Tuning issues (RESOLVED)
Halscope can only see what the encoder sees. And, the servo loop strives to make the encoder follow the commanded position. So, even if the machine is vibrating, the PID will actively strive to REMOVE the vibration from the encoder's position, even if that makes the vibrations WORSE! This is the whole issue of mechanical resonance. What you need is to make the natural frequency of the mechanical system higher than the servo bandwidth. This can be really hard to do, and is why milling machines have tons of cast iron in them. If you can't make the natural frequency of the mechanicals higher than the servo bandwidth, then you need to put notch filters in the servo system to eliminate response to the resonance. That can be done in velocity servo systems with tachometer feedback.
bevins wrote: Yes but I don't see any instability in halscope, but on the machine you do. So the interpretation halscope is doing is not that accurate in my opinion..
One way a control systems engineer would attack this is by attaching an accelerometer to the axis, and plotting perhaps servo drive against the accelerometer. Use a digital scope with FFT feature to extract the frequencies, then put notch filters over the peaks in the FFT and see if that fixes it. The accelerometer could be placed near the spindle to see what will actually occur at the cutting tool. You could also use the accelerometer to analyze the effect of improvements to the axis mechanical condition.
Todd Zuercher wrote: The picture of your problem cut looked like a direction reversal with a multi axis move.
Problems at a direction reversal strongly indicate a backlash problem. And, of course, if resonance issues are present, backlash will likely make them a lot worse!
For all intents and purposes it is very useful, todays phones have plenty of sensors and quite sensitive.
The P was simply too high and I couldn't see it in the halscope and probably what John was mentioniong. I could feel it in the axis. dropped the P and everything is working perfectly.