CFD background knowledge

Hello everybody,
I want to work myself into CFD. Having learnd the basics of CFD codes at the university and have also gained a lot of experience in ANSYS FEA for grid generation and geometry modelling, my next project shall be CFD.

I know that there are various codes and methods around all being more or less good and feasible for common PC´s.

Does anyone here has experience in CFD for yachtdesign and can give some background information on where to get tutorials, which tool have the best data interfaces to common CAD packages, running times,
RANSE or simple potential flow (what is feasible and good enough for what purpose) and so on.

I know that fluent offer trial licenses as they now belong to ANSYS inc. while shipflow and splash codes doesn´t offer trial licenses, that is another issue as well.

Hoping for a reply best regards and thnx
josch

 

Concerning Ranse codes, the first step is probably to try simple things. Mesh generation is quite different from FEA (at least in multiblock structured), and requires some time to be good at.
For a Ranse code, i ll go for OpenFoam. It has almost everything (or can be programmed), and it ll teach you much more than just use the defaults in Fluent/CFX; to make it work you ll have to understand every parameters you enter (laborious, but very important).

Potential flow codes are a bit different, they are used in some yacht design office (Ranse codes aren't). But as you've said it's difficult to find trials. If you've been to Southampton Uni, you might have a knowledge of Palisupan, this should give you a good idea of what can work.

My point of view is to focus on Ranse, this is the present of CFD (no more research is carried out on Potential flow codes).

 

Spot on, Nico, great advice as always.

It's always worth remembering that if you want to resolve small detail, the mesh needs to be fine in that area. The more detail, the finer the mesh. Mesh size is often limited by the available memory or computing power.

As a point of interest, for any serious CFD, I'd plan on using at least a 2.4GHz P4 with 2GB RAM. You can run several in parallel, of course. Most CFD in single-threaded unless running in Paralell mode, so twin cores won't help you much. Twin CPU, of course, gives you twice the capacity (effectively 2 computers).

Unless you have a reasonable-sized cluster, you can expect steady state runs to take days and unsteady runs (which you'll have to do if you want to capture wave-patterns) will take weeks.

Tim B.

 

Thank you so far, but at that stage I neither see the difficulties and problems in the hardware nor in the time as I am patient. Actually I have more doubts in getting the right numbers out of the programme.

In terms of numerical calculations my experience is that it is pretty easy to generate exciting colourful pictures, but when having a look for instance at the guys from the wolfson unit what they are doing, it seems as if there is still an enormous amount of mathematical understanding and background knowledge necesarry to do such a job properly.

What I hoped to find out ist whether CFD is still such a magic black box, which requires a half of a physics course or whether and under what amount of time and research this stuff can be managed reasonably correct by a common engineer for instance to predict the wave making resistance say +/-10%

 

When I've done it, I'll tell you how long it took.... I started in October and I'm just getting to some (potentially) good answers now!!

The answers of course, are all linked to grid quality. small grids give quick answers, but they won't be right. Big grids (800,000 cells and above) require some big hardware.

Tim B.

 

What needs to be kept in mind is that CFD is a very complex numerical tool, and yacht design a complex field.
*Numerical tool means that grid independence studies need to be done, influence of turbulence modelling needs to be checked, and so on... very time consuming.

*Its application to yacht design is complex: i ll take 2 examples: computation of wave resistance, on a yacht as speed increases, the yacht heaves (moves up or down) and trim by a significant amount. This needs to be taken into account. This add significantly to computation time. Common way around this is to compute yacht position with potential flow, and then compute with Ranse, not perfect. I have seen only 1 publication dealing with this problem. It is not that important (yet) for ship calcs, and this is where money is.
2nd example is sail forces calculation: even upwind separation is present, current turbulence models have some problems dealing with that (mainly to predict where it occurs), and this has a large influence on lift and drag results. This is even worse for downwind sails, current turbulence modelling is nowhere near achieving that. Similar problem (again), there is no money to look at that; planes don't have spinakers, or stalled wings.

But one day we'll be able to do that correctly.

 

I think it depends on how accurate one calculation hast to be.
Bearing in mind that you are dealing on a very high academic level with that problem you are aiming probably a loth higher in your results as I consider doing.
As I said it would be a good starting point to have an idea of what results can be achieved with what efford and time.
For me the only alternative is Delft which overestimates the whole drag picture of a modern yacht, some publications say that potential flow maybe combined with Ranse codes in the front and aft as Shipflow does underestimate the drag picture when compared to tanktests.

So if one would say calculate trimm with potential flow than the rest with Ranse and your result will be approx. xyz%, I think that would be a good starting point. I do know how complex the yacht design field is and I see your point in the unprecise modelling methods of turbulence phenomens. Although I again can not guess who critical in % of drag this is.

 

Actually, my interpretation of the Delft Model is pretty damn close for total resistance. (Nico has seen the results). There is no point in wasting time doing CFD when a DSYS result will suffice. Similarly, there is no point in trying to do CFD on too small a grid, because you'll never be able to trust the answers.

I'm not trying to put you off. Personnally I think CFD is a complex tool that should be used to increase understanding, and you can't do that if you can't be sure if you've captured enough detail.

Tim B.

 

I have seen some research programmes from the Australien University, who did some wave cut towing tank analysis for the Delft hulls. The article has been published at the Chesapeake sailing yacht symposium and that has prooved clearly that the Delf series overestimates the wave resistance by more than 10% especially at high Froude numbers. also when having a close look to the Delft algorithms it becomes obvious that you get the lowest resistance values when designing wide hulls with a moderate displacement at just medium froude numbers. Which is not suitable at all for high performance light weight hulls with high L/B ratios for instance as used mainly for canting keel yachts.
So with Delft you always get penalties in Drag when using high L/B ratios, which I think does not make sense at all.
Another disadvantage I see with Delft are the hull keel interaction and the consequences on the wave pattern. So the wave cut analysis results clearly show that when designed properly you may get a decrease in drag when adding a keel onto a hull, whereas with Delft you always get an increase. Another point are the viscous pressure resistance which defines the form factor those values are usually if no tank tests are performed taken as 1.3 but as far as I know can be computed by CFD.
The final point I want to get answers for is the mentioned trim and sinkage mentioned by Nico, who can not be calculated by conventional method as far as I know but as he sayd "reasonably good" by CFD.
I think by experience, if you know what you want to find out and if you ask the right questions you should be able to get the suitable answers from CFD.
And from my understanding the programme should not bother whether you model a small or a fine grid and from the required efford of the engineer the grid size shouldn`t be an engineering challenge but the assumptions and answers you are expecting are.

Please let me know what you guys think about that I may have the wrong understanding of the underlaying hydrodynamics, but for me CFD is just a tool, which can always be just as good as the user. The at first it has to be clear what you want to find out and than you can decide which is the right tool to use. Perhaps for my questions the panel code is good enough??

 

*About Delft series; if you are looking for the resistance of a hull with modern appendages, at 0 angle of yaw and heel, and at Fn lower than 0.45, the delft series are very good. And i expect it to be quite good at when looking at influence of Cp, Lcb or similars. But hull parameters have to be in the range as defined by Delft.
Hull keel interaction cannot be taken into account correctly by Delft, this would require a large number of parameters and therefore tank tests.
Potential flow (panel code) should work fine to look at that, but it is quite limited at speeds higher than Fn of 0.4.
Best results will be found using complete Ranse, i dont really know how good will potential flow + cfd would be, it has been used (and is?) on IAAC but upwind Fn is not high. I think that errors on high performance sailboats will be higher around 10per cent.
To give you an idea about how much time it would take:
*on a good personal PC (running Linux 64 bit, best CPU + 2go of ram), you ll be able to get good results (numerical error is low enough) for the resistance of a hull without appendages at no heel or yaw (otherwise flow is assymetric; a symmetry plane cannot be used; mesh size doubles). If you know what you're doing, mesh generation could take 4 hrs, and run time 3*24hrs minimum. This is without sinkage or trim, so at low speed. (delft will be better)
*work on sinkage and trim is not that complicated, it is coupling CFD with a 2 degrees of freedom solver, but not straightforward to implement and test. I'd say a 2 weeks job. And run time will increase.
*another example: a complete Volvo 70 with daggerboards, keel + bulb + rudders, at a Fn of 0.5, would require a big cluster. Mesh size will be around 7M cells. Mesh generation would take weeks. And run time a bit less maybe. I dont think it has been tried.

So in the end, i am not sure that cfd is just a tool, a yacht designer will not be able to use it by himself (and he wont be interested in it anyway). Potential flow is maybe a bit closer.

 

Wow, I am impressed!!!
That is an answer, thanks a lot as that does really help to understand both the problem of the grid modelling and also the computation times. This I didn`t know.
So the answer is than clear for me although I should go for RANSE because it gives the better results I guess I will go for potential flow maybe with a combined Ranse grid in the bow and stern region as used in shipflow.

And after your words I do now see the reason why basically no office is doing full CFD studies.
Most of the articles which can be found regarding CFD are from 2002 or earlier and in those artices the problem of the computing power can be understood. I thought that over the last 4 or 5 years the computing power has more than doubled and therefore those problems may not be an issue anymore.
Well...everyday one can learn something new.
Thanks a lot for your answers

 

I did my MSc in Southampton Uni. and I tried Palisupan.It gave some reasonable results for a rudder for varying angle of attacks but for whole body I never tried.It s also difficult to run that programme,a bit complicated.

I advise you to use Fluent which I used in my dissertation project.It gave me 80-85% accurate results compared to the towing tank test results.

 

Hello thanks for that hint. I would have wondered if such stuff can not be managed by fluent
Do you have any tutorials or other usefull documentation to get an overview about the required efford and so on??
Have you done Ranse or potential flow?? Or both in one calculation??

Best regards
josch