Closed loop stepper driver configuration help

I can't imagine the whole thing is flexing that much under cutting pressure, but I'd say next step is to mount a DTI on the table and touch the stylus to the spindle nose.

Then shove on the head and see if there's deflection.  Repeat for Y in both directions.

How much of a cut are you taking to get that much error?  Are you measuring the error by comparing the part vs. programmed size?

I'll try that when I can get a hold of a DTI. For now I went redid the gcode with a re-done tool table, and it seems to make the biggest difference so far. I'm now getting about +- .05 mm

I still think it can do better, but for now I don't want to take it apart it.

I probably need to test the spindle run-out as well at some point.

thanks for the advice it's been helpful.

do you have a caliper you trust? Take the backlash adjustment out of the config files if any is coded in there. Start with the X-axis, for instance. Chuck up a square piece of steel in the mill head, or a round piece with at least one flat. Lock or clamp your spindle. Move the head to the right say 10cm or more. Clamp the caliper to the table with the depth spike against the square or flat chucked in the spindle so that as you move to the left, the depth spike is pushed into the caliper measuring the motion. You start with a movement to the right to take up the backlash. When you move to the left, the backlash will then have to be taken-up in that direction and most likely the caliper will read something less than the motion expected. So you move to the right 10cm, push the depth spike up against the flat, move back to the left 10cm and read the caliper. Ideally you should see 10CM. You will see 10cm minus your backlash +/- your resolution. This assumes no tool table is in play as well as absolute mode, but you get the idea.

You can code a little backlash compensation into linux CNC but if you have backlash, the compensation will give you near perfect measurements back and forth, but when you cut something with the perpendicular axis moving, the head is going to drift depending if you are climb milling or conventionally milling. You will notice scallops on the finished faces. Backlash compensation in the software might also cause following errors due to the take-up, don't know on LinuxCNC but that was a fault in Mach3/4(?) because the backlash wasn't accelerated correctly. It was assumed to be loose and it would just take-up the backlash expecting no-load on the servos.

lengthy message, but you should be able to figure out your backlash this way. Pushing on the head works, but it only shows you the backlash at that particular leadscrew setting.. the other trick is to put slightly oversized balls in the ballscrew. That will take up some backlash. You can also use compression washers in your motor mounts if your screws are directly coupled. You loosen your motor, uncouple the screw, pull it out slightly, and then tighten the coupler up. then use the compression washers to tighten the motor into the mount carefully. This can take-up some backlash in the end to end movement of the screw but you don't want to bend anything.

Jerry

Thanks for the info. I'll try out the caliper method. I think the main issue I'm running into right now is tramming. But that's another discussion.