I took it all apart nothing was loose. After many tests there is no backlash something else is wrong in my kmotion configuration. Starting from ground zero tomorrow and might hook up my spare Mach controller and see if all works
I spent the entire weekend playing with my machine and there were two issues causing my problem. The first was I put in the wrong velocity and acceleration in the Trajectory tab, misreading this and putting in inch/minute data into those places where KmotionCNC wanted inch/sec. The very fast acceleration did not allow the massive gantry to stop fast enough so it often would overshoot. The second issue which you suggested was backlash. On my Y-axis where I could see the holes not drilling in a straight line there was a measured .010" backlash in the HD harmonic drive. This was surprising for such an expensive harmonic drive. I then tested my X-axis which has the exact same harmonic drive and ballscrew, and found I could only measure maybe .0005" of backlash. I then added some backlash compensation into the Y axis of KmotionCNC. I populated the lines below, and ran my drilling example test and it was much better, still off but clearly much closer. I decided to order a replacement harmonic drive since the other one seems to be what I expected, not sure how the other one could have been damaged since it have hardly been used. Anyway if when I get the new one I still find some backlash, I can figure out how to change the backlashamount and increase it up slightly until it gets much closer, but I was somewhat confused on the backlashrate. How do you calculate the rate? I just put in a guess with the "500". I was also a little surprised that the steps per unit which was a mathematical calculation was off slightly. The calculation was 2000*11 on a 40mm ballscrew, so I had 13,970 steps/unit. When running multiple tests that number needed to be 13994 on my X axis and 14014 on the Y axis where I measure backlash. More to come after the replacement unit arrives on Friday.
Thanks for the information. I could not find that in the documentation, I used the actual offset measurement and clearly that was not enough. I will play with it again while I am waiting on a replacement HD and hopefully that one will be almost zero backlash which I prefer. Someone also suggested the idea of using long linear scales and close the loop using the scale which would also address backlash and ballscrew error, something I will consider. Thanks