Typical junction of three fillets

[systembolaget]

A very typical junction - three fillets of different size. One can see in the images that the large transition surface has far too many isoparametric curves and that in some places the zebra-stripes show some wavyness. Can't attach a .wire file here?!

What could one do to improve this classic junction? Is this the right approach in terms of patch layout?

Thanks in advance for some hints...

[Anonymous]

no solution for your actual problem, but it should be possible to attach data under the reply text box, see the attached picture 😃

[systembolaget]

Problem is - the forum module rejects attaching .wire files using the attachment feature, whether from W 10 or OS X : (

[systembolaget]

Nobody's got an idea regarding the question, also the "Curvature dev=1.0" issue is raising its ugly head again... Any help much appreciated!

[Nadja.Bueckmann]

Hi @systembolaget,

you can uplad the wire file as a zip archive. Then you can attach it.

I attached an image to the patch layout I would suggest. I think with this layout the result is much better.

Best, Nadja

[systembolaget]

Hi Nadja,

thanks for the reply. The thing is that the fillets can't change as you indicate (then it would be simpler). Also, there's that dreaded max curve dev 1.0 issue. The diagnostic shade looks good at lowest tolerance. I never got these max curve dev 1.0 issues. I have seen a single span solution for the very same problem some years ago somewhere - but I cannot remember where that was. I attached the .wire file in zipped format...

[Anonymous · Accepted solution]

Hallo systembolaget

I incidently looked at this and thougt.. give it a try..

your patch layout is good I think, I would choose it in a similar way; alhough I have two slightly different variants.

First. I guess your square will work a lot better if you just trim convert (temporarily) the fillets.

Still I guess its always a bit of manual optimization at the cv's neccesary (with transform cv tool; and most important the align-tool)

I usually dont rely on the inbuild aligns of square or rail in such a corner... 

1. For the first patch marked in the picture,

  Variant 1: I basically used an extension of the fillet surface and aligned it, with a blend factor.

  Variant 2: The small fillet runs a bit further on the big fillet in a real fillet shape ... I hope the picture/model explains.

2. A rough patch for the big inverse-ball-corner-patch (2). I just throw a fast surface in there (degree 5x5), without looking too much at the alignment. (I used a blend, could also be a square)

3. Than I align the two borders with the big fillets to g2 with the align-tool as good as possible. And I always use temporarily trim converts on the initial fillets, so I get the most accurate cv-layout. red 1 and 2 in the align-picture.

Than I use the Align-History in combination with the transform-cv-tool (slide or projected) to get the cv's in the right position. Place a locator on the border to the small fillet (align 3) to get that alignment as good as possible before actually aligning it (maybe temporarily align that to see where the cv's want to be, than align 1 and 2 again, and optimize)

4. And at the very end for the alignment 3 and 4, I increase the degree and/or the spans to get that last alignments perfect. (my customer accepts higher degree rather than more spans... so I increased the cv's in my example). Often if you are a bit patience with the optimizing, its not even neccesary to increase the cv's or spans.

That sound like ages but it is just a thing of minutes 🙂

The highlight is still a bit wavy, but most of that happens following the fillets. although it still could be a bit better...

Too Curv-deviation of 1

The curvature-deviation-value of 1 sometimes (but really just sometimes) could be a calculation mistake. But especially on flat surfaces.

I'm not perfectly sure how Alias calculates. But mathematically explained... sorry for that 🙂 ... curvature is the inversion of the local radius of a surface c=1/r . on a flat surface the radius is infinitely big. While your fillets / ball-corner may have in every point probably the tiniest amound of curvature, or an extremly big local radius. But while the deviation on the border is the difference between both ... "infinite" - "extremly big" = "still infinite"

hope that helps

Greetings David

[systembolaget]

Hej David,

very much appreciated. It is extremely rare to have such a professional discussion here, compared to Rhino or SolidWorks forums that are brimming with knowledgeable people like you. The "betterment" of my initial patch layout is preferable, because it is more "constructed/obvious"; but the other solution you came up with is intriguing, I must say!

I noticed that both of your solutions also suffer from the dreaded "max curvature dev=1.0" issue, but they look just fine, no matter what visual analysis tools one uses. Thus, to me, it looks that this could be a bug or calculation limit of Alias; but then what do I know?

Without your hint, I would never have thought of trim converting the trimmed fillets and extend that little bugger forward and then align it. I have much to learn still, so it seems.

Keep up the good work!

[Anonymous]

Just one last thing for the "max curvature dev=1.0" issue you talked about:

You could have a look into one of the video tutorials from Autodesk Class-A which explains it quite good. This would be Video 3.13 from minute 14:05 on this page: http://autodeskautomotivetraining.com/product/alias-class-a/

Helped me a little to understand the issue and takes just about 2 minutes to get summary of it.

Cheers

[systembolaget]

Ok, thanks a lot for pointing out that video! After Barry Kimbal says "Well, yeah, this is one of the areas where the mathematics ain't workin' ... so close, I would leave that" - I feel much better already ; )

[tomas_lindehell]

Hej Systembolaget

Interesting corner. Here is also one possibility to solve it with 5 by 5 degree surfs.

I started out a little bit like DavidPBP but tried to achieve full Curvature Continuity with Single span surfs.

1. Freeform Blend Surfs as a start

2. The Align tool

3. A LOT of Move CV work! 🙂

Here is a Youtube playlist with some Alias Corner examples:

https://www.youtube.com/playlist?list=PLBN7j0ISn6_uX36BHzBdebrP5TlUp3uAq

Also a Zip file with a this possibly solution

Cheers

Tomas Lindehell

[systembolaget]

Ok, thanks, much appreciated!

A good exchange on here... Saw some of those videos before. I am trying to build a little "typical problems" repository for industrial designers, where Class A is not the initial goal, but rather getting the shape right. But trying to use Bezier surfaces is of course always a good idea for many reasons.

After trying both solutions for some hours, I found that with my original curves -> fillets -> trim converted I cannot achieve a 5511 surface with G2 all around, even when aligning to the two larger fillets first and not using the square's or birail's built-in continuity settings. I find a lot of "CV massage" is needed to close the gap to the large flat trimmed surface on the front. Plus the max curvature dev=1.0 thing is back. But as Barry Kimball tells us in his videos, that's just the software being drunk or on weed ; )