That's probably reduced too much. Error is now huge. Now you're not weaving but running off the road :)
Not sure why I gain and FF are reduced.
Could you try my other suggestion?
Printable View
That's probably reduced too much. Error is now huge. Now you're not weaving but running off the road :)
Not sure why I gain and FF are reduced.
Could you try my other suggestion?
Lowered P Gain to .1 to get ahead of the oscillation. Not sure just how much oscillation is acceptable. I'm sure none is best. Set I and D to zero. Increased P gain until it started oscillating. Axis did not noticeably oscillate, just the plot.I had previously been looking for the axis to go into wild movements back and forth. Backed off P gain. Increased D gain until I saw slight oscillation on the plot, then backed off. Increased P gain more, then D gain. Added I gain. Added Feed Fwd. Tried many combinations. The following is where I'm at currently. Open to suggestions.
Hi Tom,
I didn't see your last message before I posted the above. The last is of course worse than the earlier. Before the last two posts, I was increasing the P gain looking for machine oscillation. I was unsure about the plot oscillation, thinking the 'spikes' were something other than oscillation. I could raise P gain way up without machine oscillation, but making the plots worse. Even more confusing, the error plots were getting better. Now, when I start over with low P gains and raise it above 0.14, with a D gain of 20, the single 'spike' of output begins to become two. Am I supposed to keep the output to a single 'spike'? When I do, I can't seem to get the position line to follow the command line. I am obviously missing something here. Thanks for your patience. I've seen where you have pulled and pushed others through this process. Hopefully you can help me.
Just a thought. The AMC 30A8 drive is used so I don't know if the pots have been adjusted or not. I moved the #3 pot (reference gain) back and forth a little. It seems to have a similar effect as the P gain. Set it back where it started. Per the datasheet, I turned #1 pot ( loop gain for voltage/velocity modes) fully CCW. Did you find the datasheet and manual?
Thanks,
Charles
Hi Charles,
No oscillations at all is not really the goal. Oscillations can either grow (unstable), remain constant(unstable) or dampen down (stable). The best performance is normally where any oscillation dampens out over a few cycles. Or where the next cycle is only about 20% of the size of the previous.
On a side note a theoretically perfect mechanism will be linear in that it will behave the same regardless of the amplitude of excitation. However, friction, stiction, backlash, limited voltage, and such are non-linear. I think your system has quite a bit of friction and more friction when stopped. That is why it tends to go unstable while moving and is stable when stopped. Also as it goes unstable and oscillation grows at some point something changes like effective amplifier gain reduces and the oscillation ceases to grow. I just point this out to help understand the plots.
Again try increasing P until it goes unstable. Back off. Increase D until unstable. Back off. Repeat until nothing changes.
I found this in the manual:Quote:
Just a thought. The AMC 30A8 drive is used so I don't know if the pots have been adjusted or not. I moved the #3 pot (reference gain) back and forth a little. It seems to have a similar effect as the P gain. Set it back where it started. Per the datasheet, I turned #1 pot ( loop gain for voltage/velocity modes) fully CCW. Did you find the datasheet and manual?
Attachment 459356
I assume you have those switches set as shown in the Current row? Yes reference gain will adjust the overall gain and because your tuning is currently only using P gain then they will be similar, You'll want to set the amplifier reference gain high enough so that during your most aggressive moves KFLOP doesn't output near 10V (2047 DAC Counts). And not so High that KFLOP is always only outputing tiny voltages.
Hi Tom,
I do have the switches set according to the Current row. I will try tuning again as you suggest and send results. Thanks for your help.
Hi Tom,
I started again, increasing P gain to about .5 above which there was a noticeable surge in the axis motion. Started increasing D gain. went up to about 400 with no ill effects. Started raising P gain again, up to 5.0, with changes in plot, but no surging in axis. Plot got increasingly busy, but cycles always diminished. The higher the P gain , the better the plot looked. Started adding I gain, Feed fwd, and trying different profile V, A, J. Settled on the attached plots and data. Error is still high, but i wanted your comments. Is this considered stable or unstable? I assume the axis surging is unstable.
Attachments failed to work. Will retry.
Data and Plots
Hi Charles,
At what D gain did it go unstable?Quote:
Started increasing D gain. went up to about 400 with no ill effects.
Assuming that is the best you can get with P and D gains. At what level of I gain did it go unstable?
Note that the error remains behind ~ 150 counts throughout the entire move for almost 1 full second. The integrator (I gain) accumulates errors over time and corrects them. You can imagine yourself with a knob controlling motor torque. Whenever you are behind crank up the force. Whenever ahead reduce the force. The approach guarantees the average error over a long time is exactly zero. The I gain determines how fast the knob is turned. Notice throughout the move the error does reduce somewhat from around 200 to 100. I think this is due to the Integrator action. To move at full speed about 800 DAC counts of output is required. This indicates your system has significant friction. Without friction no output should be required to cruise at constant velocity. The system is requiring about 1/3 of full torque just to keep moving. We can also see to accelerate at the beginning of the moves a large peak of ~1600 DAC counts is required. But to stop the torque doesn't even reduce to zero. So friction is doing all the stopping. Notice also after the move forward when the motion stops there is an overshoot of ~80 counts and it gets stuck there. The output is at ~ +400 and there is no + motion at all. This indicates ~20% of the output can be applied and the system doesn't overcome stiction and move. Notice the Integrator, because of the constant +80 counts of error is ramping the output from +400 down to +200. If given enough time it would eventually ramp down to something like -600 and break free of stiction to correct the error. But the reverse move occurs before this. Without I gain or VFF and with a P gain of 4.5 about 150 counts of error is necessary to create the 800 DAC counts to keep moving/
So without somehow reducing friction/stiction you might increase I gain so it can reduce errors faster. Hopefully early in the move. Again increase I gain until things go unstable then decrease somewhat.
Note now that your errors are in the +/-400 range you might set the Max Following Error to 1000 so that if things go unstable the axis will quickly disable instead of thrashing itself.
VFF can also be used to reduce average error throughout the move.You might temporarily reduce I gain so it has no effect then increase VFF until the average error during the constant velocity portion of the move is zero. At V=35000 and VFF=0.03 it is creating V x VFF = 35000x 0.03 = 1050 DAC counts of output. However the D=450 acts like a pure friction drag of 35000 x 450 x 90us = 1417 DAC counts. Because we need about 800 DAC counts to move the VFF needs to generate 1417+800 = 2217 DAC counts. So I'm expecting VFF to need to be around 2217/35000 = 0.063 to help.
HTH
Thanks Tom for taking time to look it over and evaluate. I'm really struggling to make sense of it all. I usually catch on quicker than this. So many variables and how they interact! And I'm not sure sometimes what I'm looking for. The idea of stability/unstability is obviously not black and white to me. Raising gains until the system is unstable has not been a clear line of demarcation, unless the axis starts going nuts. Sometimes it seems that I can raise gains way beyond what I think is reason without knowing if the system is stable or unstable. As I said before, I'm missing something here. To answer your question, I don't know if D ever did go unstable. I Gain seemed to make the plot change drastically at about .0004, if I remember correctly.
I changed FFWD V from 0.03 to 0.063 as you suggested.
Thanks again for your patience.
Attachments failed again. Will retry.
Data and Plots
Hi Tom,
Since you mentioned that this machine appears to have a large amount of friction or stiction, I decided to try to check a few things. I disconnected the X axis ballnut from the slide. The NSK brand ballscrew had plenty of oil on it and the ballnut seems to turn on the screw as I would expect. As I have said before, this machine is mechanically like brand new. The original control had about 400 hours on the clock, and I think most of it was powered up time, without cutting chips.
I wrapped a string around the ballnut at a diameter of 1.691", and slowly pulled the string with a digital scale. After several tests, the average breakaway measurement was .7 pounds, and the average slow steady pull was .4 pounds, in both directions.
I could not easily test the ballscrew, motor, bearing assembly with the string, but turning it by hand I would estimate that it takes about the same force or less than the ballnut.
Moving the slide back and forth on the wayrods (see attached pictures) was fairly easy. It was a little stiffer near each end of travel. Starting off from near one end or the other after setting still for a few seconds was sometimes surprisingly stiff. In the center area, start up after setting was easier. Changing directions in the center area was very easy. In the first tests, measuring with the digital scale, starting the move from near one end of travel took about 24-32 pounds. Slow, steady movement measured 12-14 pounds.
According to the pictures and diagrams in the machine manual, the slide has bushings with oil grooves and lip seals that ride on wayrods. The lube system has been verified to provide Vactra 2 oil to the slide near the back at both wayrod locations. I cannot confirm that each bushing is actually wet with oil. The wayrods have an oil film on the rear side of the slide, not as much on the front side. I put extra oil on the wayrods at all four seal locations, mostly to try to lube the seals, not expecting the oil to penetrate the seal into the bushing area. I will later try to pressure feed oil into the bushing areas and test again.
After lubricating the seals, and working the slide back and forth many times, the start force from random places checked 8-26 pounds. The slow, continuous movement force checked 8-10 pounds. Changing directions in any area other than near each end of travel required the same 8-10 pounds of force as slow, continuous movement.
I don't know how these measurements would compare to other machines, but I suspect they are reasonable for a machine that has never actually been used.
Attachment 459434Attachment 459436
Hi Charles,
I'm not good in mechanics so I don't know if those measurements are reasonable or not. It seems you made some improvements so you might test again with the same settings and see if things are any better. Maybe when things are connected there is some binding?
Again it just seems odd that ~ 30% of peak torque is required to start moving.
I don't know if you have a means of measuring torque or force or motor current. You might disable the servo and command DAC commands to see that +/- 400 is required to start moving. Of course be careful as constant acceleration can quickly get to a high speed. There might be some issue with the amplifier having a deadband.
The other odd thing about the plot is as soon as we start to slow down the axis quickly falls way behind. Normally the tendency would be to go ahead. That's what Acceleration FF is used for. I expect the AFF is set too high causing this. So as soon as the trajectory starts to decelerate we slam on the breaks. So you might reduce or eliminate that. Unfortunately that will make the start of the move fall behind even more. So you should try to find a balance.
You might now add back in the I gain to see how much improvement that gets.
The plots aren't horrible. When in motion the errors are in the +/-20 count range or +/- 0.00025 inches.
I reattached the ballscrew to the X axis and ran more tests. With P=4.5, I=.0005, D=475, and FFV=.063. I adjusted Feed Fwd A up by .0001 from .0031 to .0035, watching mostly the Error Plot. Some of the errors went higher and some went lower, as you predicted. I adjusted FFA from .0029 to .0025, settling on .0028 as what looked like the best compromise. I then changed the I Gain up and down from .0005, finding more improvement in error plots with lower numbers. Raised P gain gradually to 6.5, while also gradually changing I gain, FFV, and D gain. The attached plots and data show the latest results. Looks like about 80 counts error at the pause between fwd and rev movement, if I'm reading it right. Haven't been able to improve on that.
I found that it takes at least 800 DAC counts to move the X axis in the center of travel. It takes up to 1000 counts closer to the ends. If this is too much, I may have to disassemble the slide and try to loosen it up.
I think the I gain is way too low. While it is stuck 80 counts off before the reversal the Output is stuck at 600 and there is no correction. The Output is ramping at such a low rate it would be a long time before it was corrected. One thing to be aware of is with any Integrator + and - errors sum to zero. Or the area under zero will equal the area above zero. Or basically undershoot equals overshoot. So with a system like yours which tends to lag behind at the start there will always be overshoot. With very low I gain the overshoot is typically tiny but remains for a very long time. With more I gain the overshoot will be bigger but for less time. With higher still I gain the lag might be reduced and the overshoot will occur early withn the move rather than at the end or afterward. Even though higher I gain may not look quite as good as without it it is usually best to have it. I gain is feedback where FF is open loop. Sometimes it is easy to fool yourself using FF to have small errors. But if anything changes like friction or load the result will be very poor. I gain is making corrections based on what is actually happening.
As you suggested, I gain was too low. See latest attempt.
I think that is very good except the start/stop due to the large 800 DAC counts of friction. Probably the best possible without fixing something mechanical. You might test a move at half speed to see if the "friction" drops in half or remains at 800 DAC counts. That would indicate whether the "friction" is a constant force or of a viscous nature proportional to velocity.
I'm tempted to try some sort of "dead band" compensation. Such as whenever the trajectory is moving + add 700 counts to the Output and whenever moving - subtract 700 counts. Although I've never had much luck with such things. There isn't any parameter for such a thing. But you could do it with a C Program by changing the Axis to NO OUTPUT then outputting with a C Program like this:
Code:#include "KMotionDef.h"
#define DEADBAND 700
void main()
{
for (;;)
{
if (ch0->last_vel > 0)
DAC(0, ch0->Output + DEADBAND);
else
DAC(0, ch0->Output - DEADBAND);
}
}
My second AMC drive finally came in. It is a Brushless B30A8Q. I can't find what the 'Q' means, but I set it up for Z Axis by the AMC recommended settings for Brush motors on a Brushless drive and it seems to work fine. Used the same Step Response settings as X Axis. Plots are similar, with a smaller initial error.
Tried changing Velocity on both axis from 35000 to 18000 to see if the friction looked the same, It is still around 800 DAC counts at half speed on both axis. Must be the nature of this type of wayrod/bushing configuration. Plus the fact that this machine probably hasn't ever run enough to be considered 'worn' or broken in. I loosened the set screws at each end of one of the wayrods and tried to bump it in. I was going to remove t if it would move easily. It didn't. Must be pressed into the carriage. I am reluctant to force it as I might damage the alignment. It would be nice to set up a repeating movement and let it run for a few hours, then see if it loosens up any. It does seem to improve after the first couple of cold move commands.
Have not yet tried the deadband. I need to look back at your notes on loading code.
See data and plots on Z Axis with second AMC drive and same Step Response settings as X axis.
Hi Tom,
I tried to load the Deadband code you provided, but it had a compile error. Does the red arrow point to the line with the problem? I'm not familiar enough with the code to see problem in this. I'm guessing the invalid operand for binary operation has to do with the '0' after 'DAC (' . I'm tuning the newest drive on Z axis. Will send plots shortly.
Thanks
Charles
Latest Z axis plots. Seems to like a little more Velocity FF. Lower total Output than X axis with similar settings.
Tom, do you have an example of a short axis exercise program that I could let run for a couple hours to see if the axis loosens up a bit?