New Trajectory Planner - Testers/programs wanted

Better yet - could you isolate a short snippit of gcode that does it?

I am playing with the axis xyzabcuvw config and am not seeing it in the preview. Do you see it in the preview also?

thanks
sam

The the g-code programs posted above are these same carvings.

I do not see it in the preview, mostly because the W axis is not realy shown in the preview.

This machine configuration is a servo set up, and the problem might be magnified by a tuning issue. But I kind of doubt it since it doesn't show up with the old planner.

In the above g-code 130207LWZ.ngc the first errors start to be seen around line #13-15 and #17-19. That is the first tick carved on the clock face. The problem is visible on all of the ticks arround the clock. With the same mistake on the same corners of each one. This first tick begins the carving at the top left corner and continues clockwise around the shape. the 2nd and 4th corners carved are the ones where the mistake is seen.

Sraingly enough the very last thing carved in the file is the tick right next to the very first one. This one starts at the corner farthest left, then continues clockwise. So that the 2nd corner carved is the one that was carved 1st on the very first tick. On this last tick the errors are seen on the same corners as the first one, but are the corners that are carved 1st, 3rd (and last it finishes where it started) are where the error is seen.

is there a chance we could play with your configs?

sam

Maybe. What would you like me to mess with? Would you like me to post my config?

I might have a stepper machine I could play with some as well. It's Z and W axis are fixed at a close spacing, so we don't use it for double carving. We use it more like a 2 postion tool changer.

Hi Todd,
Thanks for running this test for us! It would help debugging if I could run your config in simulation. Looking at the the position log of both Z and W should reveal any issues.

Another question on a hunch: is the servo control done on dedicated hardware?

-Rob

Here is my config.

Ok -I think I am seeing it. I am using the sim axisabcxyzuvw config with the circular blends added to the ini file..

this simple program
g64
g0x0y0w1
g1w-.1f3000
x1
w1
m30

you can see in the first screen shot - the left corner isn't blended. (compared to the right) I assume this is quite a constraint violation but I have not added acc readout for the w axis yet. I assume your sytem is step/dir servos? (I could see that type of system just overshooting at the violation and recovering)

Here is the program run with z instead of w
g64
g0x0y0z1
g1z-.1f3000
x1
z1
m30

you can see it blends both the left and right corner.

sam

My config is analog torque mode servo using a Mesa 5i25/7i77 combo to interface with some Teknic servo drives. So LinuxCNC is closing both the position and velocity loops. Hence why my servo period is 2.35kHz, as fast as I could get this PC to run. Wish I could have gotten it about twice as fast. I would have felt a little better about tuning it. This is the last time I will buy drives that can not do velocity input. In hinde sight I should have set this up with step dir, and encoder feed back, for digital velocity command, but I realy wanted to try setting up an analog system. I did digital on another machine with a 5i25/7i85 setup after this and it worked very nicely.

I think I see where the problem is.
The blending turns off (or is suposed to right now) for movements that contain axis other than XYZ. In this snipit of code:

the error occurs at the transition of a line with a W comand to a line line that does not have a W command. My guess is that the new planner is blending this transition move for XYZ but letting W do ??, causing the small gouge at that point.

I think I've fixed this in the latest commit. It looks like I forgot to check both segments for ABC/UVW motion, so it was incorrectly creating blend arcs in the case when the previous segment had W movement. Now it detects ABC/UVW motion in either segment and skips circular arcs in this case.