PID tuning step & dir servos?

Could use some guidance from the community.  First time using linear scale feedback and tuning LCNC's PID.

• 1kw S&D servos
• drives set/tuned internally to be as stiff as possible without noise/vibration
• 1um glass scales
  • scale FB sent to PID FB (loop is closed)
• Lathe Z-axis
  • ~75lbs of mass
  • 5mm ballscrews
  • MAX_ACCEL = 400 in INI and for stepgen

Attempted the PID auto-tune process in LCNC... was a bit of a mess.  Resulted in way too high P and I, plus lots of hunting/dithering.

Manually adjusting and watching halscope here's where I am:
P=125
I=200
D=0.3
FF0=0
FF1=1
FF2=0.006
FF3=0
deadband=0.0005 to stop a little dither

Attached is a 1" move at 200ipm.

What is the "f(-0.18200) = -0.00001 (ddt 0.00000)" value in halscope?  When I hover over the trace, the maximum value of the largest peak is 0.00628.
Does that mean 0.00628" of error at that point?

Am I on the right track?
Any suggestions on tightening up the f-error?
 

Just to be on the safe side, is all set and working properly without feedback to LinuxCNC?
I seems way to high, start from 0, same for D and FF2
FF1 will most probably not be 1, except if you have the output scaling done based on max velocity of servos/motors.
Use a wee bit of FF2 only if you still have spikes at start/end of move after testing the above.

@Tommy,

• Yes, it's working reasonably well open loop.
• For I, the auto-tune came up with a silly 3000.0 or so.  I dialed it way back, and also found a comment from PCW regarding a similar S&D configuration:

• FF1 - Steps/inch is 50810, and INPUT_SCALE = 25398, so I'm guessing I need to drop it from 1.0 to 0.5 to start?
• FF2 and FF3 both seem to help with the initial spikes, but I'll work on the above first.

Best I can do so far at F50 (ipm).  X-axis is a bit better.

Is the "0.00299" in halscope (the start peak) 0.0029 inches? Or some other value?

Also wondering if the wobble during steady state is the internal drive tuning and there's nothing I can do about that in LCNC.

I and D didn't seem to make anything better.  Tried really low number up to I >4000.

Any sugestions?

EDIT-  added X-scope for comparison

I suspect error spike is because of the delay from command to motion
in the drives. Unless that is something you can improve in the drive tuning,
the only other options I can think of are:

1. Reduce the acceleration. I would expect the peak error to be roughly
proportional to acceleration.

2. Use open loop (and assume the drive delays are the same
for both axis)

3. Encourage the people working on a jerk limited TP

I did notice improvement when I cut down the accel.  But I'm greedy and want a fast machine.
I've got garbage ballscrews, and no hope of hitting decent tolerance without the scales.
I'm running out of 'thumbs-up' in the couple of jerkTP threads.

I'll fiddle with the drive internal settings, thanks.

Are the ball screws that bad to cause all the wavy feedback between spikes?
If not, fix that first, then chase spikes. Feedback should track pretty flat during motion at constant speed.

tommylight wrote:

Are the ball screws that bad to cause all the wavy feedback between spikes?
If not, fix that first, then chase spikes. Feedback should track pretty flat during motion at constant speed.
 

No Tommy, they're not that bad.  I was exaggerating - they're remarkably good for Amazon.  About +/-5um along 300mm according to the encoders.

I think the waves are from the drive; it's got a 'real-time' auto tuning feature I left enabled.

After taking PCW's advice and fidding with the drive's settings I knocked the f-error in half for both axes.  They sound nice, no dithering, pretty stiff.

I'll wait to get tools installed and take some cuts before circling back to see if it needs improvement.