H.M.S. Atalanta - Drafting my own plans

[Ben752]

Hi everyone,

 

I wanted to give an update on my progress.  Normal life stuff has been busy since the last post.  My company was acquired, subsequent expansion of our office here in the U.K., met a nice English girl and more focus on my U.S. Brig Niagara.  All sorts of great things.

 

Anyways, enough about that.  As I was progressing through later parts of drafting, I came  to realize I wasn't really happy with how some of the lines were formed.  Also, AutoDesk launched CV splines for Fusion 360!

 

I ended up starting a starting over my 3d model to incorporate lessons learned about Fusion and feedback from ya'll.  Also, I stopped straddling the fence on how much to use from the TFFM plans and the original drafts -- I've decided to go using the original references almost exclusively.

 

In this second start, I've switched to the following techniques in fusion:

 

1. Using CV splines for the body plan and striving to keep the same number of control vertices
2. Defining key curves using CV splines that are fit using numerical optimization methods.

 

This last part took some R&D (I work in technology and spend a bit of time doing data science/AI type solutions).  Below is an example of what I've done relating to this.

 

I started by taking measurements off the drafts and correlating them across plans.  This yields points to fit, one challenging is that fitting CV Splines isn't a common practice.  This is partly because they are parametric which to my knowledge doesn't have an analytical solution.  What I did was use a genetic algorithm to solve for solutions that are very close using the points that match above. 

 

The result seems to be a more smoother fit that is much closer to real curve than I could manage moving points around. FWIW, there are a few naval architecture research papers that describe this technique.

 

Below is an example of a breadth curve fit to the measurements.  One interesting consequence of this is that it yields a fully constrained sketch in Fusion. 

 

 

It seems that splines also tend to work better when lofting.  I think overall it's a much nicer fit than the previous technique I used.

 

 

 

 

[SardonicMeow]

9 hours ago, Ben752 said:

I started by taking measurements off the drafts and correlating them across plans.  This yields points to fit, one challenging is that fitting CV Splines isn't a common practice.  This is partly because they are parametric which to my knowledge doesn't have an analytical solution.  What I did was use a genetic algorithm to solve for solutions that are very close using the points that match above.

Are you saying that you took the coordinates of points along the curve and plugged them into an algorithm, and the output was the coordinates of each of the control points for a spline that fits the curve?

 

[Ben752]

10 hours ago, SardonicMeow said:

Are you saying that you took the coordinates of points along the curve and plugged them into an algorithm, and the output was the coordinates of each of the control points for a spline that fits the curve?

 

That is correct.  It took quite a few tries to figure out exactly how AutoDesk implemented the spline knots (they actually confirmed it in a support request in their forums).

 

In the first screenshot above, I used measurements at each station line and it gave me the output of the 7 control vertices (with the first and last being the corresponding points I provided).

 

 

[SardonicMeow]

Neat.  I know intellectually that such a thing should be mathematically possible, but it still seems like black magic.  With a table of offsets, you could get a nice set of curves built in no time.

 

 

[Ben752]

That was my thinking!  My own inability to be happy with moving the CVs (i'd never be happy enough) led to me searching for a different solution.

 

I ended up using the Python library scipy as it has the relevant spline objects in scipy.interpolate and I used the differential evolution algorithm as the solver.

 

 

[bricklayer]

Hi Ben

Your second approach to lofting the hull seems to be the right one. Now you have a single surface.

Before it looked like a "pile of patches". There`s no bad intentions in telling you that.

Does fusion 360 have analysis tools other than zebra striping? Can you apply gaussian curvature to the hull?

It shows the amount of curvature all over the surface in colours. It shows the direction of curvature, too.

Inward or outward.

Is there a tool called curvature graph in fusion? It`s a set of "needles", that point perpendicular to a curve or surface.

Or rectangular to an imaginary tangent at a given point. The length of the "needles" indicate the amount of curvature.

 

In your post #31 two control points are close to one another. At midships. One of the points is located inside the curve.

Does the curve bend inwards over there? Is there a hollow?

 

I`m curious, because I lofted hulls in Rhino. Finally I discarded those lofts. After having applied analysis tools, none of them satisfied me anymore.

My findings are not meant to disencourage you. Not at all. I`ll follow this thread.

 

Michael 

[Ben752]

"a pile of patches" was actually pretty accurate in that it was constructed with several tangent arcs.  I had a similar experience of feeling it was just not right -- and that a different technique was needed.

 

Fusion 360 has curvature combs, draft analysis and curvature map analysis features.

Here is a curvature map analysis I ran, I haven't been able to to see much except when using the min/max settings and cranking the max values way up.  Hopefully this is a good thing (my prior hull showed more variance).

 

The CVs close together are a byproduct of using a mathematical fitting technique.  Fusion using fixed spacing for knot placement of curves.  While I'm not NURBS expert from the match side, my understanding is that if you are using uniform knots then you would need two CVs close together for it to generate the curves.  Now whether they should be split vertically like that, my guess is probably not.  I did run the fitting routine several times and it tended to favor placing them like that.

 

While i'm not totally satisfied with it, i'm trying to remind myself that I aught to start working on other areas and come back to it later.  When I intersected a plane with the hull the pattern for a frame seemed to match up well against other resources I have so hopefully that's a positive side.