An impressive CFD simulation on a Cray XE6 supercomputer

I hope that these "readers" appreciate how valuable long practical experience is
in these matters. I doubt that a NA, fresh out of university, would be able to
understand why some towing tank results, or CFD predictions might look wrong.
Or why, despite brilliant calm-water predictions and tests, the vessel will be a
complete dog in waves.

BTW, I don't include myself in that "practical" group: I had trouble with the
Lego "Pirates of the Caribbean" build!

 

Thanks for that paper AH. Very nice summary of the current state-of-the-art.

So, I understand that the critical issues of the current CFD efforts are:
1) mathematical modelling of the turbulence
2) spatial and temporal discretisation schemes
3) modelling of boundary conditions
4) seakeeping computations
5) grid resolution, hence required computing power, memory and associated costs
6) uncertainty estimation
7) lack of reliable full-scale test data for the mathematical modelling of some aspects of simulations and for a validation of CFD results
8) skill, training and knowledge-updating of the operators.

In other words, all as it was 10-15 yrs ago. We have benefited mainly from the progress in the muscular power of the hardware, which allow us to have a better grid resolution and smaller numerical errors. Is that right, or am I missing something?

How about the tank-tests side of the problem? What are the currently most critical issues?

 

Yes, that's a reasonable summary for resistance prediction, Slavi.

Here's an excerpt from a short paper Tuck and I wrote in 2008. I'm
not sure if the situation is all that much better in 2012.

After reviewing the performance of some 22 rival computer programs in 1979,
Bai [1] was forced to conclude that “. . . wave resistance predictions by
first-order thin-ship [i.e. Michell] theory are rather consistent in
comparison with experimental data and not worse than the envelope of
predictions of seemingly more sophisticated methods . . . "

Ten years later, one of us, Tuck [10] assessed that: “The situation is not
much better today". Nearly a further twenty years later, it seems that:
“The situation is not much better today.".


[1] Bai, K.J. and McCarthy, J.H. (eds),
“Proceedings, workshop on ship wave resistance computations",
DavidW. Taylor N.S.R.D.C., Bethesda, Md.,Nov. 1979.
[10] Tuck, E.O., “The wave resistance formula of J.H. Michell and
its significance to recent research in ship hydrodynamics",
J. Austral. Math. Soc. Ser. B, 30, 1989, pp. 365–377.

From:
Ernie Tuck & Leo Lazauskas
DRAG ON A SHIP AND MICHELL’S INTEGRAL
XXII ICTAM, 25–29 August 2008, Adelaide, Australia.

 

The billion cells ($) simulation would seem flawed at least in that the model appears to be fixed... surely it should heave & pitch a little,

 

If you want to see the state of the art in sailboat simulation, see http://www.cape-horn-eng.com/. I first met with Rodrigo Azcueta in Auckland 10 years ago - he was already then simulating 6-dof motions in the waves. A brilliant researcher, like the whole crew of the company.

 

It's certainly a lovely video, but it seems to be comprised of an inner
rectangular wave field and an outer field that have been inaccurately blended
at the intersection.

 

If you refer to the front page Volvo-boat simulation, that could be just an artifact of a computing capacity-saving measure. But look at the whole site, there's plenty of good info, and Rodrigo certainly claims the time of test tanks is over. The publications are good reading http://www.cape-horn-eng.com/publications.html, if you haven't seen them before. Quite an eye-opener how much intantaneous forces on the hull & appendages vary when passing over waves - not unlike what we found happening over sails. http://www.cape-horn-eng.com/whycfd.html is also a good page to start from.

 

THe costs for CFD are highly variable depending on where you go. I have found that getting my CFD simulations done in Russia is by far the cheapest. I am using the FlowVision code which allows RANS floating body simulations. The results are impressive for both displacement and planing speeds.

The cost of a resistance curve: +- USD1,000

I will shortly be submitting a paper for publication which shows the results for a 60 knots planing vessel for which I have compared CFD, tank testing and sea trial data.

 

Thanks for that info. I imagine that the said cost is relative to a single resistance curve and doesn't include any optimization study?

I'm looking forward to read your paper.

 

Good luck!
Would you have published if the three methods didn't agree well?

 

I'll say! All together...design/construction/testing/data reduction...close to $USD 1,000,000.00.


But we've also built some nice tow tank SES models and leased the Carderock basin and gotten excellent results for around $USD 100,000. The 40m SEMO SES projct comes to mind.

 

Correct. Its was €'s not $'s.
Model and resistance..

Nope, that was a good summary. But Leo has given a far quicker/better response.

 

Just as in Towing tank testing, CFD modelling costs can vary by just as much depending on the resolution and computing time, and also the CFD setup...

A 1 billion cell simulation on a cray i would argue is not really much better / more accurate in a global sense, than a 1 million cell simulation on a rented time server which will cost much less. The CFD setup is far more important in order to get meaningful results - which is why its so important to find a highly skilled CFD programmer to setup the simulation, using the correct methods in order to make a realistic prediction of the particular information your looking for.

There are many businesses doing this type of CFD modelling these days, the costs are coming down with more competition whilst the results are getting more accurate as operators work out which methods produce better predictions in the various types of flow fields... Im actually surprised its not utilized more commonly in naval architecture...

 

Got a reply on Youtube from Wave Animations on my quest about how much time was it necessary for the calculation of the CFD animation in the OP. They say that:
"This simulation took 90 hours of wall-clock time to run 30,000 time steps using 1008 processors. The wall-clock time can be cut in half by doubling the number of processors because NFA scales linearly."
So it's little more than 5 full Cray XE6 cabinets, at around 50 kW electric power each, plus some 50-60 kW for the air conditioning. Running for 90 hrs, means 27000 kWh of electric energy consumed for the calculation. Don't know the prices in US, but in Italy it would cost at least €5000 just for the electricity bill.
Just to put the issue of costs into a more complete perspective.