BallCutter

I tested Anders's new BallCutter class from opencamlib. I did a simple sphere. The picture on the right is where I have moved the sphere down by the tool radius, to show that the toolpath is a sphere.

ZigZag with opencamlib

I made the ZigZag operation in HeeksCNC work a bit better with the opencamlib option. This machining library has lots of potential.

Grain the other way

This time I rotated the wood 90 degrees and did the same test again (roughing and finishing for both "conventional" and "climb" ).
This time: CV ( conventional ): x width = 19.93mm y width = 19.99mm
CB ( climb ): x width = 20.00mm y width = 20.00mm

So, the grain direction makes a difference. I feel ready for machining the Delrin that I am waiting for.

I have attached some videos to show the difference bewteen roughing and finishing. I guess the noise is indicative of inaccuracy.

Roughing and Finishing

No picture, because it would look just like the last one.
This time I did the same test, but duplicated the operations.
I set the original operations to have "extra offset" = 0.2.
I measured the parts at 5mm down from the top.
CV ( conventional ):
x width = 19.99mm
y width = 19.94mm
CB ( climb ):
x width = 20.01mm
y width = 19.98mm

Conclusion: Doing roughing pass, then finishing makes for much better accuracy.
Also: My machine seems to cut smaller in Y, than in X. I wonder if the direction of the grain has any effect? ( I can think of a way to test this ).

Bigger tags

I did the same test of "conventional" vs "climb", but with bigger tags. These were programmed as "tag width" = 10mm, so the tags should have been 5mm deep. I measure one, it was 4.97mm deep.
The following measured at about 5mm below the top:
CV
x width = 19.82mm
y width = 19.83mm
CB
x width = 20.09mm
y width = 19.97mm

conventional vs climb

I made two profile operations with tags. I notice these tags don't hold the part in very firmly, partly because I made the program assuming the wood is 20mm thick, but it is 19.6mm thick. The tags were set to be only 2.5mm thick, but they turned out to be 2.1mm thick. I could feel a significant amount of wobble in the parts before I broke them out of the block.
I have written CV and CB on them to remind me which was done with conventional milling and which was done with climb milling; for CV the tool went anticlockwise around the square, for CB the tool went clockwise. I used horizontal feedrate of 400 and vertical feedrate of 200.
Measurement taken at a depth of about 5mm below the top.
CV measured:
x width = 19.77mm
y width = 19.82mm
CB measured:
x width = 20.11mm
y width = 20.01mm

This demonstrates how the tool get pulled towards the part with "conventional" but gets pushed away from the part with "climb".

opencamlib

I have been trying opencamlib ( http://code.google.com/p/opencamlib/ ). It seems to work OK with a couple of tests. It took 15 seconds to make the NC file for the 10,000 points in the picture. The model made 4084 triangles. It took 25 seconds more to backplot the NC file to see the toolpath in HeeksCNC. I need to change my script to have rapid moves to the clearance plane, before this is useful.