Sweep v. NetworkSrf


This is another general question that has kept me up at nights for years. In the attached file I have a set of frames that illustrates something I have seen for a long time.
If I do a NetworkSrf on them I get:

If I do a Sweep2 using the same frames I get accordion pleats
Problem Sweep 2.3dm (95.2 KB)
I am curious why the two methods produce such widely different results (especially considering those who say never ot use NetworkSrf).


You have posted an example that shows 2 really bad wtechniques to make NURBS surfaces.

It is not a good idea to make a surface where 2 edges are tangent to each other. When you do that you are inviting trouble. It almost guarantees that future operations like shelling or offsetsrf will fail.

It is also not a good to define surfaces by more curves than the minimum needed to describe the shape. One consequence of using way more curves than you need is waviness in the resulting surface. NetworkSrf doesn’t try to make the surface go exactly through all the curves so it doesn’t have as much waviness as a result of too many input curves too close together.


From my understanding of Sweep2, it uses a cross section and scales it in either 1D or 3D. Since your “ribs” aren’t mathematically similar shapes I think the accordion effect you are seeing is the tween between the two cross sections, creating the minimal surface.

If you sweep just one section along your two rails you can see that they all produce different shapes, hence it has to constantly readjust the surface to match the cross-sections - like soap film for example.


The RIBS here were generated by intersecting lines with a networksrf. That is why they have so many control points. I left those in case someone wanted to play.

Thus I can build the same networksrf using just the edges and cyan line.

I can get the same accordion even when I cut the number of frames down (again I just left them all there to play with).


Then what would be an approach to making such a surface? In this case the outer edges are all fixed by plans. The cyan line is a profile off a plan. Ending arch matches up to the next surface. The side curves are a semicircle projected on a cylinder with lines.

The interiors are all undefined by the plans. Thus the exteriors are fixed and the interiors are variable. How would one create such a surface, ideally?


That’s hard to say without seeing the connecting geometry and knowing what further downstream operations might be. The surface looks something like a half pipe that gets cut off where intersects with other geometry. If that’s the case I would model it as such.


This is what it is sitting on. Its not a half pipe, however, the sides below the curve flare out and form a planar surface.

It sits on a cylinder as shown.

(Pascal Golay) #9

Hi Jim - try: sweep the arc-shaped edge curve along the center curve and then trim the result with the pipe…?



That sweep is the best I have been able to do without a NetworkSrf. The only issue is that the edge of the sweep does not remain smooth with the sweep. I might have to live with that.


I would do as Pascal suggested but I would split things up at the kinks in your curves so that you planar parts are separate surfaces from the curved parts and you don’t end up with fully multiple knots in the surfaces.

Sweep X.3dm (112.6 KB)


Have you tried keep same height option in Sweep2?
I think it’s a check box withing the command.