[TomKerekes]

Hi Charles,

Oops. Yes the Red mark is the first line with an error. The DAC function requires an integer number of counts not a floating point double. Note you should set the max output to 2047 - DEADBAND in order to avoid any overflow issues. Try This:

#include "KMotionDef.h"

#define DEADBAND 700

void main()
{
    for (;;)
    {
        if (ch0->last_vel > 0)
            DAC(0, (int)(ch0->Output + DEADBAND));
        else
            DAC(0, (int)(ch0->Output - DEADBAND));
    }
}

To excercise the system you might use GCode subroutine with a loop count. Such as:

G20 G0 X0 Y0 Z0
M98 P100 L3(Call the Subroutine 3 times)
M2 (Stop)

O100 (Subroutine Label of 100)
F40
G1 X1 Y0
G1 X1 Y1
G1 X0 Y1
G1 X0 Y0
M99

[Charles2]

I will try the deadband shortly. Since I have not yet progressed beyond tuning, am I ready for g code programming? I thought I would need to still use MoveRel or something. Is there a way to use repetitive commands or loops at this level?

Did you check the Z axis plots and data?
Thanks

[TomKerekes]

Hi Charles,

If your axes are moving relaibly you could configure KMotionCNC and run GCode.

But here is an example C Programm Loop:

#include "KMotionDef.h"

void main()
{
    for (;;)  // loop forever
    {
        MoveRel(0, 10000);        // Move positive
        while (!CheckDone(0)) ; //wait till done
        MoveRel(0, -10000);        // Move negative
        while (!CheckDone(0)) ; //wait till done
    }
}

Plots look similar to previous to me. Maybe slightly better.

[Charles2]

I corrected and tried the Deadband on X axis. Output is now at a little over 600 to overcome the friction. See latest plots and Data.

I will try the Deadband on Z axis. I take it I edit the code to reflect DAC 1 and ch1 instead of 0.

Do you think my axis plots are ok to move on to KMotionCNC?

Thanks

[TomKerekes]

Seems to have helped some. I'm surprised 600 DAC counts is still required. Did you remember to configure the Axis as No Output? Otherwise the Axis and Program will fight each other controlling the DAC.

Yes your errors are less than 0.001 inch that should be fine for running GCode. You might read this and this.

[Charles2]

Yes, I forgot to configure axis as No Output. After doing so, the Output is about 200 DAC counts. Errors changed some.
See latest plots and data.

[TomKerekes]

You might be able to tune it a bit better now. I see at the beginning there is a jump ahead. Probably because we have too much FFs as so much is now not needed to overcome the Deadband to get moving. Although the reverse move behaves differently.

This might be part of the reason. We did a simplistic approach of when the motion should be positive apply +Deadband otherwise -Deadband. So when the forward motion stops the Output Jumps Negative. So then the integrator needs to ramp way positive to move the last little bit. And then when the negative move starts the integrator has negated the -Deadband already applied earlier.

Attachment 460052


We could be a little smarter and if moving forward apply forward correction and if backward apply backward correction and if neither none:

#include "KMotionDef.h"

#define DEADBAND 700
#define THRESH 2.0


void main()
{
    for (;;)
    {
        if (ch0->last_vel > THRESH)
            DAC(0, (int)(ch0->Output + DEADBAND));
        else if (ch0->last_vel < -THRESH)
            DAC(0, (int)(ch0->Output - DEADBAND));
        else
            DAC(0, (int)(ch0->Output));
    }
}

And still further all of the above but if not moving and the error is significantly negative apply forward correction, and if error is significantly positive apply negative, and otherwise none.

#include "KMotionDef.h"

#define DEADBAND 700
#define THRESH 2.0
#define THRESHFE 5.0


void main()
{
    for (;;)
    {
        if (ch0->last_vel > THRESH)
            DAC(0, (int)(ch0->Output + DEADBAND));
        else if (ch0->last_vel < -THRESH)
            DAC(0, (int)(ch0->Output - DEADBAND));
        else if (ch0->LastFollowingError > THRESHFE)
            DAC(0, (int)(ch0->Output - DEADBAND));
        else if (ch0->LastFollowingError < -THRESHFE)
            DAC(0, (int)(ch0->Output + DEADBAND));
        else
            DAC(0, (int)(ch0->Output));
    }
}

I'm not sure how much time you want to spend on trying to use tricks to overcome mechanical problems. For me its fun but rarely helps much.

[Charles2]

Thanks Tom.

I guess I need to create another copy of this Deadband routine for the Z axis (ch1)? So I would substitute '1' for '0' after 'DAC' and 'ch'. Correct?

Then I change Axis Mode output from 'DAC Servo' to 'No Output' , and lower max limits output by the deadband amount.

Please explain about the 'No Output' change.

I take it that these changes will be written into the Forever Loop later?

[TomKerekes]

I guess I need to create another copy of this Deadband routine for the Z axis (ch1)? So I would substitute '1' for '0' after 'DAC' and 'ch'. Correct?

Correct. Probably better to make a function that will work with a specified axis. But let's see if it helps first.

Please explain about the 'No Output' change.

Not exactly sure what you're asking. But with No Output mode the Axis computes what it thinks the Output should be but doesn't do anything with it and assumes something else will handle using it for something.

I take it that these changes will be written into the Forever Loop later?

Correct.

[Charles2]

Installed the latest Deadband correction routine. Output to overcome velocity friction is about 350 DAC counts. Tried changing Feed FWD values, but it seems to like the higher values.

Comments welcome.

See latest plots and data.

[Charles2]

Copied Deadband routine to Thread 4 in C Program screen. Saved to a different name. Changed all '0' to '1'. Changed Axis mode output to 'No Output' and Max limit to 1347. Plots not so good. Ballscrew coupling is rocking back and forth slightly without command. Tried lower and higher FF, no help.

See Plots and Data and Deadband routine.

[TomKerekes]

Hi Charles,

I'd prefer to work on one axis at a time. Its less confusing. And if it doesn't help significantly then it isn't worth doing.

Going back to axis 0 - again that looks to surge ahead during acceleration and surge behind during deceleration and otherwise looks nice and symmetrical. I'd be surprised if reducing AFF wouldn't help. You said it doesn't could you post a plot with it reduced?

[Charles2]

You are correct, as always. I kept lowering FF acceleration and velocity until the point of diminishing returns.

See latest plots and data.

[TomKerekes]

Well now its the opposite it falls behind at the beginning and overshoots at the end. Increasing AFF doesn't reduce that?

[Charles2]

Yes, increasing AFF does help. I raised and lowered AFF by small amounts and found what I thought was a median. Then lowered and raised VFF looking for a median, where all errors were minimal. Then tried different AFF settings again. Back and forth from AFF to VFF. Best overall results were as seen in the following plots and data. Sometimes when running the same settings over and over, the results would change a lot, mostly the return positive error would go from about 34 to 160 or more. If I keep running the same move, it usually eventually returns to about 34. I'm guessing that is mechanical, but I don't know what to do about it.

[Charles2]

Thought you might want to see plots and data of the rerun of the same settings where it changed so much.

You may notice in both sets of plots that I changed the Motion Profile Velocity from 35000 to 40000 and the Acceleration from 350000 to 400000. that didn't seem to cause any problems.

[TomKerekes]

Hi Charles,

It seems that last plot really got stuck. After the move forward it stopped with error -60. So the deadband program should be adding +700 to the output, The Integrator ramped up the Output to 1000 before it broke free, Which would be 1700 to the DAC. Nearly full torque.

The results being inconsistent strongly indicate a mechanical problem. I think stiction is normally caused by rough surfaces meshing together under load.

Besides the occasional stiction causing a temporary error of +/-0.002in or less the errors seem quite low +/-0.0003 or so. These are theoretical errors at the encoders. There will be other things that add other errors. Getting these to zero my not add significant benefit. What are your requirements? Do you know the original specifications?

[Charles2]

Hi Tom,

I have heard the term stiction used in machine tool maintenance, referring to the axis slides sticking to the ways on initial movement, sometimes due to lack of lube, sometimes due to the use of the wrong lube. Anyway, not knowing what lube was used in this machine, I drained the lube tank and filled it with 5-30 motor oil, hoping to dilute whatever is in the system with a less 'sticky' oil. As I have said before this machine has wayrods with bushings as a slide system. There is probably a good reason that this system is not seen on many machines, at least that I know of. I'm thinking this system has more friction than more conventional slide systems. I clamped off the lube pressure bypass and forced the pressure up by manually pumping the plunger. I used your exercise routine to run the x axis back and forth almost full stroke (560,000 counts, or 7 inches) for almost 4 hours, periodically hand pumping the lube system.

When I reran the plots, things had changed. See Data and Plots.

See picture of machine specs. Next post

My requirements are simple. I want the best accuracy, repeatability, and speed possible. I'll take what I can get.

[Charles2]

This is the Machine Spec sheet from the Warner/Swasey QCM-6 manual.

[TomKerekes]

When I reran the plots, things had changed. See Data and Plots.

Looks like Output required to move dropped from ~200 to near zero. But since we're adding in deadband of 700 that means it dropped from 900 to 700 or about 20%.

This is the Machine Spec sheet from the Warner/Swasey QCM-6 manual.

Nothing about precision or accuracy. 300ipm = 5ips = 400000 counts/sec. 10X faster than you are testing. We knew with the lower voltage it would be slower. But you could probably go faster if you need to.

I'd say configure KMotionCNC and try some cuts.