View Full Version : Servo Tuning - any experts here ?

08-09-2016, 12:12 PM
Having finally got my Bridgeport conversion up and running, i got to fine-tuning.

I did a home-all tonight and noticed the X axis was wobbling when doing the back-off part of the home sequence so it seems my inept attempt at auto-servo tuning was lacking somewhat. In the graph i could make it do the wobble-thingy at about 13% speed and it looked bad.

Reading the CSMIO manual, the tuning section is a bit vague but it seemed i needed to reach oscillation first then back off. Having seen what that does, i thought it safer to tune for the lowest following error and fastest dead-band arrival as i had read on various other papers, there does not seem to be any "scientific" method to tuning, just get it smooth, fast and oscillation free.

Tweaking the values in P-I-D order, i have now got far better following errors and as yet no wobble;)
Values are now...
X P=20,000 I=50, D=0, ff=65 Following error 135 at max speed
Y P=20,000 I=50, D=0, ff=65 Following error 105 at max speed
Z P=12,000 I=65, D=0, ff=60 Following error 253 at max speed

These errors are much lower than before and it seems happy at all speeds now so will leave it there for now.

Is there any advice for tuning or is it all suck-and-see??

08-09-2016, 03:45 PM
It's Black Art with bit of Using the FORCE thrown in.!! . . . . If you have stable and are happy just go with it. Tweak it occasionaly when your in Manic mood and fancy bit of self Mental torture. . .Lol

08-09-2016, 05:42 PM
Lots of experimenting!

There is some science behind it, but ultimately it's mostly trial and error.
Increase P until things become unstable, knock it back a bit, increase D until it becomes unstable, reduce it quite a bit, repeat. Add a little bit I, then tweak P&D.
FF can then be applied if needed to help reduce following error even more, however it's not always the best solution, and you should try to get tuning the best you can before applying FF.
You should also try testing the tuning under load, as dumping a big weight on the bed will make a noticeable to performance.

However, that's what I've learnt using Dynomotion, and they give you various test/plot/parameter screens for testing purposes, so you can see exactly how things are working. You're also able to apply various filters to remove oscillations at certain speeds/frequencies, which I don't think you can do with CS-Labs.

08-09-2016, 08:10 PM
Ok, thanks.

It ran well tonight so i'll leave it alone for a bit i think.

There do seem to be a lot of differing methods though ;)