Tuning Results Confusion

I am in the middle of doing a 3 axis CNC mill retrofit with a KFLOP/KANALOG combo, I finally got everything moving and "tuned" so it appears to be extremely smooth but I can't seem to figure out why my measured velocity graph looks so terrible. My actual position vs commanded position looks really good, but I can't seem to clean up the velocity. I have tried different encoders, different wire (shielded of course), separating feedback channels in different wire looms, different methods of powering the encoders, and finally, different grounding locations for the board AND the shield wire. I am starting to think I need to just leave it alone. Anyone have any advice on this matter?
Thanks in advance!

Hi EvanR,

Measured Velocity is normally noisy due to quantization error of the encoder. KFLOP samples the encoder every 90us and computes the velocity based on dx/dt where dx is the encoder change and dt is 90us. So the resolution of measurement is 1/90us = 11,111 counts/sec. However your noise seems to be ~ +/-40,000 counts/sec which would correspond to a few encoder counts of noise not just 1 count of quantization. Let me guess that you are using the CUI/AMT capacitive encoders instead of a true optical encoder?

Saving and attaching the raw data would allow us to zoom in and see things more clearly.

Hey Tom,

The mill came with Dynamics Research Corp encoders PN#: 152-021-500-18UJ

I cant find much on them online anymore but the closest i can come are these:
https://ecatalog.dynapar.com/ecatalo...769.1602014459

I also purchased these just because I was not sure if my encoders were any good (during my trial and error):
https://www.amazon.com/gp/product/B0...0?ie=UTF8&th=1


As far as the step response file, it appears to be too large to post it here. can I email it to you?
Thanks again

Hi EvanR,

Strange they would have that much noise. Do they ever lose/gain countsr? Are they differential? Where/how are they connected? What is the resolution?

Regarding the file: if you zip it (then add a .txt extension) and/or capture less time it should be under the 1MB cnczone limit.

Tom,

I agree with you completely on the noise, I was really surprised also. I thought it was from being in the same liquitite as the motor power & tach but if the shield was properly grounded I would think that would drain any incoming noise. Regardless, I changed the signal cable and just temporarily ran it outside of that liquitite, only to have the same result. I do have a VFD for the spindle in the same cabinet, however, it is only a 2HP drive and it was not even powered at the time (just in case that was your next question)..

The encoders are a 500ppr differential line driver, but they do not loose counts. When I was determining the resolution I set my scale at zero, completed 1 full rev and my count was 2000. i then reversed 1 full rev and i was back at zero.
I have them connected 5vdc and 0vdc connected to the KAnalog board on JP8 via a 2 conductor shielded cable, then I have a+, a-, b+, b- on JP1 of the KAnalog. I tried to isolate the a & b channels with separate (shielded) cables but nothing changed so I returned to the 4 conductor cable.

Hopefully the step response file gives you a bit more of an idea what's going on.

-Evan Rice

Hi Evan,

It seems strange. Its seems like where it should be simply moving slowly +1 instead it does -1, -1, -1, +5

Attachment 449990

Maybe something out of alignment with the optical mask to to the optical disk?

That is really interesting. I am going to try to change that encoder to a brand new one then review the same plot and see if the issue (-1, -1, -1, +5) clears up. If it winds up being the encoders are the problem, what brand do you recommend for the best results?

Hi Evan,

USDigital makes good encoders.