Michell v CFD and EFD: Round 2007

[Leo Lazauskas]

Attached are two graphs comparing some very recent (Sept. 2007)
CFD predictions with experimental results for a Wigley hull.
Also shown are predictions using Michell's thin-ship theory.

The graphs show the total resistance coefficient Ct=0.5*rho*U^2*S,
as a function of length-based Froude number, Fr=U/sqrt(g*L), where
g=gravity, L=ship length, rho=water density, U=speed, and S=wetted
surface area. The hull has L=1.905m and L/B=8.

The top plot is for deep water; the second is for water of depth
0.3175m.

Experiments are from:
Millward, A. and Bevan, M.G.,
"Effect of shallow water on a mathematical hull at high subcritical
and supercritical speeds",
J. Ship Research, Vol. 30, No. 2, June 1986, pp. 85-93.

"CFD" refers to the predictions of CFD-SHIP-IOWA 4.00. See:
Sakamoto, N., Wilson, R.V. and Stern, F.,
"Reynolds-Averaged Navier-Stokes simulations for high-speed Wigley
hull in deep and shallow water"
J. Ship Research, Vol. 51, No. 3, Sept. 2007, pp. 187-203.
(This paper also used the experiments of Millward and Bevan for
comparison.) CFD calculations were performed on U.S. DoD
supercomputers. No timing results were quoted, but clearly they are
substantial or results for more speeds would have been included in the
paper.

The curves in the plots were produced using Michell's thin-ship theory
for the wave resistance and two skin-friction methods - the ITTC line
and Grigson's method. (Hullspeed should give very similar results).

Michlet 8.08 took about 5 seconds to produce each curve (81 speeds)
on a 3.2Ghz dual core PC under Windows XP. The attached Michlet 8.07
input file can be used to produce very similar results to those in the
graphs.

Leo.

[yipster]

CFD beeing Computer Fluid Dynamics, and -had to look this one up- EFD Engineering Fluid Dynamics
whats the difference and Michlett is much faster, wish i was, am gonna paste that wigley in tho, a great work and thanks again

[Leo Lazauskas]

Originally Posted by yipster

CFD beeing Computer Fluid Dynamics, and -had to look this one up- EFD Engineering Fluid Dynamics
whats the difference and Michlett is much faster, wish i was, am gonna paste that wigley in tho, a great work and thanks again

Sorry, Yipster:
CFD = Computational Fluid Dynamics
EFD = Experimental Fluid Dynamics
Michell = Dead D00D.

[marshmat]

Michell = Dead D00D.

Dead d00d whose algorithms give numbers pretty close to reality in this (admittedly somewhat idealized) case, in about 0.003% of the time a complex CFD code takes to get only marginally closer (if at all) to the experimental result, and which can be done on a machine worth 1/1000 as much.

Nice work, Leo. We physics nuts thrive off such data!

There's a Michlet 8.08 now? I have yet to find any bugs in 8.07, so you must be tweaking something fun....

[Leo Lazauskas]

Originally Posted by marshmat

Dead d00d whose algorithms give numbers pretty close to reality in this (admittedly somewhat idealized) case, in about 0.003% of the time a complex CFD code takes to get only marginally closer (if at all) to the experimental result, and which can be done on a machine worth 1/1000 as much.

Nice work, Leo. We physics nuts thrive off such data!

There's a Michlet 8.08 now? I have yet to find any bugs in 8.07, so you must be tweaking something fun....

There's still a lot of room for improvement for both CFD and Michell-like (or "inviscid CFD") methods. CFD can give useful results for flow near the hull that Michell methods aren't able to handle. Michell's strength is in prediction of far-field wave patterns and wave resistance.

Sure, the Wigley hull is an idealisation, but if CFD and other methods can't give reliable predictions for that simple hull, what faith can you have in their capabilites with complicated geometries?

Michlet 8.08 is my "working" version. When I have time I move useful routines from my other codes into Michlet and then release it when it seems ripe. Michlet 8.08 has a couple of extra hull series, e.g. Series 20 uses 20 parameters to describe the hull. This allows for more complicated, (i.e. wobbly) geometries than Series 1 to 8. Optimising 20 parameters can be a bit slow, and ultimately the hull shape ends up smooth so that fewer parameters could have been used in the first place. I'll try to release it early in the new year.

Regards,
Leo.

[marshmat]

I've spent enough time with Fluent to know better than to trust CFD results of any kind.... but it is a very useful tool, especially in the later stages of fine-tuning a shape.
I'm quite impressed, really, that Michlet works as well as it does. A relatively quick algorithm based on well-defined equations, that works for a remarkably broad range of shapes and gives results that, within a reasonable design envelope, are a pretty fair approximation to the real thing.... not something you come across every day.

[Leo Lazauskas]

Originally Posted by marshmat

I've spent enough time with Fluent to know better than to trust CFD results of any kind.... but it is a very useful tool, especially in the later stages of fine-tuning a shape.
I'm quite impressed, really, that Michlet works as well as it does. A relatively quick algorithm based on well-defined equations, that works for a remarkably broad range of shapes and gives results that, within a reasonable design envelope, are a pretty fair approximation to the real thing.... not something you come across every day.

Yes, it still holds up very well. According to E.O. Tuck, Michell's thin-ship theory has been called the first ever "engineering design" formula, in that it gives an output quantity of design interest from an input that specifies the actual (unrestricted) design geometry.

After Michell published his theory in 1898, it was promptly forgotten and only rediscovered about 30 years later. I'd hazard a guess that it is one of the most cited Australian research papers.

Leo.

[yipster]

Far as I’m concerned Michell isn’t dead than and hope you dont mind me liking visuals too
A while back I compared some hard chinned slender hulls in Michlet and was very impressed
Ah off course, a talking horse; Dead is Dood in dutch as 死者 is dead in Chinese
I’m flattered and sorry for the confusion taking it for D00D (stands for dude in cyberspace)
since i'm on it now can i ask if Michlett accepts keel and rudder appendages from freeship ?

[Leo Lazauskas]

Originally Posted by yipster

since i'm on it now can i ask if Michlett accepts keel and rudder appendages from freeship ?

Sorry, it doesn't accept appendages from Freeship. You have to treat them as part of the hull by supplying offsets.

Regards,
Leo.

[Guest625101138]

Originally Posted by Leo Lazauskas

Attached are two graphs comparing some very recent (Sept. 2007).....

.....The curves in the plots were produced using Michell's thin-ship theory
for the wave resistance and two skin-friction methods - the ITTC line
and Grigson's method. (Hullspeed should give very similar results).

Michlet 8.08 took about 5 seconds to produce each curve (81 speeds)
on a 3.2Ghz dual core PC under Windows XP. The attached Michlet 8.07
input file can be used to produce very similar results to those in the
graphs.

Leo.

Leo

No need to convince me of the virtues of Michlet but since you started the topic - I do not understand the strict distinction you are drawing between Michlet and CFD. This may not be the place for a maths lesson but how wrong is someone in describing Michlet as CFD?

What is the clear distinction between Michlet and CFD? What is at the core of programs described as CFD that is different to Michlet?

One of the interesting observations I have made using Goszilla is that a boat constrained only by displacement and power will have very little power absorbed in wave making so the viscous drag dominates. As long as you have a good idea of what shape will minimise the waves it really does not much matter how accurate the calculation of the wave energy is because it is going to be low relative to the viscous drag. I realise this is a paradox but as long as the wave energy for a particular shape is in the ballpark and RELATIVELY higher or lower than the next shape it will lead to the optimum solution without the need to have great precision in the wave energy calculation.

On the other hand, if there are constraints other than displacement and power then the wave drag could have much greater significance so in this case it is nice to have good precision in wave power calculations.

These days I am more interested in optimised props than optimised hulls - I am satisfied with Godzilla. So another plug for the prop optimiser. Happy to assess, build and test resulting designs.

Rick W.

[Leo Lazauskas]

1. The main distinction between Michlet and CFD is that CFD divides the
flow field into very many small volumes and then attempts to solve the
"Navier-Stokes" equations within those volumes. Michell's thin-ship
theory is based on a simpler subset, called "Laplace's equation", and
it does not divide the flow field into small volumes.

In thin-ship theory, the hull is represented by an arrangment of
"sources" and "sinks" on the hull centreplane. Michell's genius was to
show that the strengths of the sources and sinks were related to the
longitudinal hull slope.
The whole procedure can be reduced to evaluating certain integrals
accurately.

The great problem for CFD is one of scale. Close to the hull, in the
boundary layer, the small volumes must be tiny. But if the same
size volumes are used everywhere, then astronomical numbers must be
used to estimate the flow far astern of the vessel.
Another difficulty for CFD is that it attempts to model turbulence
inside each small volume. Like time, turbulence is perfectly understood
until one thinks about it or attempts to write a sensible equation.

2. Regarding your observations that viscous drag dominates...

There's no paradox. It depends on the speed regime you are designing
for. At low speeds, viscous drag dominates. In fact this is where
Michell's theory is not at its best, so it has been great to hear that
you have found reasonable agreement with predictions. Incidentally, do
you use the ITTC line or Grigson in your work?

For many monohulls, the wave drag hump is at F=0.564, which is well
above the Froude numbers long hulls operate at.

A last point. Without constraints, you should always be able to find
at least two optimal vessels: one with low skin-friction and high wave
drag; another with high friction and low wave drag.
Constraints push the design one way or another.

3. These days I am more interested in optimised props...

About 15 years ago I went to ask E.O. Tuck for some advice on a
vertical axis wind turbine I was working with. He laughed and said that,
while he supported my attempts, he was still trying to understand a
single flat wing at small angle of attack. What a typically useless nerd answer, I thought at the time. Soon after that I abandoned wind turbines and
started modelling simple flat wings...
I haven't finished that yet.

4. I know very little about CFD. I thought it would be of interest to present recent comparisons with experiments and theory to put things in perspective for any prospective users of the methods. I didn't say anything about the experimental data in my plots, but they could be crap too. (Just in case anyone was thinking of putting their faith in towing tank deities). I was also moved to post because the CFD results were by Frederick Stern and colleagues. F. Stern has been a great force pushing for proper verification and validation of CFD methods. He's the closest thing to Penn and Teller in CFD

Hopefully others can help you out with the CFD definitions if I've written anything too barbarous.

All the best,
Leo.

[Guest625101138]

Leo
Thank you for the maths lesson. I still have a view that Michlet is "Computational Fluid Dynamics". It could be more precisely described as full-scale CFD as opposed to finite element CFD.

I was reasonably adept at Laplace Transforms when doing studies. I actually used them in my first year of employment to assess a large DC drive stability problem. Soon after that I discovered programmable calculators and started working in discrete time domain - never looked back and never used Laplace Transforms since.

I have never changed the viscous setting from ITTC1957. So I have not compared the results with Grigson.

I feel the ITTC number is reliable as low speed correlation is good. My performance data indicates that Michlet produces low numbers when the wave drag is about equal to the viscous drag but this is typically around 4X my design power level and measurement accuracy is also low at that level. THere are significant biomechanical factors as well. I have got very little direct measured power data. Most of my power measurements are dependent on my body calibration. This has good repeatability to within 3% but is no better than 5% accurate in absolute value. There is also test variation in wind and waves and these are significant when you are designing low drag hulls. It can take a few weeks to get conditions calm enough to make reliable comparisons.

I have some direct power measurement on some recent testing that I posted last week:
Electric Boat Data
I still have some gaps in loss determination so am working on this but there is plenty of room to find the answers. Most obvious is the wind and waves. Another is that I assumed perfect trim with Michlet and this is not the case. Also I only have the low-load losses for the controller - I could go on.

Rick W.

[Leo Lazauskas]

Originally Posted by Rick Willoughby

Leo
Thank you for the maths lesson. I still have a view that Michlet is "Computational Fluid Dynamics". It could be more precisely described as full-scale CFD as opposed to finite element CFD.
Rick W.

I didn't put much maths into it because a lot of it is at my limit. I'm very lucky to have Ernie Tuck as a long-time mentor.

A better description is that thin-ship theory is a "boundary integral method". (Throw in "classical hydrodynamics" and you'll have respectable, bespectacled girls swooning at upcoming Xmas parties). It is also an "inviscid" theory whereas CFD takes viscosity into account. On the other hand, viscous modifications can be introduced into Michell's integral which is what I have been working on recently.

Leo.

[CGN]

Hi Leo, many thanks for Michlet and glad to hear form you on the forum, I would like to ask you if you think there is any developments besides CFD to calculate wave resistance for "fat" boas like tugs or trawlers (and those not within the scope of Michlet), or the calculations do not allow to develop a way to applied such to "fat" boats

Thanks and keep with the great work

[Leo Lazauskas]

Originally Posted by CGN

Hi Leo, many thanks for Michlet and glad to hear form you on the forum, I would like to ask you if you think there is any developments besides CFD to calculate wave resistance for "fat" boas like tugs or trawlers (and those not within the scope of Michlet), or the calculations do not allow to develop a way to applied such to "fat" boats

I don't know the answer.

My guess is that Neumann-Kelvin methods, non-linear methods, and CFD techniques were all, in a sense, developed to handle "fat" boats and flow situations where thin-ship and slender body theory fail.

Each of these alternatives have problems. Neumann-Kelvin is not consistent. Non-linear methods can sometimes fail when waves are too steep (i.e. on the verge of breaking); CFD requires enormous computer resources; experimental methods are very expensive, and scaling up from model-size to full-size is sometimes very uncertain.

I've heard opinions that numerical ship hydrodynamics is 50 years behind the equivalent state in aerodynamics. That's not surprising given how wonderful flight is, and how many researchers are involved in the aerospace industry.

Regards,
Leo.