Hi Danny,

> is MoI eventualy going to implement most of these filleting features?

It should be already, as far as I know... aside from variable radius filleting which I will need to cook up an interface for at some point here.

It's easy to get a kind of wrong overall picture just by looking at a few examples though, much of the difficulty in filleting involves handling a myriad of special cases like many different corner configurations and that only really comes with a large investment of time to handle each kind of thing on a case by case basis.

One thing that I don't currently specifically enable in the IntegrityWare engine is the option to do a piecewise or global merge of the fillet surfaces, which kinds of booleans them into place I guess rather than having them trim into place.

I guess it is possible that could help with a couple of specific cases, but it also seemed to slow things down by quite a bit so it did not seem to be automatically "better" to do it in all cases.


For the most part, I think that filleting problems come from corner configurations that are not handled, not continuing a fillet across the seam of a closed surface (which seems like it may have worked in some cases in the past but is not working 100% anymore) and difficulties with getting a clean intersection between 2 extended fillet surfaces in near tangent conditions.

Of these, I'm hoping that the "crossing the seam" case may be the best potential to try and focus on getting fixed later. That seems to be the most common problem, so if it can get fixed I think it would be a pretty major gain.


- Michael

Ok, so my goal is to relace these points here in green, with the smallest amount of change to the curve in red. (arrows that is)



It was my understanding that you would need to place a point on either side of the existing point, as close as possible, to keep the curve from changing. So I zoom in, and can place my points very close.



But I cant place them "Exactly even"...For either side of the point needing deletion, and for the opposing curve that I want to do the same to. I would want the 2 curves to be the same, and not slightly different.



It's a darn "Conic Curve" again! :O

Hi Burr,

> It was my understanding that you would need to place a
> point on either side of the existing point, as close as possible,
> to keep the curve from changing.

Not necessarily "as close as possible", just fairly nearby it.


> But I cant place them "Exactly even"..

That's ok - it probably doesn't matter that much if they are exactly even or not.


> I would want the 2 curves to be the same, and not slightly different.

Well, if you're going to delete a control point, the result is not going to be 100% completely identical. But if you placed additional knots just fairly close to either side of the place you want to delete, the final curve you get with the deleted control point is going to be very very close to the original one.

When you say "not slightly different", do you mean you do not wish to allow even a 0.0000000000001 difference between the curves or something like that?

If you are really concerned about it for some special reason, place 2 or 3 knots to each side of the one you are going to delete, that should reduce the change to the curve even further when you delete the point.


Probably the easiest thing is to use the new Rebuild command that will be coming in the next beta - it will allow you to specify a tolerance level for how accurate the rebuilt curve hugs the original input curves, and it refits any segments that are tangent to one another as a new single segment automatically.

I've attached the results here of running Rebuild with a tolerance of 0.01 .


At some point I also want to probably update the merge command to maybe make it work to merge tangent segments into larger segments without rebuilding, but it is not yet clear to me if merging lines and arcs together into single curves is a good idea or not.

- Michael

Thanks Michael,
I didnt mean the .00000000001 thing. I just experienced that if the curves were different at all, that the swept surface had wiggles. Since I was sweeping it to a point there. I ended up using RebuildCurve to get my result, but I added about 80 points to be sure that that tangent area where the conic meets the line segment was true. I thought there was something I didnt know about how to delete that one point properly to get exacting results.

I've been using rebuildcurve alot lately and the new command sounds very cool and powerful. I cant wait to see it! (50 bucks to send it to me now) hehehe!

I didnt think of just putting in more knots. Good advice. Thanks for the time, more in my toolkit,
Burr

Hi Burr, yeah every time you add in an additional knot in that area, it will reduce the stretch of the curve that is affected by the control points in that part of the curve.

The new Rebuild command has 2 modes to it, a "Refit to tolerance" mode where you give it a distance value, and it refines the rebuilt curve to use as many points as necessary to achieve that tolerance, or a "# Points" mode where you just tell it how many points you want it to use, that mode is like the current RebuidCurve plugin.

- Michael

I looked at that file. Noticed it used the "more points" through the curve area, then dropped off at the line segment (which the rebuild doesnt do) though having both will be nice, to be able to fit an exact number of points when desired. Very nice.

I did notice, with the 3 curves you sent back....I did a coons patch with network and noticed that the picking can control the direction! Can you elaborate on this a bit to clarify it for me? I could figure it out with those three curves and place it where I want, but didnt grasp what it was doing, in case there were some different types of curves involved. :O

Thanks.

Hi Burr, well a Coons patch is fundamentally a 4-sided thing.

When you do it with 3 curves there is an implicit fourth curve involved which is taken as a curve with all of its points collapsed down on to the same spot.

The location of that singularity point will have an effect on the shaping of the generated surface. Just turn on control points of the generated surface and you'll see how the points collapse down to a single point on one side, that's the difference.

It seems like the case you have here, your path curve is actually an exact 90 degree arc, with radius = 10.15, is that correct?

In that case it may better for you to use Revolve, just selecting one of those pieces as the revolve profile. That's usually the most efficient and accurate way to construct something that pivots on a center point and sweeps out a path of an arc.

- Michael

I was just poking for the knowledge of how it worked. I can control the collapse point by "order of picking". 2 are not good, and one produces nearly the same result as sweeping does. I was just inquiring for my better understanding of what was happening. But it is not nessasarily a method I would attempt to construct something.

No need to go any further with it. i thought it was a mechanism I couldnt find in the reference. The reference just said, using network with 3 or 4 curves is a coons patch.

Hi Burr, the singularity goes to be opposite of the first curve that was picked.

So for instance in your arrangement you probably would want the singularity/pole point to be at the very top, that would be most similar to a surface of revolution and make the 2 similar curves behave more symmetrical to one another:



So to do that, pick this curve first:



Now if you turn on surface control points, you will see that they kind of start at that first pick curve and collapse down to a single point opposite of it like this:



If you instead pick one like this:



Then if you turn on control points you can see that it will collapse down towards the opposite of it, but that gives a kind of asymmetry:




But really revolve or rail revolve is much better for something that pivots like this.

- Michael

That was it. Thanks. (first curve picked is swept through.)

Does this hold true for a 4 curve selection? So first curve slected is swept through 2 rails towards the opposite 4th curve in the selction? (So the only real determining factore of picking the curves would be the first one)

Hi Burr, no it doesn't really work like that for 4 curves - for a 4-curve Coons patch the curves are touching each other end to end forming a ring.

The ones opposite one another are not determined in that case by selection order, only by their relative placement in the ring which is a geometric arrangement, not a selection based one.

So like if you numbered the curves going around the ring like 1, 2, 3, 4 like this:



It pairs #1 and #3 opposite one another, and #2 and #4 as the other opposite pair.

In that case it is all determined by how the curves touch each other.

- Michael

Also in a Coons patch there is not exactly a "sweep" happening.

It's more like a Loft between the opposite pairs of curves, then those 2 lofts get combined together into a result surface.

There isn't really a sweep happening, it's more like a membrane attaching itself to all the curves.

There isn't one set that is the "profiles" and one set that is the "rails" for this particular kind of surface generation, they are all more being combined equally in a sense.

- Michael

Hi Anthony,

> It would be great for me and for other users
> to see precisely the workflow you've used to
> build the "boiteaproblem" :D
>I mean if you've got a bit of free time
> (I know it can be time consuming huhu)
> if you could take some screenshoot along the
> way it would be REALLY nice :D

I haven't forgotten, here is the run down for the body of the watch, I actually did this with a slightly different strategy than the previous model I did.

1. Line up the reference curves, I used the magenta line in the center to do this and place the center of the watch face on the origin,
it will be easier to work with.
Draw a circle around the outside (green) and use the second magenta line as reference to create the red profile curve, it's a simple line and arc.
Then transfer this red curve to the (top view) reference curves and orient the right way.


2. Revolve the red curve about the center to create a solid. (Construct > Revolve. with Cap ends)


3. Boolean difference so we just work with a quarter of the watch as it is symmetrical around the X and Y axis.
If you like there is a alternate Revolve script which allows you to revolve any angle segment, in this case 90°.


4. Draw the profile curves as shown with sharp corners, using the reference curves as a guide, join (Edit>Join) these curves together.
*Tip: if you are on V2 you can lock the reference curves so they don't get in the way or disturbed and still be used to snap to.


5. Draw an arc, (Draw Curve>Arc>3pts) as shown using the reference curves to snap to end points.
Make sure the curve is at the level shown in the second picture.


6. Draw a line (green) tangent to the inner curve of the reference as shown.
Again, make sure it is on the bottom level of the solid.


7. Extrude (Construct>Extrude) this line in Z+ (up) using the reference curves for the height.


8. Extrude the arc, but this time using the 'Set dir' button in the Extrude command and drag a construction line from
the middle of the bottom arc to the middle of the top arc of the watch to set the direction.
This surface should end up on a slight angle.


9. Select these two edges.


10. Create a blend (Construct>Blend) between the two surfaces and then join them (Edit>Join)


11. Now use the top curves, drawn previously, and the newly created surfaces to do a Boolean difference as show.
Hide or delete the pieces not needed.


12. Create a fillet on this edge first, I used 0.5


13. Select all these edges and create a 0.25 fillet.


14. Draw a line aligned with the top edge of the radius as shown.


15. Boolean difference the solid with this line and hide or delete the top part and line.


16. Mirror the quadrant through X and Y axis, then Boolean union the pieces together.
Done!


Cheers

Thanks so much Danny, for taking the time to explain your workflow - I'll definitely be saving this in my MoI tips and tricks document since once again I've gleaned insight on NURBS modeling techniques.
(Shame I've been ruined by poly modeling for so long...)

-Will

Superb and tricky tutorial!
Cool use of colors
Bravo!

PS You can find this tutorial in the French Section ;)