Changed behaviour of GeomAPI_ExtremaCurveSurface in pythonocc 0.17

[jnwalther]

Hi everyone,
I am using pythonocc to compute the intersection points between a B-spline surface (Geom_BSplineSurface object) and a set of vectors (Geom_Line objects), using GeomAPI_ExtremaCurveSurface. This used to work like a charm in version 0.16.5, but after installing version 0.17.0 several intersections fail:

The number of failures seems to be somehow dependent on the length of the line evaluation interval [0,wmax]. Also, the surface's circumferential coordinate starts and ends at the top where the failures occur, so I don't know if this might have something to do with it. GeomAPI_IntCS manages to find the points, but is much slower than GeomAPI_ExtremaCurveSurface.

Have there been any changes in the function between versions which could explain this?

[tpaviot]

@kngbuzzo Thank you for this feedback. pythonocc-0.17 is based upon oce-0.17.2, which contains many changes since the previous oce-0.16 (FYI this is the step from official OCCT671 to OCCT680). OCCT680 release notes (https://www.opencascade.com/sites/default/files/documents/Release_Notes_6.8.0.pdf) does not mention any change to the Geom_ExtremaCurvedSurface function/algorithm, however there might have been many other changes to functions/classes used by the Geom_ExtremaCurvedSurface class. Your use case seems to be complex, would it be possible that you provide a simple example that reproduces the issue ?

[tpaviot]

@kngbuzzo Are you sure about the spelling of Geom_ExtremaCurvedSurface, I can't even fin such a class in pyocc-0.16 or 0.17

[jnwalther]

@tpaviot Thanks for your answer. Geom_ExtremaCurvedSurface was a typo and should say GeomAPI_ExtremaCurveSurface as in the title. Sorry about that. I have tried a second example, which is just a regular cylindrical barrel, and all the intersections work fine.

[tpaviot]

@kngbuzzo It will be very difficult to debug without an example that show the error. There might have some parameters to tune, but unfortunately I don't use to work with this class and can't be helpfull.

[jnwalther]

@tpaviot I'm sorry I don't have a good example to share. The algorithm seems to work great with all my simpler models... Btw. I went through the release notes and they mention new global optimization algorithms for extrema computation in their highlights. More specifically, the Extrema_GenExtCS class is said to have been changed, which, I suspect, is somehow being used by GeomAPI_ExtremaCurveSurface. In a way this makes sense, because the more I think about it I am certain that my problem is related to the initial guess for the optimization, which is what has been changed, from my understanding. So I guess that answers at least part of my original question.

[jf---]

@kngbuzzo , we appreciate your feedback, but lets have an example, or close this issue; its not in an actionable state...

without being able to reproduce the issue, this will take too much resources on our end I'm afraid...

[jnwalther]

@jf--- thanks for bringing this back up and sorry for the delay. I came up with an example to illustrate my problem based on an ellipsoid, which you can find here: https://github.com/kngbuzzo/pythonocc_examples. For 2500 intersections I get the following output:

pythonocc 0.16.5:

Time for intersections 0.480000019073
[]
0 failures

pythonocc 0.17:

Time for intersections 2.21000003815
[[  2.00000000e-01   3.00321910e-04   0.00000000e+00]
 [  1.15714286e+01   3.79488064e-01   4.89295676e-02]
    ...
 [  5.21836735e+01   4.00839802e-01  -5.16825693e-02]
 [  2.00000000e-01   3.00321910e-04  -7.35576530e-20]]
107 failures

The black dots indicate failures (obviously). Also note the different timings! I hope this is a little more useful to you.

[jf---]

that's perfect @kngbuzzo , thanks so much!
Can you share the script please, than I can delve deeper into this, whether this can be overcome in pythonocc or should be escalated to OpenCascade.

Perhaps you can share it as at gist.github.com, or just paste it here if you like...

Thanks a lot for this!

[jnwalther]

@jf--- , I added the script in a repository under my name, the link to which I have provided above (https://github.com/kngbuzzo/pythonocc_examples). Is that ok for you?

[jf---]

Shoot, I misread your earlier comment, sorry @kngbuzzo & thanks again!!

[jnwalther]

Cool, thanks for looking into it @jf--- !