HobbyKing.com New Products Flash Sale
Reply
Thread Tools
Old Mar 09, 2011, 08:21 AM
Registered User
Joined Feb 2011
12 Posts
OK, here we go - complete BP3434 parameterization for ya enjoy
bp3434

It all seems to be working well enough - a not-exactly-perfectly-xy-scaled example is shown:
Name: airfoil example.png
Views: 131
Size: 6.1 KB
Description:
According to the paper, BP3434 tended to take longer to converge than BP3333, but was able to duplicate a wider range of existing designs - both factors due to the larger number of parameters. I may implement the BP3333 sometime in the next few weeks, but I have some more pressing optimization problems I need to get cracking on
madact is offline Find More Posts by madact
Last edited by madact; Mar 09, 2011 at 08:33 AM.
Reply With Quote
Sign up now
to remove ads between posts
Old Mar 09, 2011, 10:19 AM
Sink stinks
Montag DP's Avatar
United States, GA, Atlanta
Joined Apr 2005
4,547 Posts
Thanks, madact. I'll take a look at those references.
Montag DP is offline Find More Posts by Montag DP
Reply With Quote
Old Mar 14, 2011, 05:40 AM
tau
Registered User
Joined Aug 2006
44 Posts
@Dan & Mark

I've followed this thread on and off and it's quite interesting what you guys are trying!

Just as a hint: maybe it's worth looking at the DE (differential evolution) algorithm as well.

We are using it quite frequently for the optimization of technical (structural/mechanical) problems with up to 60 (or even more) parameters and it has a very good reputation of being efficient(fast) and reliable. The source for it is freely available as Matlab, C and other codes as well.

It has been quite successful in the past winning competitions on an academic level too.

Good luck & keep posting, it's really interesting to read about your progress!

Tunc
tau is online now Find More Posts by tau
Reply With Quote
Old Mar 14, 2011, 07:51 AM
Registered User
United States, KS, Andover
Joined Oct 2005
539 Posts
Thanks, Tau, I'll look into that.

I'm about ready to return to a GA method. For some reason, my swarm method hasn't performed the way I know it should. It seems to get stalled and can't find the global minimum. I'm sure it's something I've done. I've had good luck with the GA, so I want to try that again. I got distracted by an actual airplane build, of all things...
markschaffin is offline Find More Posts by markschaffin
Reply With Quote
Old Mar 19, 2011, 09:00 PM
Registered User
Joined Mar 2011
2 Posts
Hi,

I found this thread and would like to ask something more in general.
How often is xfoil used for aifoil optimization from you guys, engnieeres and co?
Is there a big need for it and if there is, for what kind of tasks? model planes, drones, planes?

I ask because if there is a big demand for it, I would consider to start a single sub-project for such cases on my www.aerospaceresearch.net/constellation, because computing power wouldn't be a problem, what was mentioned on a previous page.
And I would like to have a subproject where people can set up their parameter and it will be computed on Constellation and you will get back your result.
But therefore it should be more frequently used.

Best regards,

Andreas
hornig is offline Find More Posts by hornig
Reply With Quote
Old Mar 23, 2011, 07:15 AM
Registered User
Joined Feb 2011
12 Posts
Quote:
Originally Posted by hornig View Post
I found this thread and would like to ask something more in general.
How often is xfoil used for aifoil optimization from you guys, engnieeres and co?
Is there a big need for it and if there is, for what kind of tasks? model planes, drones, planes?

I ask because if there is a big demand for it, I would consider to start a single sub-project for such cases on my www.aerospaceresearch.net/constellation, because computing power wouldn't be a problem, what was mentioned on a previous page.
The work I'm doing isn't with XFoil, but with OpenFOAM, which is a different kettle of fish entirely. I have used XFoil before though...

Xfoil is a 2D panel method with boundary layer analysis. This is very useful for optimizing airfoils in their 'normal' mode of operation, but doesn't handle flow separation, eddies etc very well at all, e.g. wings in stall, the trailing edge of laminar flow airfoils, etc. It is, however, pretty good for optimizing airfoils in normal modes of operation - and it will model transonic and hypersonic flows in the right hands. Relatively easy to pick up for modellers and other 'amateurs' like myself, too, as it needs only a basic knowledge of aerodynamics (things like viscosity, Reynolds numbers, etc that you need anyway).
Being a panel method, calculations tend to go quite quickly (unless you're plotting high resolution polars, of course )... distributed computation could still be useful for genetic algorithms with large population sizes. You'll have to let the others comment on how useful that would be, though, the last time I used it 'in anger' was over 5 years ago, and I remember it being quite snappy back then...

OpenFOAM is another can o worms - it's a general-purpose (very general purpose) finite element package. I'm pretty sure you'd need a degree in computational physics to get the most out of it (I'm just scratching around in a small area of it), and it's really easy to "shoot yourself in the foot" with it, but it can do static and dynamic analyses, RANS modelling, large eddy separation, structural analysis, mixed-phase fluid analysis, moving meshes, you name it... it's primarily intended for 3D simulation, though it can handle 2D as well. Having it available on a distributed net would be nothing short of awesome, but the input geometry can be quite big (~70MB in one of the more interesting tutorials), the output can be very big, and it really, really wants a high-specced machine running it, if you hit memory limits in particular (not too hard to do, especially in 3D with an overly-high mesh resolution), things can go downhill fast, performance wise... On the other hand, it can give results that XFoil can't for airfoils, such as noise analysis & eddies, analysis of airfoils in stall, multi-element airfoils etc. and used in 2D case it has relatively modest requirements - though transferring geometry data might once again be more of a bottleneck than processing.
madact is offline Find More Posts by madact
Last edited by madact; Mar 23, 2011 at 07:24 AM.
Reply With Quote
Old Mar 23, 2011, 10:37 AM
Registered User
truffaldino's Avatar
Joined Jul 2007
334 Posts
Quote:
Originally Posted by hornig View Post

I would consider to start a single sub-project for such cases on my www.aerospaceresearch.net/constellation, because computing power wouldn't be a problem,
And I would like to have a subproject where people can set up their parameter and it will be computed on Constellation and you will get back your result.
But therefore it should be more frequently used.

Andreas
Hi Andreas,

It would be nice if you could start a project to run CFD simulations for whole wing or airplane rather than an airfoil ("virtual 3D wind tunnel"). For instance something like XFLR5, but with Large Eddy Simulations, rather than vortex lattice method supplemented with x-foil polars: User will send you an aircraft geometry in AVL/XFLR format and will get polars as an output. Mesh can be easily generated by some program from XFLR data and calculations could be done using open-source codes. The end user will only need to know how to deal with XFLR which is extremely easy to use (one needs to spend just a couple of hours to master it).

As was mentioned above, XFOIL/XFLR is quite accurate when you deal with reguar airfoils/ wings in normal mode of operation. At low reynolds numbers, XFOIL is even more accurate than regular 2D RANS models in CFD simulations, but when you have something non-standard XFOIL/XFLR5 will not work: It will not handle strong spanwise flows, vorticity regions, nonregular geometries, will not take properly flows past fuselage into account etc. So demand for a such kind of simulations could be quite high.

Truffaldino
truffaldino is offline Find More Posts by truffaldino
Reply With Quote
Old Mar 24, 2011, 03:14 AM
Registered User
Joined Feb 2011
12 Posts
Yeah, if someone who knew their computational fluid dynamics were to come up with a wrapper that hooked up XFLR to OpenFOAM's solvers for compressible and incompressible flow, with solvers tuned for a selection of reynolds and mach numbers, and with a choice of a couple of 'standard' mesh resolutions and refinement parameters, it could be highly useful - some "nonstandard" designs (e.g. blended wing designs, ring-wings and other fun stuff) could then be simulated fairly accurately without a great deal of ultra-specialist domain knowledge in simulation... and the same goes for doing things like noise and stall analysis of conventional designs...

One thing I don't like about XFLR is the way the fuselage frames are set up - it makes it very hard to achieve curvature continuity. But I guess I'm biased by working primarily on fuselage-like problems just at the moment and it's probably the easiest program to pick up that I've come across for this kind of thing.
madact is offline Find More Posts by madact
Reply With Quote
Old Mar 24, 2011, 05:02 AM
Registered User
truffaldino's Avatar
Joined Jul 2007
334 Posts
Quote:
Originally Posted by madact View Post
Yeah, if someone who knew their computational fluid dynamics were to come up with a wrapper that hooked up XFLR to OpenFOAM's solvers ...
Why don't we guys to try to do this? I found that there is people here that works with OpenFOAM, Fluent, CFX etc. At least mesh generation could be done without giant efforts. We can try, for instance, to do AVL/XFLR to gmsh geometry file converter: it quite feasible for an unstructured mesh. In case of success we could go on with a coupling to a solver...
truffaldino is offline Find More Posts by truffaldino
Reply With Quote
Old Mar 24, 2011, 09:46 AM
Registered User
Joined Feb 2011
12 Posts
I'm kinda snowed under 'till June, but a couple of tips for anyone who wants to take some of this on:

I've been using snappyhexmesh for meshing in OpenFOAM (there are computational speed and accuracy advantages to hexes compared tp tets such as gmsh spits out, and it does a rocking job of boundary layers and custom volume refinement once you work out the right parameters) but it would make mincemeat out of trailing edges - though I have seen a nice util which might fix the shortcomings: http://openfoamwiki.net/index.php/Contrib_snapEdge and it would be pretty easy to apply automatically to wing sections coming out of an XFLR converter or similar.

I'd recommend STL ascii or binary files as a surface mesh transfer format - it's dead simple and read/written everywhere.

If you can get octave into constellation, I'd highly recommend it - grab the 'nurbs' package from octave-forge and the iges converter from the Matlab file exchange, and you have a simple handler for all your NURBS needs. You can generate a surface mesh using nrbeval(), dump to a file, and use a simple C/C++ program to convert to STL, then send that off to gmsh/snappy or whatever.


Probably time to take this to a new thread so we don't annoy mark & montag too much
madact is offline Find More Posts by madact
Reply With Quote
Old Mar 24, 2011, 12:10 PM
Sink stinks
Montag DP's Avatar
United States, GA, Atlanta
Joined Apr 2005
4,547 Posts
Quote:
Originally Posted by madact View Post
Probably time to take this to a new thread so we don't annoy mark & montag too much
No, this stuff is interesting, and not annoying at all. It is a little off-topic, though, so feel free to start a new thread if you want. I haven't really done anything with my optimization since the last update. I have too many other things that need attention.
Montag DP is offline Find More Posts by Montag DP
Reply With Quote
Old Mar 24, 2011, 04:29 PM
Registered User
zummi's Avatar
Cape Town, South Africa
Joined May 2009
149 Posts
Quote:
Originally Posted by madact View Post
I'm kinda snowed under 'till June, but a couple of tips for anyone who wants to take some of this on:

I've been using snappyhexmesh for meshing in OpenFOAM (there are computational speed and accuracy advantages to hexes compared tp tets such as gmsh spits out, and it does a rocking job of boundary layers and custom volume refinement once you work out the right parameters) but it would make mincemeat out of trailing edges - though I have seen a nice util which might fix the shortcomings: http://openfoamwiki.net/index.php/Contrib_snapEdge and it would be pretty easy to apply automatically to wing sections coming out of an XFLR converter or similar.

I'd recommend STL ascii or binary files as a surface mesh transfer format - it's dead simple and read/written everywhere.

If you can get octave into constellation, I'd highly recommend it - grab the 'nurbs' package from octave-forge and the iges converter from the Matlab file exchange, and you have a simple handler for all your NURBS needs. You can generate a surface mesh using nrbeval(), dump to a file, and use a simple C/C++ program to convert to STL, then send that off to gmsh/snappy or whatever.


Probably time to take this to a new thread so we don't annoy mark & montag too much

Don't underestimate the complexity of the problem you are trying to tackle. Generating a mesh using SnappyHexMesh is probably going to be the least of your problems.

Currently, no single RANS turbulence model is going to do a good job of predicting the correct behaviour at low Re numbers. You would get reasonable results for low angles of attack, but the current models are not going to do a good job at predicting separation.

So it really depends on what you are trying to get out of the CFD models. If you want to determine the behaviour pre-separation then you would probably get a reasonable estimate using RANS turbulence models.

From literature, the trend in simulating post-separation behaviour seems to heading the route of DNS and LES type simulations, which has it's own set of problems. Anyway, that's just my 2 cents worth.

Cheers
Imraan
zummi is offline Find More Posts by zummi
Reply With Quote
Old Mar 24, 2011, 07:25 PM
Registered User
Joined Feb 2011
12 Posts
Quote:
Originally Posted by zummi View Post
From literature, the trend in simulating post-separation behaviour seems to heading the route of DNS and LES type simulations, which has it's own set of problems. Anyway, that's just my 2 cents worth.
True 'nuff. I'd like to do some modelling on my problem with LES using OpenFOAM's 'pimpleFoam' solver, but I'm not expert enough to set it up without a fair bit of research and fddling around

Just because it isn't perfect doesn't mean it's not worth doing, though
madact is offline Find More Posts by madact
Reply With Quote
Old Mar 25, 2011, 11:19 AM
Registered User
United States, KS, Andover
Joined Oct 2005
539 Posts
Quote:
Originally Posted by Montag DP View Post
No, this stuff is interesting, and not annoying at all. It is a little off-topic, though, so feel free to start a new thread if you want. I haven't really done anything with my optimization since the last update. I have too many other things that need attention.
I haven't had time to touch my optimization stuff either. I guess I'll see how well I documented things.

These ideas are interesting, but maybe more challenging than it appears. I'm staying quiet, as I do this stuff for a living and really can't talk about the methods we use. I will say that for the things we (modelers) do, XFOIL is a real value. RANS methods are not likely to do the trick. Transition models, while they have been studied, have not been developed nearly as far as turbulence models have. In reality, very little laminar flow occurs in full scale airplanes, and most RANS modelling assumes fully turbulent flow. Viscous grids (3D) are very challenging to do well. I've seen lots of claims of easy grid generation, but in practice, it is usually not that easy. It really depends on the physics you are trying to capture. Truthfully, at our low Re, it's very hard to say what the truth really is, anyway, which brings us back to the value of XFOIL...
markschaffin is offline Find More Posts by markschaffin
Reply With Quote
Old Mar 26, 2011, 06:50 PM
Registered User
Joined Mar 2011
2 Posts
Hi Madcat, Truffaldino andthe rest of the gang,

Quote:
Originally Posted by truffaldino View Post
Hi Andreas,

It would be nice if you could start a project to run CFD simulations for whole wing or airplane rather than an airfoil ("virtual 3D wind tunnel"). For instance something like XFLR5, but with Large Eddy Simulations, rather than vortex lattice method supplemented with x-foil polars: User will send you an aircraft geometry in AVL/XFLR format and will get polars as an output. ...
That's in the pipeline http://scr.bi/gpTt5l but it will take some time, because BOINC, the infrastructure we use, doesn't support nodes, so one instance could be solved on one client. and there you will hit resource limits very fast.
but if the mesh and dat will be small, that's doable. but I/we don't have experience with openfoam.
that's why I liked xfoil, because it's easy to start with.

perhaps anyone here would like to help and start such a subproject. It could be you project that's only hosted on Constellation with all credits. I think you have the experience in this forum, you just have to bundle it!

Andreas
hornig is offline Find More Posts by hornig
Reply With Quote
Reply


Thread Tools