Hi Henk,

 

BTW a good Bode Plot is not absolutely necessary.  It is only a tool to help you.  Many Controllers (like your Panasonic?) may not even provide it at all.  I hope you appreciate the capability of KMotion to create a full spectrum Bode Plot at the push of a button which in the past would have required a number of circuit changes, signal injectors, scopes, and a lot of time and labor.

 

I was just reading on the LinuxCNC forum that they were discussing on how a Bode Plot would be useful for tuning.  And they included this:

 

>>Drawing on another one of my hobbies... Here is a description of some

>>techniques for this sort of thing, including a link to a writeup from HP
>>(from the 1960s) on using their wave analyzer.
http://bnordgren.org/seismo/lpmeas3.pdf

 

The article give an idea of the difficulties and includes this quote which I really appreciate

 

The difficulty with cut and try design techniques on feedback loops, is that to some degree the

various components of the design all interact. If changing one parameter results in oscillation, it

still cannot be said that that parameter “caused” the oscillation. It is quite possible that some

modest changes elsewhere in the loop may solve the problem perfectly well. Trying to get a loop to

work, “blind”, is one of the more frustrating activities I can think of.

 

Anyway back to your case :}

 

As I said in the last email I think you are mainly not using a high enough Amplitude of stimulus to get a reasonable Bode Plot.  I didn't see where you attempted higher Amplitude.

 

If I did the math correctly your system parameters regarding max speed are: 

 

3000/4 * 5 = 3750mm/min (148ipm) = 62.5mm/sec = 125000 counts/sec

 

You have been testing with small moves and low speeds (4000 counts/sec).   This is fine but you should also test at higher speeds.

 

As I said in your last email the 2nd to last plot from your previous email is quite reasonable by simply using P gain.  You didn't zoom in (left click drag) or plot following error.  But it seems to settle to 1 count which would be 0.5um.  So why don't you go back to that condition and let's see how well that works for you and if we can possibly make small incremental improvements.

 

Regards

TK

 

 

 

 

 

From: Henk Sterk <henk.sterk@...>
To: DynoMotion@yahoogroups.com

Sent: Sunday, May 6, 2012 4:19 AM

Subject: Re: [DynoMotion] Kflop tuning, which filter to use

 

Tom,

 

Spend an other day in the workshop, try to figure this out.

 

The screen  caps I send you yesterday, have little notes in the right bottom corner, that where you can see when I changed the settings in the Drive, the season I left the other in, so you can see the difference.

 

The reason why there is a change in Dac counts, is because I try to see if there would be a difference, using a higher count rate out of the MSDA drive (send to max) I must have forgotten the put it in the note (sorry).

 

 

I've tried something new today, I used my 4th axis (currently not connected to the machine (no load), which is also a MSDA drive only 200 Watt) used the encoder feedback from the drive to close the loop to Kmotion (all settings on the MSDA where factory default where needed). I did a Bode plot, and ended up with the same result, position not following the command in the Bode Plot, and the magnitude plot not crossing the 0 db line.

 

Don't know what to do anymore, I've tried about every setting there is (I think) from min to max (machine vibration).

Are you sure there is not a bug in the firmware, communication between Kanalog and KMotion?

Are there other people with this set-up?

 

As a last bit of info:

 

Max Servo rpm=3000

Belt step down ratio =4

Glass encoder counts = 2000/mm

Ball screw 5mm/rev C3.

 

Also a "Panasonic MSDA.txt" was included with the last mail, which is an export of the servo settings.

 

I don't want to give up on this (motion controller), as I know this should work, and the result is going to be amazing.

 

Many thanks for your patience and feedback.

 

Regards

 

Henk

 

 

 

 

On Sun, May 6, 2012 at 2:42 AM, Tom Kerekes <tk@...> wrote:

 

Hi Henk,

 

I'm having trouble following all that you did.  It isn't clear to me which plots are after "fixing" the amplifier settings and which plots are with motor feedback vs glass scale.  The last two plots are confusing.  Supposedly nearly same settings (P=0.3 vs P=0.35) and much different response.  The strange thing also is that in one case only ~14 DAC counts are needed and in the other ~45 DAC counts are needed to go at the same speed?   These are both very small voltages to the Amplifier.  But you are testing a relatively low speed 4000 counts/sec = 4mm/sec.  Is that correct?  What speed do you need to go?  It may help to reduce the input range of the Amplifier so that 10V in gives your max desired speed and not something much higher.

 

To get a good Bode Plot you need to look at the Time domain and Frequency Domain with the same settings.  In general I think your amplitude is too low and the frequency cutoff too high to get a good measurement.  The green plot and the right green scale it the output that is being driven to the Amplifier in DAC Counts.  In most of your measurements it is driving with 5 counts or less this is only 5/2048 =  0.2% of full scale!  Or 20mv out of the 10V range to the amplifier.  Also having the Output (green) follow the Command(blue) is not a good thing at all.  This will occur whenever you have only P Gain and no movement of the motor.  It is necessary to have the Position(red) follow the Command(blue) at least somewhat.  If it doesn't, then we are exciting the system outside of it's ability to respond and not really getting useful information.  Sorry it is so complicated, but that is just the way servo control is :}

 

As expected connecting the feedback from the motor rather than the linear scales is much easier to control.

 

You shouldn't have a problem in Mach3 that you don't have from the Step Response Screen.  If you do it just means that you are simply not testing with the same acceleration or distance of motion.

 

I don't think a D gain of 1 is doing anything useful so keep that set at 0.

 

I'm not familiar with the Panasonic Drive or any of its settings.  The better you can make it control velocity the better the position loop will be.

 

But basically the 2nd to last plot (#26) looks reasonable (maybe add in the small amount of I = 0.00001).

 

Regards

TK

 

 

 

From: Henk Sterk <henk.sterk@...>
To: DynoMotion@yahoogroups.com

Sent: Saturday, May 5, 2012 2:40 AM
Subject: Re: [DynoMotion] Kflop tuning, which filter to use [17 Attachments]

 

Today's progress.

 

I found an issue that improved some of the things:

There is an acceleration and deceleration setting inside the MSDA that was set during first trails  (jog at installation), and apparently  interfere with closed loop controllers (now set to zero).

This fixed the issue in the Bode plot, where now the command and the output following each other (the position line remains an issue).

 

I've also tested the Pole-Zero filter (as suggested), taking P all the way down combined with the 2nd low-pass 2 filter, that did not seem to improve our Bode Plot (open Loop) razing the Blue line above the 0 db.

 

Then I eliminated the timing belt and glass scale encoder, by connecting KMotions input direct to the drivers output (short cutting second loop), this improved the Bode plot a bit (Blue line), but still not what we are looking for (see screen caps) and also no improvement on the position graph in the Bode-Plot.

 

Mach3 is a lot more sensitive using G0 command, even had to take back the P Gain to avoid oscillations on deceleration (had to Kill the drive several times).

 

Can't quiet understand way we cant get an accurate position graph in the Bode-Plot we receive the counts very well in Mach3. If G0Y200 is given, it nicely race to 200.

 

I've also include a printout of the Panasonic settings.

 

Hope this will get us a bit closer (sure ill be an expert by the time we are done, lol)

 

Many thanks

 

Henk

 

Kuala Lumpur Malaysia

 

 

 

 

On Sat, May 5, 2012 at 2:04 AM, Tom Kerekes <tk@...> wrote:

 

Hi Henk,

 

I probably would not have "increased the velocity integrator" (whatever that is) in the Panasonic Drive.  We mainly want fast damping (velocity feedback) to kill oscillations.  An integrator is usually used to improve accuracy or persistant errors which for damping purposes don't really matter.  And an Integrator is highly de-stablizing.

 

Some minor comments: 

 

Your screen shots have critical info covered up.

 

A D gain of 0.001 will have no effect.

 

Please repeat the settings I've decribed in the previous email (lead/lag + low pass filter).  When you had violent oscillations you should have reduced the P gain to 0.1 or whatever to eliminate them.  There should always be a low enough gain to make it stable.

 

The belt could be a problem.  Unfortunately it is a Catch 22.  As we need the servo to perform reasonably to get a good Bode plot - but we need a good Bode Plot to see how to improve the Servo...

 

Regards

TK  

 

From: Henk Sterk <henk.sterk@...>
To: DynoMotion@yahoogroups.com

Sent: Friday, May 4, 2012 3:27 AM
Subject: Re: [DynoMotion] Kflop tuning, which filter to use [8 Attachments]

 

Dear Tom,

 

What am I doing wrong?

Spend most of today running test and making screen captures, (I'm starting to pull my hair out).

What I've done today:

* Yes I've read the links you presented me, it's getting a bit clearer but there are still holes for me.

* First re-tuned the MSDA (Panasonic) drives (using auto tuning) no setting was changed.

* Then increased the machine stiffness setting .

* Increased the velocity gain, till it started vibrating, then backed off a bit.

* Also increased the velocity integrator.

 

* All done inside the MSDA Drive.

Then turned all filters and settings off, with only the P=1 (inside KMotion)

There was a bit of improvement shown (less vibrations in the Position graph).

Then, turned on the lead-leg compensator (Pole-Zero) at n1,n2 at 15 and d1,d2 at 60hz but had to turn the machine off as violent vibrations appeared. so increased the numbers to a level where all vibrations where gone n1-n2 =58 d1-d2=230.

 

Also applied Low pass filter 2nd on Filter 2, as well as a normal low pass filter (all improved a bit). 

 

Then played around with a lot of settings al little steps (see Screen caps attached).

 

It sort of looks like its getting there (or it's in there somewhere).

 

One major issue still remains, when we look at the Bode Plot, I can't seem to get the blue line to cross the 0db (to get a 30 Hz Bandwidth), also can't get the blue to follow the red in the Bode-Plot Time-Domain (how important is this, and are there exceptions?).

 

Mach3 is responding well though.

 

I still have to mention that there is a Timing Belt between the Servo Motor and Lead screw.

 

Hope this will give you a picture of what is going on.

 

PS

 

Are there other people in this Group that have a similar set-up as what I've got (DAC in Cascade)?

 

Really hope you can help.

 

Many thanks

 

Henk 

On Fri, May 4, 2012 at 4:55 AM, Tom Kerekes <tk@...> wrote:

 

Hi Henk,

 

That doesn't look good.  I worry that your Amplifier is not well tuned.  You will need to test that using the Panasonic Software.  That is not our responsibility.  With a good/tight velocity loop handled by the amplifier simple P gain alone used for the position loop should give reasonable performance.

 

But otherwise:  lets shoot for a bandwidth of about 30Hz.  So set the lead/lag compensator to be centered around that.  Yours is much too high.  Try N1 N2 = 15Hz and D1 D2 = 60 Hz.

 

Also add a Low pass filter #2 of 2nd order 500 Hz Q=1.4 (this will greatly reduce spikes on the output - green plot)

 

For now keep I=D=0

 

Increase P until it goes unstable, then reduce until it is well stable.

 

Collect Plots (if you save as type *.png they will be much smaller).

 

Have you read the other threads about how to get a reasonable Bode Plot measurement.  There must be a high enough amplitude such that you get significant movement (red plot) without saturating the Output (green  plot).  Also the frequency cutoff needs to be reduced at the same time so that the blue plot is moving slow enough that the red plot somewhat tries to follow it.  If the red plot doesn't at least look like it is trying to follow the noisy blue plot the result will not be meaningful).

 

Regards

TK

From: Henk Sterk <henk.sterk@...>
To: DynoMotion@yahoogroups.com

Sent: Thursday, May 3, 2012 6:47 AM
Subject: Re: [DynoMotion] Kflop tuning, which filter to use [7 Attachments]

 

Dear Tom,

 

Thanks for the reply.

 

I've spend a lot more time on my machine today, and made some progress.

 

* I've used a Pole-Zero Filter n1-n2 at 58 Hz and D1-D2 at 300 Hz, this was by trail and error went all the way up and down, and this is where I ended up.

* Only P at 1 is left I and D only gave more instability.

* Try to use a low pass filter at various frequencies, but that did not get rid of the oscillations either.

 

Now the positive, Mach3 reacts really nice and solid G0 no overshoot or large oscillations, I had before the Pole-Zero Filter.

 

The question is, do you have any suggestions to further improve this system (please look at the attached screen shots) and how much more accurate you think we can get this?

 

Many thanks

 

Henk

On Wed, May 2, 2012 at 3:35 AM, Tom Kerekes <tk@...> wrote:

 

Hi Henk,

 

Nothing special looks wrong, but here are some comments.

 

Not sure how well your auto tune works so we don't know how well the velocity loop is being handled by your amplifier.

 

Your wiring is very clean :}

 

KFLOP is unaware of motor currents as those are handled externally in the drive in your case.  So select the plot modes which show Output (DAC) rather than currents.

 

D Gain has the main purpose of providing positive phase to increase phase margin.  Your D gain is so small it has no effect in the ~40 Hz range.  You can see that as the plot of the Compensator Phase on the Bode Plot Screen.

 

Here are some threads and topics that may help get better results and plots

 

http://www.cnczone.com/forums/dynomotion_kflop_kanalog/142000-optimizing_servo_tuning.html

 

http://www.cnczone.com/forums/dynomotion_kflop_kanalog/142770-servo_tuning_snapamp.html

 

http://tech.groups.yahoo.com/group/DynoMotion/message/2123

 

http://tech.groups.yahoo.com/group/DynoMotion/message/4233

 

Regards

TK

 

From: Henk Sterk <henk.sterk@...>
To: DynoMotion@yahoogroups.com
Sent: Tuesday, May 1, 2012 8:32 AM
Subject: [DynoMotion] Kflop tuning, which filter to use [4 Attachments]

 

Hi Tom,

 

My set-up is as follows:

 

Second hand EDM rebuild to cnc.

Kflop+Kanalog.

400 Watt AC Panasonic Servo's in Velocity control (auto tuning)

Glass scales on the table to close the loop to the Kflop (1 Micron resolution).

Using Mach3

 

I've been spending a lot of time tuning this system, and I'm very close (I think).

This is what I've got so far:

 

* Auto tuned the Panasonic drives using Panasonic software.

* Adjusted P I D  and Low Pass filter values, getting Command and Position very close.

* Very good position control and stiff

 

 

Still and up with oscillations.

 

So I've gone back to minimal settings, and hope you can tell me a bit more how to interperet the Bode plot, find attached file.

 

Many thanks

 

Henk