managed to confuse offs: http://jmkasunich.com/pics/offs-3.png
oh, I think I know what I did - I offset so much that the pocket disappeared - so offs turned it inside out (sort of)
you may not have noticed this, but offs doesn't quite get everything right yet :-)
oh, I've noticed
I'm fixing up a couple things now
I ruined a part yesterday with a wrong arc (no preview, hand edited gcode on the BOSS)
got it right on the second try though
our preview is so nice.
no big deal. learning experience!
I'd be basically hopeless without it
(the lesson is don't hand edit gcode and then not test run it)
o-word programming loops and such would be insane
I actually have a hacked up emc that accepts (some of) the BOSS gcodes so I can get a preview
woo-hoo - got the last of the "fur" out of the program
but I didn't bother - it was just a simple change!
little loops and excursions from what should be a square corner
offs roundoff errors or something
ah I see
some are only a few thou
making a better finishing pass?
I'm gonna run the last manifold with 0.010 stepdowns instead of 0.050, to try to get rid of the tool lines
I know its gonna take a long time, but its only one part
when I re-ran offs with the finer steps I got some fur
that will be neat.
2200 lines of g-code now (was just over 1000)
look out, that's probably over 100 feet, might not fit
(that's what BOSS says)
I think it should be safe to up the feedrate a bit for these very light passes
from 12 to 15 ipm should help a little
was that a 1/4" mill?
ah, that's small
roughing is 1/4", that is at 12 ipm, and zero to 0.100" width, 0.075 depth per pass
roughing leave 0.025, pre-finish takes that to 0.015 (and gets into the corners that the 1/4" mill couldn't), with 0.060 stepdown
finish takes off the last 0.015 and uses the real fine stepdown
and here we go.....
do you ever leave it and let it cut, or does it not have your trust that much yet?
for light cuts I leave it
during roughing I squirt it with wd-40
I don't leave any farther than this keyboard, 8 feet away
unfortunately I'm here now because I got a following error
last pass of roughing (I roughed a little deeper this time, that last loop must have been too small somehow
but ferror from just a funny path?
maybe quick successive invocations of backlash - two reversals very close together?
seems like backlash + screwcomp + steppers is not well tested
I just manually edited out the last roughing pass (it was cutting air anyway), gonna start again
we know comp can generate 150% accel
rough and pre-finish done, its started on the half-million finish orbits
there are definitely places in the path where blending is failing - indicated velocity drops, and I can see it stop (or at least almost stop)
probably places where offs generated paths with very short segments
that'll sure do it
I have G64P0.010
but that doesn't help if they are arcs
the tolerance affects non-tangent intersections, right? (assuming the segs aren't too short)
not quite half way down, and looking nice
I don't understand the question
G64P0.010 means you can round non-tangent intersections, right?
(tangent ones can be exactly followed)
yes with up to .010 diversion from the programmed endpoint
also, if a series of straight moves stay within .010 of a line, it throws out the intermediate points in favor of the longer line
so why did you say "that doesn't help if they are arcs"?
I don't think my g-code contains any such series of lines
you aren't seeing much or any of this secondary benefit
it's the 'naive cam detector' code
its only a benefit if you have crappy cam, which I don't have
yep that's pretty much why we wrote it
I'm pretty sure the hesitation I'm seeing is at an arc-arc transition
one of the arcs is short?
the original path has arc-arc-arc, both transitions tangent - the offset path is no longer tangent at the intersections
the middle arc gets shorter each time I move down (and increase the offset)
the exact tangency doesn't matter much at all
at some point I bet the middle arc disappears
hope it actually disappears, and doesn't change into an infinitely small arc
I'd have to examing the g-code to be sure
maybe later - its almost done
+ (command ".pedit" (entlast) "e" "b" "n" "g" "x" "x")
I must say that is quite purty
near the end as the orbits got small it was doing one every few seconds
[02:30:32] <jmkasunich> http://jmkasunich.com/pics/port-fine-machined-2019.jpg
compare to http://jmkasunich.com/pics/port-machined-2014.jpg
you can see a line running up at about 2 o'clock - that is one of the places it was pausing
what's in the gcode there?
line to arc
is one overly short?
the arc runs all the way around to about 12 o'clock (I'm cutting CCW - climb milling)
the line is about 1/8" ling
about 0.5 seconds at 15 imp
I wonder if it still does it if you use G64 only
my accels are only 4 ips^2
15 ipm is 0.25 ips, so it only takes 0.0625 seconds to accel from stop
might be a bug/unfortunate feature then
its dog-walkin time here (trying to not stay up so late) - maybe tomorrow I'll extract a single pass around the loop from the program and see what it does in more detail
that would be great
goodnight, I should do that too.
actually, I can post one loop right now - its less than 20 lines
no hurry, I won't look at it tonight anyway
ok, tomorrow it is
[02:50:40] <jmkasunich> http://www.pastebin.ca/985550
I started to annotate it, and realize that there are all kinds of weird things in there
really tiny arcs, including some G2's where I was expecting only G3, etc
also two consecutive G1s with the same endpoint
won't mean a lot without the corresponding offs input path
that's a certain exact stop
I bet these are more like offs bugs than emc bugs
yeah, the doubled lines certainly
Guest220 is now known as skunkworks_
[14:44:49] <cradek> http://www.fricktion.net/~mfrick/lathe/els/ElsDoc.sxw
[Global Notice] Hi all! As you are aware we're preparing to move to a new Services package -- thanks to the hard work from the Atheme guys, in particular jilles (and nenolod) we are now ready for you all to come help us make sure they're approaching ready for production! Please join us over at the testnet. Connect to irc.freenode.net on port 9001. Thank you and have a great day!