Delft Hull Series

Thanks Mikko, interesting, but they are using a sledge hammer to crack a nut. Extrapolating from tank tests of 3 geosims to full size!!!
I stay with my simple boundary layer computation.

 

Hello Urlich, I think the paper reads very well. The presentation is excellent.

I am still a bit uncertain about the robustness of the way you went about selecting the regression terms, though. I'll say up front that regression technique is not my thing.

The narrow, short term goal is to separate the models' measured total drag into scalable components so that the resistance of full scale hulls can be predicted. And you have no full scale (or other scale) hull resistance data with which to validate the final components.

You begin with an improved viscous estimate. It includes a girth integral, a hydraulic diameter, and some sort of base area term for the separated case. You then subtract to get a residual resistance, and do a regression study on the residual only. And you do this separately for each Fr. This is where I suspect there is a problem.

If you did the regression over all the Fr at once, then I think you can argue that terms in viscous model and terms in the wave model are adequately noncorrelated because one scales with Re and the other scales with Fr. But if you are only looking at one Fr, I don't think this idea holds. I'm not sure how well it holds even if you are running multiple Fr.

I think the regression studies for a single Fr should include the viscous model, or its factors, when looking at correlation figures and selecting terms. In addition to the correlation with your residual Y, You should also test for correlation with Viscous Resistance and its factors -Integral G (piece wise, say 2 pieces along the length), D, and Base Area where applicable. Perhaps a practical way to test this is to just go back and run your forward selection algorithm on the total resistance and compare the sets of terms.

It's all rather murky in my head, but I just can't see that the selection method is the one best suited to the particular goal of scaling by Fr.

One other quibble. You really ignored the angle terms in your model. Just because they are dimensionless, doesn't mean they can't be normalized. Take the deadrise angle, for instance. It has a high correlation with Ax. To eliminate that, find a deadrise angle that is representative of Ax and divide the actual by the proxy. You could look at a quadrilateral with the same beam, draft, and area and use it's deadrise as the proxy.

 

Phil, I guess we discussed something similar already 4 month ago. I am glad that you appreciate my presentation. I do not understand your rational, that let you say:

I am using the basic method of W. Froude that has not been questioned during the last 100 years. The separation into viscous- and waveresistance is valid for a single test point at one specific speed and the extrapolation to full size is straight forward. It is another story, if one is searching for the relevant hull parameters, that determine the waveresistance. In this case one should look at a whole range of FN and expect that the parameters with the major influence are the same.
There is no need to determine the viscous resistance with the help of any kind of regression. The boundary layer calculation is much more accurate.

In the internal process of the matrix inversion all parameters are normalized with their statistical means. An additional normalization outside of this process does not change the results.
Uli

 

My bad. Normalized was the wrong term. The correlation between the two can be greatly reduced.

 

"the basic method of W. Froude that has not been questioned during the last 100 years"

My understanding is it has been questioned many times over the last 100 years, and found to be very useful but not exact.

Reynold's number can affect wave drag, particularly at low Re as in tank testing of small models. Changes in boundary layer thickness will affect wave formation, though the effect may be very small. Changes in separation and the viscous wake which may as Re changes can have a considerable effect on wave formation. The latter effect may have been responsible for the difference between predictions based on tank test results and the actual performance of Mariner, a 12 meter with a submerged second transom.

Possibly a larger effect will the changes in viscous drag with change in Froude number. As the waves around the boat change with Fr the velocity field and pressure near the surface will also change. This can affect the boundary layer and separation locations.

 

This is the reason why I wrote in my paper on page 2 that "Froude scaling to full size is never 100% correct".
If you read my paper you will see that I tried to correlate the residuary resistance to the Reynolds number, because of the impact viscous effects can have on the wave making. Unfortunately the Delft test matrix did not allow this regression because of selection bias. This is described in detail in chapter 4.3.

 

I didn't mean to interfere with your project but I find it kind of interesting that even form drag should be scaled from model to fullsize. The differences in the form drag coeff K are significant in the paper.

Interesting also how skin friction is related into the boundary layer close to the hull, while form drag is related more into the vorticity in the wake further away.

Another thing that puzzles me is how you could allow for the fact that for sailboats with long aft overhangs, wet area changes so much from low Fn (where the water is not touching the overhang) to intermediate Fn where the overhang is all wet, but it hardly contributes to waterline length yet... The reference wet area you calculate your friction over is the static one, and with zero trim. At speed, and trimmed, the wet area can be rather different in size (I should look at that), and certainly very different in shape.

 

This thread is very helpful for me. One of my questions is always "did I communicate my ideas properly". It seems that I have to add a few more lines to my paper.

In my opinion Froude's idea, to separate viscous- and wavedrag and to use this in tank testing, was the most brilliant step in the history of naval architecture. The trouble is of course how to determine the viscous resistance. As you and David in post#20 explained, there are many effects caused by the wave elevation along the hull and also Reynolds number effects on the wave making and the interaction between waves and boundary layer (e.g. separation). It is therefore impossible to determine the "true" viscous resistance. I think it was a wise and practical decision by most towing tank operators to define the viscous resistance to be the drag of the double body, i.e. the drag without waves, assuming the water surface at the still water level. All the additional effects that you and David listed are then lumped into the "residuary resistance".
Because of these viscous "left overs" in the residuary resistance, there is for sure a dependency of it on the Reynolds number. I was hoping to find this out when correlating the Delft residuary resistance to the Rn. Unfortunately the experiments at Delft were designed in a way that masked this dependency (see chapter 4.3).

The best way to measure the so defined viscous resistance, is to tow a submerged double body in the tank, or measure it in the wind tunnel. There are many reports about this practice in the literature (e.g. Inui, Larsson). The method to calculate the viscous resistance from flat plate data and multiply this with a form factor is only a very crude replacement, at least for yachts that do not have a long midship section with parallel sides. The reason why, is shown in figure 1 in my paper. If no double body test is available, a better way is a computation that makes use of a RANSE-solver or any other kind of boundary-layer calculation.

I use a slight adjustment in that way, that I calculate the trimming moment that is caused by the pulling force at the towing point at a defined hight above the still water level + any additional trimming moment imposed by moving weights on the model. This trimming moment is applied in the hydrostatic calculation at rest and changes the form of the wetted area. The effects of the changing attitude of the hull at higher speeds is of course still contained in the residuary resistance, as described above.
Does this make sense to you?

 

It has been known for a long time that overhangs can be a problem for
many non-linear programs. Most of those programs start with a linear
solution. The first iteration might produce a wet under-side of an
overhang. After trim has been calculated, the next iteration is made
and it might predict that the overhang is dry. After the trim is
estimated a 2nd time, the next solution might predict a wet overhang...
Numerical damping can help sometimes. Or, look at the experiments
you trust, construct the mesh for the CFD code accordingly, and hope
the solution gives you results that agree with the experiments.

I agree with Uli that Froude's hypothesis was a wonderful advance.
Measuring viscous drag and wave resistance separately and then adding
the results should get the total to within 10% for very small models
and maybe to +/-5% for larger models. Differences can be quite large
at very low Froude numbers when transoms are not running dry, and
repeats of experiments can be very scattered. The experiments with
NPL hulls made by Couser et al bring this out very well.
It is unfortunate that, as Uli noted, the Delft experiments are not
repeated at some Fn. To get as close as he has with the Delft data is
impressive, and I doubt he can do much better given the many
uncertainties with the experiments e.g. different BL trip locations,
different or unknown towing points etc.

Rn effects on form factors has been noted by the ITTC in the last
couple of their conference proceedings. Maybe someone will suggest a
way to proceed at the next conference.

After 100+ years of experiments and beard-scratching I don't think the
problems have been fully resolved, despite the excellent results shown
in advertising brochures for some codes.

 

Remmlinger, It would be helpful if you include an appendix with high quality hull sections drawings. I know that is asking much in comparison to what we are paying you....

This thought was brought on by Table 2 where Pearsons(Cx, T/L1) = -0.77. It begs the question "Do long skinny hulls in the Delft series have less full sections?"

 

The CAD-files are all available at
http://dsyhs.tudelft.nl/dsyhs.php
in the form: IGES, point files, Maxsurf MSD and JPG linesplans
I guess I do not have to copy that.

r=-0.77 indicates that Cx increases as Tx/L1 decreases. In other words a small draft causes a fuller main section, whereas a large draft is combined with a main section that is closer to a triangle. I guess this happens, if you keep volume, length and beam constant and just increase the draft.
Uli

 

Yes that is clear.

Regarding Fig 3, Pearsons(BM/L) is -0.7 to -0.8 for 0.45<Fn<0.65. This implies that drag reduces as hull volume is moved below the water surface. Extrapolation beyond the Delft hulls points toward bulbous bows and SWATH.
However, could this correlation also be a result that a more full hull has less beam for a given length, volume, and draft? The reasoning being that higher metacentric height for a given section area and draft implies a more triangular section with wider beam.

 

I think one must be extremely careful when looking at Pearson's rhos in this manner. In the case of the Delft models, the rhos are not representative of the independent design choices a designer might make, but instead represent a measure of the parameters that were used to create the child hulls from the parent models. One important feature of the Pearson's rhos are that they are invariant with respect to linear scaling of either or both variables. If a child model was constructed using a linear transform of the parent model's geometry, then it is as if there were two copies of the parent in the sample. To a fair extent, there are only three different models, with many copies of each, being looked at by the Pearsons.

 

Yes exactly my point.

So maybe the only parameters that should be included in the prediction are those used to create the variants (child hulls) within a series? then after that, a representative "NewParent" could be produced for each series which would have equal length, volume, and righting moment compared to the other NewParents. The drag of each NewParent could be estimated for each Fn interpolating fron the towing data for it's series. Finally the NewParents could be compared and a system of parameters that quantifies the differences between each series NewParent could be constructed and then correlated with the drag estimates.

 

The thread hasn't really looked at drag prediction. That is a different kettle of fish and I was trying to deflect the conversation away from that. Those statistics are a great temptation, but looked at singularly, they are misleading.