Servo Purchase Recommendations, Yaskawa?

More
23 Nov 2021 03:14 #227364 by ihavenofish
pure straight g64 path, smooth as butter.

Please Log in or Create an account to join the conversation.

More
23 Nov 2021 03:17 #227365 by Michael
Is the issue specific to analog? I would imagine it would also be a problem for closed loop pulse control.

Please Log in or Create an account to join the conversation.

More
23 Nov 2021 03:29 #227369 by ihavenofish
linuxcnc generally isn't used as closed loop with pulse drives - although i think its technically possible.

i assume analogue is not the issue, but linuxcnc directly controlling the position loop, with hardware that cant do what linuxcnc is asking.

were talking a lot of hypotheticals here. i have mine and other cases of un-tunable machines, but they all have in common old slow drives. for you, there's a pretty straightforward path. get new high performance delta/yaskawa/whatever drives that take both analogue speed and pulse control, and try it in analogue. if it works, yay. if it's like mine, just swap to pulse mode with a different mesa card. still beats forking out for a high end control with proprietary drives.
The following user(s) said Thank You: tommylight

Please Log in or Create an account to join the conversation.

More
23 Nov 2021 14:19 - 23 Nov 2021 14:27 #227406 by Todd Zuercher
Open vs closed loop and analog vs digital, have nothing to do with the problem. And with old slow hardware it is actually less of a problem. Jerk induced error is relatively insignificant and not really a problem when the acceleration and feeds are low as with most older milling machines.

I'm not sure what Ihavenofish was trying to show us in his videos. But as loud as his screws are I would be very suspicious of their state of wear. I recently had to have ball screws rebuilt on two machines that were causing large vibrations at direction changes. Even though I could not measure any backlash with a dial indicator. The screws were still badly worn and the backlash wear only showed up under heavy loads, giving the worn screw a springy feel, but they still tracked true under no load for traditional backlash testing. (This was on Fanuc controlled machines with jerk limiting.) Once the screw's sponginess was fixed the cuts were nice and smooth again (and movement was significantly quieter.)
Last edit: 23 Nov 2021 14:27 by Todd Zuercher.
The following user(s) said Thank You: tommylight

Please Log in or Create an account to join the conversation.

More
23 Nov 2021 15:15 #227411 by ihavenofish
the noise is the not the ball screws, it is the servos. they be very loud.
you'll have to read my other thread and the scope plots. at every exact stop move the motor direction inverts and slams the axis, or in worst case, resonates and squeals violently.

an any speed and any acceleration and any kernel speed (i tried down to 100hz and up to 10khz) this inversion exists. it cannot be tuned out. it is not caused by backlash or screw wear as the machine had only 0.0003" or so.

the point is, velocity control, for me, with those drives, simply doesn't work. i got similar reports from many other people, often with similar (brother) machines.

so, this is either a result of very slow drives, or a result of linuxcnc's overall control loop.

do you have a machine with linuxcnc controlling analogue drives in velocity mode? if so, what specs / performance.

Please Log in or Create an account to join the conversation.

More
23 Nov 2021 16:06 #227417 by Todd Zuercher
My point was I couldn't "measure" any backlash on these machines by jogging to a dial indicator in the normal way either, but when I applied 50-100 lbs or so of force to the axis then I could measure more than +/- 0.01". And the resulting motion looked and sounded like bad servo tuning, but it wasn't. I have seen similar behavior on different machines with different controls

Please Log in or Create an account to join the conversation.

More
23 Nov 2021 17:30 - 23 Nov 2021 17:34 #227424 by ihavenofish
backlash is irrelevant. the servo is being reversed on exact stop moves. it is reading the encoder on the motor and backlash doesn't come into play at all. i expect would do the same with no load at all. it is a function of linux commanding a speed, reading back the encoder finding it had moved not enough (jerk) overcorrecting, finding now its moved to far, and back. it could not be tuned to do any better than that initial overcorrection. in worse cases it would resonate. the z would always resonate, it never because remotely stable.

my machine behaved fine on the original control.
Last edit: 23 Nov 2021 17:34 by ihavenofish.

Please Log in or Create an account to join the conversation.

More
24 Nov 2021 20:47 #227529 by Todd Zuercher
The encoder feed back is on the motor on my machines as well, and the delayed and oscillating motion of the table was still picked up by the encoders and looked like reversals from out of tune servos as well.

I understand that you say you didn't have the problem with the old control, but the very brief video you showed also appeared to have much slower motion than the clip showing the problems (and there is no way to compare acceleration settings from the video.)  If you've increased the speeds (max velocity and or accelerations) from the old system It wouldn't be surprising that you would have these kind of problems. 

Also are you using velocity commands to the drives or torque?  I think a single PID loop torque command system looses the advantage of using feed forwards for tuning the PID in Linuxcnc.  (I am of the opinion that is a significant disadvantage and would make fast changes in velocity a problem.)

Please Log in or Create an account to join the conversation.

More
24 Nov 2021 22:25 #227537 by ihavenofish

The encoder feed back is on the motor on my machines as well, and the delayed and oscillating motion of the table was still picked up by the encoders and looked like reversals from out of tune servos as well.



I understand that you say you didn't have the problem with the old control, but the very brief video you showed also appeared to have much slower motion than the clip showing the problems (and there is no way to compare acceleration settings from the video.)  If you've increased the speeds (max velocity and or accelerations) from the old system It wouldn't be surprising that you would have these kind of problems. 

Also are you using velocity commands to the drives or torque?  I think a single PID loop torque command system looses the advantage of using feed forwards for tuning the PID in Linuxcnc.  (I am of the opinion that is a significant disadvantage and would make fast changes in velocity a problem.)


"The encoder feed back is on the motor on my machines as well, and the delayed and oscillating motion of the table was still picked up by the encoders and looked like reversals from out of tune servos as well."

no looks like, IS. its on the scope. as i said, and every speed and every acceleration. all the identical behavior. 0.01G and 1ipm, same reversal and "bang" or resonance. 0.5G and 700ipm, exact same bang, exact same scope plot.

"I understand that you say you didn't have the problem with the old control, but the very brief video you showed also appeared to have much slower motion than the clip showing the problems (and there is no way to compare acceleration settings from the video.)  If you've increased the speeds (max velocity and or accelerations) from the old system It wouldn't be surprising that you would have these kind of problems. "

not even close. the clip of the adaptive clearing is 400ipm programmed rate (so, between 200 and 400 real speed). on the original control it is 787 ipm (the max speed of the machine). acceleration isn't explicitly stated in the old control, but the manual suggests it is 0.5G (100ms to 4000rpm or similar, i don't remember the number), which is what i set linuxcnc to although as above, i tried accelerations down to 0.001g and up to 1g and nothing changes at all.

"Also are you using velocity commands to the drives or torque?  I think a single PID loop torque command system looses the advantage of using feed forwards for tuning the PID in Linuxcnc.  (I am of the opinion that is a significant disadvantage and would make fast changes in velocity a problem.)"

velocity mode.





 

Please Log in or Create an account to join the conversation.

More
24 Nov 2021 22:33 #227538 by ihavenofish
in any case, how did you fix the motion?

Please Log in or Create an account to join the conversation.

Moderators: piasdom
Time to create page: 0.114 seconds
Powered by Kunena Forum