Can Rhino model any object which polygonal packages like Maya are able to create?


In both cases, you can get any mesh. What then is their difference? Why is SubD worse in production?

(Tom) #82

here is a little shortcut on a high render mesh used in automotive:

you can see how the tessellation is based on a underlying surface patches and how its exactly represent the desired curvature. you cannot do the same without surface modelling or with subdivision. Subdivision would not be based on curvature, its just subdivides each mesh face more and more. 1 face becomes 4 faces, 4 faces becomes 16 faces … and so on. There is no logic saying you tessellate based on curvature.

(Nathan 'jesterKing' Letwory) #83

Aye, but it is up to the creator to create the necessary control faces to achieve what is done automatically when a render mesh is created from the nurbs surface. And while the mesh generation works, it doesn’t win prices for beauty, nor manageability.

If a designer takes the time quality of approaching similarity could be achieved. If it is a good way to do is a different issue.

(Tom) #84

If you can create everything you can do with Nurbs true. However I do not believe it gives you any benefit in doing it with sub-d. If you like to achieve same quality, I doubt sub-d will be faster. In addition to that, when it comes to detail modelling, Sub-D is limited, because Sub-D is based on averaging polygons in between. In T-spline its always G2 in between patches. But especially at situations shown in my images, this is not the best thing to do. On strong curved parts, an overall smooth curvature is more important than always having G2 transition. You don’t want wavy sections. This is the fundamental disadvantage of sub-d yet.
In addition, beauty and manageability is not important for nurbs tessellated meshes. Isn’t it accuracy and efficiency?

(Tom) #85

but before comparing apples with oranges…

if you do industrial design or engineering use surfaces, if you do cgi or concept design use polys.

(Nathan 'jesterKing' Letwory) #86

In full agreement :slight_smile:


Here I do not understand something, what is the problem of logic?

(Tom) #88

the logic of sub-d is called catmull-clark subdivision. Its a very “dumb” process.


Has been explained n times above; that’s a polygonal model and, with polygons, one can’t model a product with draft angles, component deviation control and curvature control. Most importantly, from polygons, which are inherently flat, one cannot derive G code for CNC milling of prototypes and, later, tools.


Well, you can, with the right CAM software. The results will be as faceted as the mesh is (at best).


Yeah, well, no.

Fist, any tool path, you extract from your well defined Surface may or may not be an exact section. You can expect a spline that fit’s the surface within a given tolerance.

Second, G-Code is basically just a list of 3D coordinates for the tool of your choice. The path can be interpolated in linear or circular fashion. That means, your precise spline is essentially broken down to polylines and (poly)arcs. Which finally is converted to tool steps.

The advantage of NURBS is that you can model with precise control over any possible aspect of your surface and then decide on the precision of the mesh and polygonal path to represent your model down the line.

If you want to have control over production surface quality, you can either define poly model down to a resolution that fits your maufacturing tools or you just need to accept, that between your design and your smooth production surface the curvature is constrained by the properties of a lot of catmull-clark iterations.


Well. NURBS more conveniently and more qualitatively than SubD in production means (in calculation of exact control, control of curvature, a corner). A grid at NURBS more logical (density of a grid depends on complexity of a surface). And NURBS doesn’t depend on mesh (mesh is formed atop).


But what is the use of that? A faceted aircraft seat, lacrosse stick or ski boot? No thanks!


Sorry, pal, but there is no “NURBS grid” or “polygon grid”. And, no, there is no “mesh formed atop”; you can generate/model a polygon mesh without a prior NURBS surface patch network present.


You’d probably never notice it it was well done… It’s all a matter of tessellation. Mesh facets smaller than you can detect.

You listen to digital music don’t you? In the process the sound has been divided up into tiny bites (tessellated) smaller than humans can detect and then reassembled. Same with digital images. If the pixels are small enough you can’t tell they’re there.

Add to that what @HaLo was referring to - even if you have an absolutely perfect “Class A” surface, that info is going to get first tessellated by the CAM software into tiny bits of straight lines (that’s all CNC machines understand). Most CAM software actually mesh the NURBS data behind the scenes.

Those tiny 3D line segments, each represented by a line of text in your machine code, are then going to get interpolated physically by the tool using a system of motors, linear and rotary axes, measuring scales, feedback loops and motor control circuitry. All of that ends up being converted into discrete physical movement commands to the cutting tool happening at micro/millisecond intervals.

The result is it looks “smooth”; inertia and the constraints of the physical world actually help the smoothing process if they’re well managed and don’t induce overshoots. More irregularities are however introduced at the workpiece itself by tool vibration and thermal effects. Then the molds often end up getting polished out by humans… Then actual parts get made from the molds. Lots of opportunities for deviation there.

So, like the digital music you hear or the images you see, even your carefully designed smooth, mathematically defined Class A NURBS surface data has actually been converted into discrete elements and then reproduced as a physical part using a fairly complex process - with possible error introduced at every step of the way. If the error is kept below the threshold of perceptibility, (or acceptability?) at all stages, no one cares.

If you go back to the beginning and start with a quality mesh object instead of NURBS and repeat the process above, you will not be able to tell the difference between the two parts - the error will be below the limit of perception/acceptability.



Not a single client I consulted for, and I’m talking about large international businesses, ever used polygonal models for prototyping and production of tooling. That’s why in engineering departments in Europe and Asia, and at toolmaking companies in China, Taiwan or Germany, you predominantly see SolidWorks, Creo and Catia.

Ask Lear Seating, Recaro or Airbus, Electrolux or Volvo Trucks about using polygon software in design engineering and production tooling. Omega or Rado watches with polygon software? Fugeddaboutit. The meeting will be rather short.


And how it is possible?

How? I see NURBS grid, I can see mesh.

(Nathan 'jesterKing' Letwory) #98

The question wasn’t which package is best, but AFAICT is it possible. With a lot of hard work, and the necessary polygon modelling tools, it is possible. Feasible? Not the issue.


All usefulness of Nurbs for very accurate production data aside – every technology which rules a market for decades forms strict conventions. And such may turn out quite painful at times.

I have worked for firms who perceived the strict Nurbs only pipeline in Tooling / CAM as a bad limitation. Footware makers who developed their lasts and the whole shoe design as high resolutions meshes and who wanted to send their files to Asian contractors. While one obviously can mill from meshes, none of their contractors accepted polygonal data.

In the end I helped them shoehorn their detailed meshes into ugly and detail-reduced Nurbs – with the sole reason to make their data compatible. Obviously the first thing the CAM program will do with these files is re-tesselate them…


I think the problem is not so much machining form a mesh if it’s dense enough but building the tool and extracting splitlines from a mesh.

If I exported a mesh from Rhino setting the max distance, edge to surface to something like 0.001mm, import into Fusion and setup a CAM op, enable smoothing so you get arc fitting you’d find it hard to tell it was from a mesh. Building a mould or die from the mesh would be more of a problem, don’t know how you’d get on with a solid modeler like Solidworks.