Cfd

[Gades]

Hi, I'm interested in CFD, and I was wondering if there is any freeware (or available for students) that would admit as input a profile from a CAD file, or an xls file.... in fact any form of file; as soon as I can define my own shape. For example a NACA-like section, if it's a 2D CFD; or the intersection between keel and hull, if it's a 3D CFD.

Also, I'd like to know if anyone knows of a CFD written in C. What principles do you need to write a CFD?

do you need the code or you need a 3d model of an actual boat?
i'm looking almost for the same but if you need a model to test i can provide one

[Steve Hollister]

Hello From New Wave Systems

"What principles do you need to write a CFD?"

If you have to ask, then you don't want to know the answer. Many people spend their their whole careers working on CFD.

You can go to our web site www.newavesys.com in the "Links" section under technical journals and symposia to find many links related to CFD, including an execlent recent tutorial article by Bob Beck (my old advisor at U of Michigan) and Art Reed. This will get you up-to-date on potential flow theory and RANS codes.

Generally, the CFD codes are very expensive if they are commercially available. Usually, companies will contract with a CFD company to do a particular analysis. The only inexpensive CFD program (free actually) that I know of is called Michlet, by Leo Lausaskas, but it is just for slender hulls, like canoes and kayaks. There is a like to his web pages from our link site.

A recent book that you can get on the subject is called "Practical Ship Hydrodynamics" by Volker Bertram, Butterworth-Heinemann. Check for this on amazon.com. However, if you get headaches when you see integral signs, then this book is not for you. In fact, skip the thought of writing a CFD program.

Take a look at Michlet, however, since it it relatively easy to use and our hull design software outputs a Michlet input geometry file.

Regards,
Steve Hollister

[Stephen Ditmore]

Not that I'm an expert, but I want to just add to that explination the words "Navier-Stokes Equation." It's my understanding that this equation is the theoretical basis of CFD.

[Steve Hollister]

Hello Again,

Not that I'm an expert either (and I even specialized in hydrodynamics for my master's degree!), but simplistically, there are really two parts to CFD - potential flow (based on Del-Squared-Phi - irrotational flow) and turbulent flow (based on the Navier-Stokes equations - Reynolds-Averaged Navier Stokes Equations - RANS codes). From the little I have read lately, many in the CFD world have been trying to model both of these areas using "super" RANS codes. I understand that this has not been going very well. The models become HUGE and still take an enormous time to solve even using the fastest computers. There is also the sticky problems of verification and validation. The best bet is to read the Beck and Reed paper I mentioned before. Perhaps I should do that too, if only I can set aside a few hours (days!) to read it.

Steve Hollister

[Stephen Ditmore]

If someone would like to check out http://www.e-cfd.com/ , I'd be interested in what you think. My understanding is that the "digital physics" used by the Exa Corporation is based on a different algorithm than the standard Navier-Stokes based codes, allowing it to run faster, though perhaps sacrificing some accuracy.

[Gades]

>>If you have to ask, then you don't want to know the answer

Well, actually I do know about some of the equations you're supposed to use. I've studied Navier within my Mathematics and Hydrodynamics subjects at the University.

>>Check for this on amazon.com. However, if you get headaches when you see integral signs, then this book is not for you

I've covered double and triple integrals theory, I'm sure I'd enjoy that book. The problem with CFD books that I've seen is the price!! At least for a student


I've just been looking through your stie www.newavesys.com and I think I've got some download to do. Thanks!

About the different methods that you've mentioned, I've read that one of the points with CFD is to know which one you should apply for each case. At least this is what I was reading yesterday in a paper called: APLICACIÓN DE HERRAMIENTAS CFD AL DISEÑO
HIDRODINÁMICO DE UN VELERO TIPO COPA AMÉRICA, wroten by the team who worked for the development of the Spanish America's Cup boat.

Three years ago I went to a visit to the towing tank installations at El Pardo (Madrid), and they mentioned something about CFD. We were told that it'd be a long time til this technique could replace the actual towing tank tests; but, how closed are we? Or how much can we achieve by now?

Thanks for the replies.


P.D.: it's the third time that I try to reply to this thread, but my computer seems to have problems with this url; 3 out of 4 times that I try to login my computer crashes.

[Steve Hollister]

I just looked at EXA Corp and its "Digital Physics" technology. It's hard to dig through the PR and hype, but I do like any solution that separates the complexity of the geometry from the complexity of the lattice or grid used for calculations. In the traditional meshing process, difficulties arise whenever two surfaces meet or intersect, since you always want the meshes of the two surfaces to meet exactly, vertex to vertex. The mesh can get very weird near the intersection. Those who have seen STL triangle representations of intersecting surfaces know what this means.

Their solution is to divide the whole 3D domain into a refined, gridded cube of voxels and surfels. These are geometric modeling terms often used when modeling digitized models, like internal organs of humans taken from MRI machine data. Then, it seems like instead of solving the N-S equations directly, they simulate the actual movement of the molecules (voxels) of the model statistically. (I really need to study it some more.) My gut feeling is that it is more accurate in some ways, but less accurate in others - there is no free lunch. It seems like for any kind of accuracy, you would have to model it with billions and billions of voxels and surfels. If the voxel grid is not tuned to the shape of the model, then it must have to be very, very large.

I don't know how this applies for vessel CFD, since boats and ships have to deal with the free surface of the water. For traditional marine CFD methods, the program solves for the wave pattern along the side of the hull; then the program has to remesh the hull to match this new waterline shape and then do the calculation all over again. This keeps going until an equilibrium solution is reached. The wave surface is why it is unlikely that general, commercially available CFD codes can be applied to boats, unless you are looking at geometry that is well below the free surface.

Perhaps with the EXA approach, one could model the free surface with a different type of "surfel". Philosophically, I like the idea of the approach. In many applications, I see mathematicians trying to handle an overall problem by solving a set or system of equations. For complex geometric problems, I think there is a great potential for solving problems using algorithmic or simulation methods.

Steve Hollister

[Steve Hollister]

In response to Gades' post,

From what I've been told, about the best repeatable accuracy that you can achieve in a towing tank is about 1/2 of a percent. The accuracy from a CFD program is fairly unknown because there are no good standard test cases for validation. However, there are two kinds of accuracy: absolute and relative. Even if a CFD program cannot calculate absolute resistance numbers within 3-5 percent of the real, full-size vessel, it still might be accurate enough on a relative basis to compare boat A versus boat B. If the meshing and CFD calculation are relatively fast, then you can process many systematic changes in hull shape. Although the absolute values of the resistance are not as accurate as you would like, you can still spot trends and identify promising hull shapes. That was the idea behind the SNAME paper I did on Automatic Hull Variation and Optimization.

Outside of the techniques described at the Exa Corp. site for "PowerFlow", the best place to start is the paper I mentioned by Reed and Beck.

Regards,
Steve Hollister

[Stephen Ditmore]

Steve:

As usual I'm really enjoying your insights (sorry I haven't become a customer yet, but the day will come!). My (actually a student at Trinity College working with me) experience using FLUENT to model lift/drag on a sailboat keel is that the drag numbers appear to be way off, mirroring the experience of John Letcher et. al. when they used VSAero to design keels for Dennis Conner in '88, and others I've spoken to. Do you have any suggestions for getting accurate drag numbers for keels?

Also, did you see the paper by the German group on hull optimization presented at the last Chesapeake Sailing Yacht Symposium? I'll take a look at your paper when I have a chance (though that may not be 'til late January).

[Gades]

I wanted to share this extract from the paper I've mentioned before. It's about the decision they took for each section of the study:

· Resistencia viscosa de la quilla: (flujo 2D) código potencial con acoplamiento de capa
límite 2D. Calculo de Cd 3D integrando el Cd para cada sección.
· Resistencia inducida de la quilla y viscosa e inducida de casco-bulbo y timón: Código
de paneles 3D con acoplamiento de capa límite cuasi 2D.
· Resistencia de ola del conjunto: Código de paneles con superfície libre.
· Comprobación resistencia total del conjunto de apéndices: Código de Navier-Stokes

I'll try to translate more or less (I guess you don't understand Spanish):

· Viscous resistance of the keel: (2D flow) potential flow with coupling/joining? of the layer.
· Induced resistance of the keel and viscous and induced resistance of the hull-bulb and rudder: 3D panels code with coupling/joining? of the cuasi 2D limit layer
· Wave resistance of the whole: panels code with free surface.
· Check the total resistance of all the apendices: Navier-Stokes code.


Steve: so, according to what you say, I could calculate that between model A and B I get an increase of 10% resistance (for example), and apply this knowledge to the results I've got in previous towing tank tests. Then you could make let's say 10 different CFD model/tests and two towing tank tests, to get accurate results?

[Stephen Ditmore]

I searched for CFD/drag prediction software at the software section of this site. Unfortunately it appears Michlet http://www.maths.adelaide.edu.au/App...et/michlet.htm is no longer available for download, but the links below look promising:

http://www.hsva.de/services/cfd/wave/
http://capella.colorado.edu/~laney/software.htm
http://www.zwakenberg.de/hulldrag/
http://www.techsail.com/

[Steve Hollister]

Let's see, where am I?

1. ...getting accurate drag numbers for keels? I don't have a lot of direct experience evaluating the results of CFD. My work has mostly been centered on converting the hull geometry in our software to a proper CFD mesh. I do know that VSAERO is a potential flow code that doesn't take into account the free surface. The company that developed the software - Analytical Methods (I think) kept developing the software and added a WHIP (Wave-Hull Interaction Program) to take care of the wave surface. Generally speaking, prople have used potential flow codes for evaluating lift and drag comparisons for airfoils "far away" from the free surface. I don't know how far "far" is and I don't know what the loss in absolute or relative accuracy is.

2. paper by the German group on hull optimization at CSYS. I dug it out of my BIG pile of papers I really have to read and started to look at it again. (You know, the most interesting projects to work on are the ones that nobody will pay you for.) It is interesting to see that the authors are from TU-Berlin and have worked with Dr. Horst Nowacki, one of the professors I worked with when I first started at Michigan 25+ years ago. Dr. Nowacki has always been interested in the process of automatically creating hull shapes from form parameters like Disp, LCB, Cp, Sectioanl area curve, etc. (See the book he edited and contributed to called: "Computational Geometry for Ships published by World Scientific in 1995) It's surprising that these techniques have been developed over 30 - 50+ years (the Lackenby technique for hull variation I use was developed in the early 1950s), but you don't see them commonly implemented in commercial software. Well, actually, the authors of this paper are trying to do this. See their web site at www.friendship-systems.com. There are also some good technical papers there in English and German. I approach the problem a little bit differently. I assume that you already have a parent hull shape and I apply hull variations (stretching, shrinking, shifting, etc) to the hull to create a child hull with a specific displ, LCB, Cp, parallel middle body (for ships), and draft. Without a parent hull to start with, their technique will only generate one style or class of hull shape, unless they write a different program for each style. With my parent hull approach, the parent hull shape defines the style or class of hull shape that will be created. Once the child hull has been created, then you can run the shape through an analysis program, generate the appropriate results (measure of merit, etc.) and then automatically generate a new hull shape. I could go on, but you should read their article and my article to get a better idea of what can be done. I did add my hull variation software to a version of ProSurf that I have on my computer to test its ability to take a parent NURB surface hull shape and generate a child shape with target values for displ, Cp, LCB, etc. I wanted to get it into the current version, but it's not quite done and will have to wait for the next version.

3. Different CFD methods for different problem areas on the hull. Yes, this is the most realistic way of doing CFD if you have to get some kind of real answers now. The idea of a super RANS code program that can do everything sounds nice, but it may not be practical if you have to get real numbers for a real deadline. Often, what sailboat design groups will do is to use the simple LPP/VPP analysis to identify promising areas of development. Then, these shapes will be fed into the CFD program(s) to narrow down the promising cases. Then, the large (6+ meter) models are towed in the tank and compared. Finally one or two full-size boats are built, wired up, and the results are measured. The problem with the America's Cup is that you are dealing with very minor differences in performance - so close that it is unrealistic to expect that kind of accuracy from the performance analysis programs. That is why I think that groups need to spend more time analyzing the complete parametric design space of the America's Cup rule using some sort of automatic hull variation connected to a CFD analysis program.

4. For many other CFD links, see the Technical Papers/Journals links on my Links page at www.newavesys.com

5. Michlet - I checked and you're right - Michlet is no longer provided due to lack of donations. Leo, say it isn't true! You should just charge people some fee rather than wait for donations. Oh well, my program now has an output format that can no longer be used!

Regards,
Steve Hollister

[Stephen Ditmore]

Thanks again to you both. Another question for either of you going back to 1988 America's Cup research: Dan Greeley, working for the NYYC's America 2 campaign, summed the angular momentum in the predicted turbulent wake of hull and appendages in order to determine the induced drag using a "Krevitz plane" behind the vessel, normal to its direction. Is this something that can be done with today's CFD codes, and would it make sense?

With respect to the free surface, I saw an interview with Dave Egan on the web after the last AC in which he basically said (I think) that one really needs to use both types of analysis. The logic as I remember (reliability questionable) was that the keel is initially analized and detailed without the free surface in what amounts to a virtual wind tunnel, then the wavemaking drag and interference effects for the hull/keel combination are assesed using free surface modelling. Would that make sense to you, Gades? You might be able to locate the interview by searching on Dave Egan's name. As far as I know he still works for SGI.

[Stephen Ditmore]

Leo's site: http://www.maths.adelaide.edu.au/App...azausk/leo.htm
, is interesting. I hope the Michlet problem gets resolved!