Originally Posted by
ebrewste
What does it sound like? Clicks at the jerk points? If so, I would guess that is your acceleration feedforward (aff) doing its job. Aff causes a current (force) to be output proportional to acceleration. Since you are running trapezoidal moves, Those acceleration curves have square edges. Its like hitting your ball screws with a little hammer. Since feedforward is not a closed loop thing, it's bandwidth is only limited by your current loop performance. So to calm down those clicks, you either need to back off on your aff and sacrifice following error performance on acceleration portions or change your trajectory so you aren't commanding square acceleration profiles. Or put a low pass on your speed loop output. It will limit performance, while reducing the clicking, but you might find a happy medium. However, it will limit your closed loop bandwidth. The one practical downside of those clicks is that they might show up as ringing in the surface finish because they excite the mechanical structure. Or it might not...
I think you are generally getting quite nice performance. You may even think about backing off your system bandwidth. I had an experience when doing servo tuning for a recognizable VMC manufacturer. -- Don't set your bandwidth too high or you will shake the head / column on higher jerk portions of your trajectory. I had come from a background of always shooting for maximum bandwidth / minimum following error, but with max servo bw, those high jerk sections would get the head / column moving, as the servo jerk was like hitting the mill with a hammer. When we backed off on the system bw, contouring tests would have drastically less ringing shown in the tool paths. It was a mixed bag -- I could clearly measure the increase in servo following error, but that was less objectionable than the ringing left in the cut. That ringing was absolutely not measurable by us and was absolutely invisible to running your fingernail over it, but it was easy to see.