer
Tutorial Publisher
Donator
Joined
Mar 21, 2008
Messages
2,687
Reaction score
1,337
Points
128
Location
Saco, ME
Website
mwhume.space
Preferred Pronouns
he/him
Are you using a structured or unstructured grid?

I believe it's an unstructured grid.

---------- Post added 05-01-16 at 06:08 PM ---------- Previous post was 04-30-16 at 08:38 PM ----------

Are you using a structured or unstructured grid?

From his email.

Hi Matthew,

Thank you for your interest in my 3D VWT software.

I assume you read the brief technical description on following page: http://microcfd.com/3d.htm

The CFD solver and turbulence model are "proprietary", i.e. I will not get into the details of the underlying methods (I am sorry, if that disappoints you). The algorithms are very "basic" to run on GPU.

The grid is uniform Cartesian with cut cells at the surface and a "proprietary" boundary layer model. The resolution of the Cartesian grid is doubled inside the "box" compared to the far field. The boundary layer is not actually resolved, so you cannot compare it to an RANS solver. Commercial solvers like SolidWorks Flow Simulation work similarly to shorten computation times.

The shocks are resolved through a localized upwind algorithm, which adds sufficient numerical viscosity for a stable solution, even if the shock oscillates back and forth, like in a cavity flow:

http://microcfd.com/movies/cavity-p.htm

The shuttle orbiter validation results I have online were NOT produced with the free STL sample files I distribute. I used a slightly simplified and watertight model, which I can send to you, if you like. I also have some publications with flight data.

http://microcfd.com/gallery/shuttle-orbiter-lift-drag.htm

The STL model you used in your orbiter model online is a bit crude in terms of triangulation, and the thickness of the fins (relative to the body) is probably right around one grid unit (see log) and thus may be underresolved.

[...]
 

boogabooga

Bug Crusher
Joined
Apr 16, 2011
Messages
2,999
Reaction score
1
Points
0
Proprietary!?
:huh:

Sorry, n72.75, but in my opinion that sounds like a CFD toy, not real CFD.
 

n72.75

Move slow and try not to break too much.
Orbiter Contributor
Addon Developer
Tutorial Publisher
Donator
Joined
Mar 21, 2008
Messages
2,687
Reaction score
1,337
Points
128
Location
Saco, ME
Website
mwhume.space
Preferred Pronouns
he/him
Proprietary!?
:huh:

Sorry, n72.75, but in my opinion that sounds like a CFD toy, not real CFD.

It probably a good approximation of most "normal" problems. What I'm trying to do is pretty "out there". And honestly because I'm more interested in the CFD itself than the actual results, I think I'm going to play around with openfoam et al. again. I want control over my meshing and my solver.

I mean no disrespect to the guy but I think it's a bad move not making the source code available.
 

Urwumpe

Not funny anymore
Addon Developer
Donator
Joined
Feb 6, 2008
Messages
37,588
Reaction score
2,312
Points
203
Location
Wolfsburg
Preferred Pronouns
Sire
I find it a bit disappointing that you can not mesh your own grid around a model. Usually, that is part of the art of CFD analysis, knowing where you expect most action and need a finer grid.
 

Thunder Chicken

Fine Threads since 2008
Donator
Joined
Mar 22, 2008
Messages
4,327
Reaction score
3,248
Points
138
Location
Massachusetts
I mean no disrespect to the guy but I think it's a bad move not making the source code available.

At very least the physics models and numerical algorithms should be provided. Most reputable CFD vendors document these in detail. Keeping the source code proprietary is understandable - explaining what the code is doing is another matter entirely.

I'd move on. This might be great CFD software, but without a validation it is impossible to say.
 

boogabooga

Bug Crusher
Joined
Apr 16, 2011
Messages
2,999
Reaction score
1
Points
0
What Thunder Chicken said. CFD numerical algorithms are a source of academic research and several high-quality ones are in common use. I doubt that this guy is improving much upon them, except perhaps to optimize for lower performance computers.

It is also suspect that you provide a regular STL files and the software magically generates a CFD grid.

It would be nearly impossible to publish any results if the numerical scheme is unknown. Also, I find it unlikely that any professional would stake their career as such.


I can only conclude that this software is intended for CFD amateurs- which is, of course, exactly how it is being used if the goal is only Orbiter add-on development.
 

Jon

New member
Joined
Jun 21, 2016
Messages
1
Reaction score
0
Points
0
MicroCFD Validation

I stumbled upon this post while doing a Google search on "microcfd validation" and decided to add my two cents.

I have been using the MicroCFD virtual wind tunnel for some time to reveal the shock structure of supersonic aircraft designs, and I have not found a commercial CFD tool that runs faster, offers better resolution, is easier to use, and still affordable at the same time.

Its grid is uniform Cartesian, i.e. highly structured, consisting of equally spaced cubes in the near field and far field. The size of the cubes in the near field is the same as the (1 gu) diameter of the stream particles shown. The color cuts are not interpolated and thus reveal the grid structure and resolution. Auto-generating a Cartesian grid around an STL geometry is not uncommon in commercial CFD software; neither is boundary layer modeling.

Although the MicroCFD flow solver is poorly documented, the developer does provide validation examples for 2D, axis-symmetric and 3D flow that can be easily verified, if one goes through the trouble of recomputing these cases, or at least a random selection of data points:

http://microcfd.com/gallery/naca-airfoil-lift-drag.htm
http://microcfd.com/gallery/g7-bullet-drag.htm
http://microcfd.com/gallery/shuttle-orbiter-lift-drag.htm

I would not go through the hassle of trying to learn OpenFOAM and generating my own mesh, which may or may not yield more accurate results, since you are ultimately limited by computing power, not solver technology, when analyzing complex 3D supersonic flow.

Anyway, that is my "humble" opinion.
 
Top