I have plenty of the same part to make.  More than I can cut in a day.  For starters, my system is a C32 board from CNC4PC that is fed from an ESS.  This goes out to my Dugong DG2S16035 drivers to my 26Lb/in servo motors.  This is all on a Bridgeport Knee mill that was originally made as a CNC.  My problem is that after setting my Dro's as Persistent DRo's and deactivating the zero upon home feature ( how come that doesn't stay off all the time when the check turns to an X???).  I have to deactivate the zero upon home so that when I home upon machine start up, which if I don't, the soft limits have a hissy fit, I won't lose my position on the dros.  I need all that because after all, if the dros stay the same I will maintin the same part Zero.  Well, the problem is this, after all that trouble, say every twenty parts or so the position moves by about .010" it seems in X and Y axis.  I am running at a fairly low MMR and an average speed of around 15 IPM, 3100 RPM, .514 Axial DOC and .032 Radial DOC w/ a 3/8 carbide 4 flute e mill.  If there is anything I can do about my zero being lost I would appreciate it.  I have little oscillation in my system,  My  servo drivers max error is 1500 steps, my encoders are set to 512 PPR and the encoders are read in 4X mode.     

Few things I am unclear about.
1. Why have auto zero disabled when homing?
2. Are your switches accurate
3. Are you losing position during part runs or is it only after restarting Mach?
4. If during part runs are you disabling the drives at any point during these runs?
5. Your following error is high but what that equates to in actual axis distance is unknown, for example it is nearly 3/4 a turn of the motor so that could equate to quite a distance depending on gearing and screw pitch.


Hood

1. Why have auto zero disabled when homing?
     In order to try and maintain the same physical zero that I established on the part the day before.  When I home, there is a noticeable position change by around .020-.010". 

2. Are your switches accurate
     These aren't dedicated homing switches.  I do have another set of homing switches on each axis, but not enough inputs yet to utilize them.  They are Baluff switches, Made in Germany.   

3. Are you losing position during part runs or is it only after restarting Mach?
     I check to see that my cutter is touching the same points I previously zeroed on every say 25 pieces .  I see that I have to travel another 10-20 thousandths in order to touch the workpiece.

4. If during part runs are you disabling the drives at any point during these runs?
     There might have been an E stop somewhere but for the most part, No.  Cycle stop seems to do a good job when the tables start moving where theywant to move and not where they are commanded ( I already had to junk two pieces because when I hit cycle start, instead of following the commanded path the cutter just went full into the work piece at way too high a speed and feed)

5. Your following error is high but what that equates to in actual axis distance is unknown, for example it is nearly 3/4 a turn of the motor so that could equate to quite a distance depending on gearing and screw pitch. 
     Theoretically its 10 rotations per inch, but the values derived from calibration are a different story by say +/- 25 step pulses.  If that is because of the pulley set up (The pulley on the ball screw and pulley on the motor shaft supposed to be 2:1), I am not sure.   

 

1.
I still dont really follow on the auto zero, maybe I am not understanding how you have your work offset configured but normally you home, move to where you want the XYZ zero to be then Zero the XYZ DROs and thus your work offset is achieved. Assuming you do not zero these again then the offset should be the same  the next time you start Mach and home, assuming of course you have persistent offsets chosen.

 2.
If switches are not accurate then really you need to touch off a known point to set work offsets each time you start Mach.

 3.
Kind of sounds like its either the Step pulse that is the wrong state for your drives or possibly just the large following error you have.

4.
Ok you say Cycle Stop does a good job, are you pressing that whilst in motion or just at a standstill?
Table moving to wrong position? Is it showing that wrong position in toolpath view?

5.
Sounds funny that if they are indeed 2:1 and I presume its original ballscrews?, that you have to have different values in motor tuning than from calculated values.
Personally I much prefer calculation as that way it is equal in accuracy to the hardware you have where calibration is only as accurate as your measuring.

Hood