Hi Martin, I tested the existing setback mechanism and currently it just does not generate good enough results to enable it.

Right now it generates a corner that's structured like this, it has pretty noticeable dimples in it:



When these kinds of corners are instead constructed with an n-sided patch then it might be feasible to enable setback corners at that time too.

- Michael

Hi Mike, it's not necessarily a bad idea to try and make things more regularized for the n-sided patch to work better (it definitely gets easier on it the less extreme amounts of variation it has to deal with), but I don't think that a combination method like you're talking there will be able to work very well. The problem is the surface shaping in the star shape is not good in these areas too (highlighted in yellow here):



Basically the star method often only has a nice shape right at the immediate vicinity of the outside boundary edges only...

Here's a bit more complex example so you can see how just sticking another patch in the middle won't really be an overall solution:




I guess actually thinking a bit about it now after being away from the research for a bit, a big part of how these ridges get formed is that the star blend process starts by putting in some initial curves that connect the outer boundary area to the center point, and then those become the edges of the surfaces that are generated. If those initial central-connecting curves are just a little bit too low or high it will then make for a wave in the surface shape as things both interpolate that curve and also maintain smoothness to their neighboring patch.

The problem is that it's pretty difficult to know how those connecting curves should be shaped from the start, it could be theoretically possible that some kind of iterative method that tunes those central-connecting-curves a bit after each result to reduce wiggles could help improve things. That would involve quite a lot of testing and research work though. As near as I can tell the CAD industry in general has given up on this kind of star blending approach though in favor of the n-sided patch method.

- Michael

Michael.

Thank you very much for experimenting with the setback feature, and what a shame that it's not quite up to the mark as yet. Hopefully, a bit more development by the Solids++ people should sort that out and let you get to work on the n-sided patch for this kind of stuff.


Thanks again for checking this out.


Martin (2).

Just for interest, here's a picture of various Radius/Setback settings done in ViaCAD.

Hi Martin,

> Just for interest, here's a picture of various Radius/Setback settings done in ViaCAD.

Do you still have the file with these in it? If so can you post a STEP file with them so I can take a closer look at the surface quality?

- Michael

Hi Michael.

I didn't save the file, but I'll do another for you soon.


Martin (2).

Hi Michael.

Here's a similar file to the one I posted a screen shot of earlier.

I've zipped STEP, IGES and SAT versions of the file, along with the native ViaCAD file for comparison (version 5.5, so it should load Ok even if yours is fairly old), as I got some issues with missing surfaces etc. when I brought them into MoI. I thought it might assist in case it turned out to be a MoI import problem.

I hope this helps you in some way.


Martin (2).

Thanks Martin - I think the blend surfaces may be missing in SAT format because they're probably specified as a "logical blend" where instead of storing the actual NURBS surface what is stored is just a kind of tag saying "make a blend between these edges" and then the receiving application needs to calculate the actual blend. You may be able to set some option at export time to stop that from happening and having it write out NURBS surface data for everything, possibly by specifying an older SAT file version (maybe ACIS v8 or v7 or something along those lines).

The main thing that I wanted to check was surface quality, and there are several problems with the blend surfaces generated by ViaCAD, with some ripples and bumps and lack of tangent continuity... It's difficult to see these problems in ViaCAD's display since it tends to use a pretty rough triangulation which tends to make it hard to really analyze the surface quality very well.

You can see some of these problems in these areas here (all following screenshots in MoI):




It's possible to get a really really clear image of every nuance of a surface in MoI by exporting to a polygon mesh format like OBJ format, and then cranking the mesh density up super high so that you can be sure that there are plenty of polygons being used in every nook and cranny of the surface, that will then show you the surface in super high definition. So for example I cranked the mesh export up to this level of density:



All those little tiny black lines are the outlines of little teeny-tiny polygons, this is how you can be sure that you're seeing a really good image of the surface without polygonal artifacts coming into play and obscuring things. With every polygon about the size of one pixel it basically eliminates any polygon shading blending artifacts. Then when turning off the edge display and and only showing the shaded result you can see this is the super clear view of that same surface:




There you can pretty easily see the problems near the boundary edges - there's a bunch of bumps and the patch surface does not meet up with the adjacent surfaces totally tangent, there's a sharp area where they meet, particular at the top and bottom juncture areas.

There are similar problems with every single one of your other examples, here are some screenshots:












I've also tried some various test cases in ViaCAD myself and seen these same kinds of problems too, so I thought I'd ask you for your file to see if your own examples were of good quality or had these problems.

This is kind of troubling in general because I've sometimes thought that maybe I should license the ACIS geometry kernel to improve filleting in MoI, but if this is the best quality that the n-sided patch mechanism in ACIS can deliver I'm not sure that it will really be that great, although I guess an argument can be made that it's better to have this than nothing... and also some kinds of usage purposes like if you're making an STL/3D printed result the size of the lumps are small enough that they're probably less than the size of the printer resolution itself and so probably don't matter for that case. It's not nice for high quality rendering though.

It kind of turns out that MoI's high quality surface display is almost something of a curse in this situation - just the regular display without doing the high density export generates enough triangles that you can see these surface defects during regular modeling, while other CAD programs have such a lower density triangulation that these defects get obscured and many people are just unaware that there's a problem until maybe they try to export something to a renderer and make it reflective and see that it looks bad in a close-up in the renderer...

I guess that possibly a good n-sided patch mechanism is rarer than what I had previously thought. Or maybe just all these cases are just too difficult on it to handle very well, although even just doing a setback corner on a box has problems. The file valise.3dm posted at the very first message of this whole thread has a whole bunch of lumps and bumps all over it...

- Michael

Here's a patch srface from rhino so you can turn on control points to see what it's doing. I still got a little "dip" in my tangent connections, but I think it was more of a poor choice for UV count to produce a good surface. I later tried 9x9, which was using the compound of 3 to get more equal. It "looked" like a better result. I would certainly have to study up on it to use it well.

Hi Burr, yeah that's not all that bad, I think it may be G1 but even with your improved one there still seems to be a bit of a wiggly dip right near the boundary edges:




The other thing is that it doesn't work all that well to have a zillion parameters to adjust for these when they're happening as a part of filleting, because there may be a variety of different shaped corner junctures being used all in one single fillet operation and it would be pretty awkward to try and integrate some kind of corner patch parameter adjustment to every single fillet corner...

Would you guys let me make the MoI display look cruddier so that you won't be able to see these kinds of problems? ;)

- Michael

""""""""I think it may be G1 but even with your improved one there still seems to be a bit of a wiggly dip right near the boundary edges:"""""""""

Yeah, that seams to be from that "flip up" in the upper corners of the untrimmed surface created, and I wasnt quite sure which of the 8 something parameters were to control that! That surface was using a 30x30 UV setting, that I think allowed it to flex there. I didnt save the 9x9 to re-check it in MoI.

I guess thats why we wait so long for n-sided patch. It will be it's own complex area adding that dreaded extra UI.

Please dont degrade the display to fool me. I get that in a couple other apps already :o

Hi all.

I haven't got much time to add to this today, other than to include a screen shot of the ViaCAD file I posted in the zip, and which has had part of the STEP file imported and superimposed on the right of the image.

The point being that there's a big difference in surface quality between the two, so I think ViaCAD must be messing up in some way when it exports to other file formats.

I'm not saying the original ViaCAD effort it perfect, just that it compounds whatever faults are there when it does an export.


Martin (2).

Hi Martin, it could be possible that it's kept as a kind of "analytic blend" in your first example that looks better, rather than it calculating a full NURBS surface which it needs to do for exporting to other applications. The problem is that it is also likely that it will calculate a NURBS surface result in various other situations as well like when there's an offset or intersection that needs to be performed on the analytic surface. So if there is some kind of degradation happening at export, it's pretty likely that the same kind of degradation will also happen in a variety of modeling operations as well.

In fact if it is not quite as bad in the initial display in ViaCAD that's quite a big can of worms in itself because it means you're not at all operating with a "What you see is what you get" system - that's quite troublesome if things look ok at runtime but then fall apart with major changes in surface structure and quality when you go to do some final step like export...

But also another problem though is that it's hard to see anything for sure in those screenshots that you posted due to the very poor quality of the display - notice how there is an abundance of jaggedy triangulation artifacts everywhere that you look in every screenshot - the triangulation being used to display the surface is just too coarse to be able to really see what's happening with the small details of the surface. This often times can completely disguise surface quality issues with ripples and bumps.

Think of it like this - say you have a surface that has a bunch of little squiggles in it, when viewed from the side it might be shaped like this:



But say this surface gets triangulated for display coarsely with only a few triangles say like these 5 points here:




When this happens the shaded display (which is just done purely from the triangulation) can easily look like the surface is all nice and regular and smooth, but what really happened is the triangulation is too coarse for you to actually see all the true details of the surface.


Now compare those screenshots that you posted where you can see all kinds of triangle artifacts, to this one here that I took in MoI:



Notice how in this screenshot you cannot see any trace evidence of triangulation whatsoever?

If you can't get the ViaCAD display to this same level of quality removing all triangulation artifacts, and getting to almost one triangle per pixel (or close to that) you won't really be able to say for sure if the original surface before export was actually less wiggly in some way or not...

- Michael

So does exist a perfect N sided patch mecanism somewhere?

If yes, just copy past only this perfect 3dm "corner" inside Moi and resize it to the model ?
with oriente / line/line
(copy the half dozen of possibilities for have a sort of little familly of corners for adaptation to the corners wanted)

That maybe can avoid some times of difficult research?

Hi Pilou, it could be possible that a Parasolid based modeler like SolidWorks or NX might do a better job.

But even on a box the shape of the patch is a little different depending on the particular radius and setback values used, it's not like there is only one single corner shape that fits every kind of case...

The type of solution that you're talking about with just reusing one specific surface would only work on one shape and one set of parameter values.

- Michael

Hi Martin, I played around some with your original ViaCAD file inside of ViaCAD and yup definitely it looks a lot better there, and I think it is probably due to it being handled as a special analytic blend surface at runtime in ViaCAD and then having an actual separate NURBS surface result generated for exports.

You can also see a NURBS surface generated if you turn on surface control points for that one surface in ViaCAD and move just one single control point (on some far off corner of the patch so as not to actually change much with the edit itself) - that converts the surface into a regular NURBS surface directly in ViaCAD and if you then paste back the original blend you can see that the control-point edited surface has those ripples in it and is slightly differnet in shape than the analytic blend.

The analytic blend seems to be handled by a lot of mechanisms though, intersections and offsets do not seem to degrade it so that means that they implement the special blend surface type deeply into the ACIS system and it's able to do a lot of processing on the kind of "ideal blend surface" and not need to use the NURBS surface version for too many places except of course for doing a file transfer out of the ACIS system to some other system that wants to deal just with plain NURBS surfaces, at which point the messier NURBS surface approximation of the blend gets used instead.

This does not bode all that well for trying to integrate ACIS into MoI just for some particular operations like fillets though, because in order to get the nice n-sided blend behavior the surfaces have to stay as those special ACIS blend surface types and not get converted into regular NURBS. So that means keeping all things as ACIS data all the time and not just using bits and pieces of ACIS to calculate NURBS surface results.

So anyway I think that's what's going on there, that there's a deep treatment of these n-sided blends as a special surface type in ACIS which seems to work nicely if you stay all inside the ACIS system for all the operations involved but the NURBS approximation result is not all that great...

- Michael

I found out how to do a super dense mesh in ViaCAD - you can do it by right-clicking on a surface and picking "Change object type" to a mesh type, that has some controls including edge length control which can be used to make a really dense mesh object. I did that with the original blend surface and it definitely does not show the same kinds of problems as what you get from export/import.

So I'm reasonably sure that it is indeed different than the exported one and that it's probably treated as a special kind of surface in ACIS and when it gets that special treatment it's able to look nicer than it does when converted to NURBS.

Anyway, it's pretty good for me to know about this particular inter-op problem with ACIS,

- Michael

And in ViaCAD I am seeing some operations that end up using the NURBS surface, offsetting does seem to do it - if I take the nice looking N-sided patch by itself and thicken it into a solid with a thickness of 10 units, those same bumpy areas appear in the offset surface (here converted into a dense mesh):



So it seems that sometimes you get things calculated off of the ideal logical blend surface, and other times stuff happens on the approximated NURBS surface that's along with it.

- Michael

I know this isn't what you were looking for but this kinda gets you there:
(i patched a bunch of surfaces together and joined, although I realize this process is generally frowned upon)

Hi Michael.

Another splendid explanation as to what is going on here - thank you.

So it looks like ACIS can make nice surfaces, but is apt to throw out a lot of the good work further down the line. I think that may explain why I've noticed changes to (complex-ish) faces of solids when I've done something like join them within ViaCAD.

My question is whether this hugely degrading 'feature' is entirely down to ACIS, or is likely to be some unfortunate error in how ACIS is being implemented? A case of "Snatching defeat from the jaws of victory", if ever I saw it!

http://doc.spatial.com/qref/ACIS/html/modules.html for anyone with an interest and can comprehend it :)



Cheers.
Martin (2).

Hi Martin,

> My question is whether this hugely degrading 'feature' is entirely down to ACIS, or is likely to be some
> unfortunate error in how ACIS is being implemented?

Well, the "error" part is just that the actual ACIS "fit NURBS surface through n-sided boundary" mechanism seems to have a bunch of the problems that are just pretty typical to that class of operation which is a tendency to have bumps and wiggles in the generated result. Wiggles are pretty much a natural byproduct of constraint based fitting mechanisms.

But they are able to mask that problem in many cases because they do have a nicely performing "analytic n-sided patch" surface which is able to be used instead of the NURBS surface for quite a few operations. Any operation within ACIS that is able to use that analytic surface instead of the NURBS surface will behave really nicely. Any operation that actually needs to operate only on a NURBS surface rather than a more abstract analytic surface (export being a prime example) will then have the less nice fitted result come into play.

Whether this is good or not I suppose is open to some interpretation and dependent on what in particular is being done - I guess if you only end up doing the particular things where the analytic surface is able to be used then the lesser quality one is totally masked and never comes into play and you could say that is a good result for that particular case. If you do end up doing something that needs the NURBS result then this sudden quality shift is probably pretty surprising though. And it's bad news for NURBS data exchange certainly.

- Michael