Gmoccapy & 7I96 jerky movement

Hi,
I have a curious problem with my 3 axis router that uses Gmoccapy and a Mesa 7I96. Generally everything is working great. I've got the MPG working, touch switch for tool length measurement, VFD speed via the PWM output.

However sometimes when running gcode the movement becomes very jerky. I've got a programme milling some letters of words where I've seen it. Increasing the feed rate makes it worse when it happens.

However, it does't happen every time, and when it does, restarting linuxcnc but not the Mesa solves it at least temporarily which makes me think it probably isn't Mesa related.

I tried decreasing the servo thread frequency from 1khz to 500hz but that didn't change anything.

Does anyone have any ideas?!

Configuration files attached...

Thanks,

Oliver

I would try plotting the following error so see if its latency related

Also adding:

# latch stepgen positions 100 usec before nominal read time:
setp hm2_7i96.0.dpll.01.timer-us -100
setp hm2_7i96.0.stepgen.timer-number 1


To the hal file should improve smoothness if there is significant jitter.



 

Hi,
I tried adding the lines to the hal file, but the problem is the same.

CPU usage from the Gmoccapy process is around 25% when it's happening. The machine is a core I3 with 4Gb of ram.

Attached is a screenshot of halscope when the error is happening. The ferror always seems to be at "10"...not sure of the units. Irrespective of the commands...

 

You are plotting the f-error limit setting, not the actual f-error,
which would be:

joint.0.f-error

(for joint 0, usually mapped to the X axis)

OK I managed to capture it in the attached.

The error isn't there when linuxcnc is first started, but after a number of runs through the same gcode it starts, and then continues until linuxcnc is restarted. I can pause the running programme and restart it, and the jerky movement continues.

It's pretty obvious once it's started because it makes the whole machine vibrates, but increasing the feedrate makes it even more apparent...

Pic of the machine in question attached just for interest.

OK so those errors (+ and - 0.002 mm) are negligible so the
stepgen is doing what's expected, that is, it's following LinuxCNCs
commanded position accurately.

Other possibilities are electrical (how is the 7I96 connected to the drives.
Some drives don't like differential connections as there is less drive current
Also might be a mechanical intermittent.

The 7I96 is just connected single ended to the optocoupler inputs for the step/direction on each drive....

The fact that a linuxcnc restart always makes it work properly again for a while makes me think it must somehow be related to the software on the PC?

Oh I did the same with halscope for the y axis and it looks the same, very small ferror...

Very strange...

Can you get a plot of velocity at the same gcode section
when its working and not working?

If LinuxCNCs commands are the same(which I would expect),
its likely a hardware issue of some kind.