Servo Tuning... Poor finish with deadband, and dithering without. Any suggestions?

[mmurray70]

Hi Guys,

I've always noticed a better surface finish on my parts in single axis moves compared to interpolated. I just fixed some bad thrust bearings which made a big difference but im still noticing that single axis cuts have a better finish. So i started to look at servo tuning. Seems like the deadband setting is causing my servo to lag behind and then overshoot creating a slight oscillation while moving at a constant speed. Im pretty sure this is what im noticing in my parts too. With a deadband of 0 the position error looks alot better. See attached pics.

But theres definitely an increase in the noise from servo just sitting still now. Is this anything to worry about? Should i leave it like this or can this damage the servo? Maybe damage commutator from sitting in home position or something? Should i try and find a balance that is fairly good for both or just not worry about it?

Can i change deadband setting on the fly? Maybe set deadband to 0 when i start my spindle and then back to 2 when spindle stops so I dont have to listen to it while setting up machine? Its not all that bad, i just really liked how my servos were pretty much silent before.

Similar Threads:

• Newbie- Poor finish on ABS
• Poor Finish on new machine
• Poor finish
• Need Help!- Poor surface finish
• Servo Motor Basics - Dithering

[TomKerekes]

Hi mmurray70,

Interesting. I'm surprised several counts of error would be noticeable. What is your resolution? But adding deadband is like adding backlash and will add error and be less stable.

I wonder if the larger dither might actually cause more wear than the higher frequency single count dither.

btw have you tried other deadband settings such as Range 2 Gain of 0.1?

Yes deadband settings can be changed on-the-fly. Keying off the spindle sounds like a good idea.

There is an AntiServoDither__ALL.c example that monitors for very small following error for a period of time and then reduces PI gains. I worry if your system really needs such small following errors (+/- 1 count) to cut well that might not be a good approach.

Another test might be all axes stopped for a long time.

Regards

[mmurray70]

My resolution is very good, at 121,920 counts per inch. So this works out to less then 0.0001 variation, but it doesnt take much to see a change in surface finish. And its amazing how much we can see and feel. For example, feeding a half inch endmill at 0.005" per tooth leaves a nice finish on aluminum, but you can still clearly see and feel every flute mark. And this would work out to a difference of only 0.000013 peak to valley between teeth!

I will try lower gains in the deadband area and take a look at AntiServoDither__ALL.c but I think ill be safe enough to just switch it out with spindle on and off by the sounds of it. I cant wait to try another aluminum job now and see if it actually makes a difference.

See attached pics for what i am talking about. You can see the cut across the bottom is super smooth (X only) and has a slight pattern when cutting the angle on the side. The pattern i am seeing seems to match the frequency in the step response graph too. Its more noticeable in person, but you can still see it a little in picture.

[mmurray70]

Tom, what exactly am i changing when changing deadband gain? Is this replacing the P gain value or I gain or is it setting a factor of original gains or what?

[TomKerekes]

Hi mmurray70,

Beautiful parts and educational comments. Thanks for sharing.

The Deadband Gain is applied to the error. So for example with Range of 5 and Gain 0.1 an error of 3 would be treated as an error of 0.3. See below:



Regards

[mmurray70]

Thanks Tom! I played around with different gains and by the time i got the oscillation gone, it was starting to dither again. After some more testing, Its really not all the bad. Im going to set it to change with spindle on to zero and then back to 2 counts when stopped and that will be good for me.

I ran a test piece and there is a definite improvement. Still not as great as the X axis cut, but i would say at least 50% better. Im happy with that.

One more thing, since were talking about changing servo settings on the fly, it looks like its ok for me to set P and I gain on the fly too right? I might try reducing gain in my MPG program. Still getting a slight vibration on 0.010 increments. Its only minor, but i like to have things perfect. lol

[mactec54]

My resolution is very good, at 121,920 counts per inch. So this works out to less then 0.0001 variation, but it doesnt take much to see a change in surface finish. And its amazing how much we can see and feel. For example, feeding a half inch endmill at 0.005" per tooth leaves a nice finish on aluminum, but you can still clearly see and feel every flute mark. And this would work out to a difference of only 0.000013 peak to valley between teeth!

I will try lower gains in the deadband area and take a look at AntiServoDither__ALL.c but I think ill be safe enough to just switch it out with spindle on and off by the sounds of it. I cant wait to try another aluminum job now and see if it actually makes a difference.

See attached pics for what i am talking about. You can see the cut across the bottom is super smooth (X only) and has a slight pattern when cutting the angle on the side. The pattern i am seeing seems to match the frequency in the step response graph too. Its more noticeable in person, but you can still see it a little in picture.

I think he means Encoder counts per rev, this is the important part

[mmurray70]

I think he means Encoder counts per rev, this is the important part

Encoder counts per rev is 48,000 with 10mm pitch ballscrews.

[mactec54]

Encoder counts per rev is 48,000 with 10mm pitch ballscrews.

That's a strange number, what servo motors do you have

[mmurray70]

Its the stock DC motors on a Fadal 4020 and the encoders are Teledyne Gurley model 8225-6000-DQSD.

[TomKerekes]

ok for me to set P and I gain on the fly too right?

Yes. Of course changing P may result in a step change in the output based on current error and your Filters.

Regards