Z offset can't remain consistent from print to print.

Hey everyone.

I installed auto bed level systems on 3 of my printers. I went through dozens of iterations on the best servo/switch/bracketing designs. I nailed down my auto bed level repeatability of the probe to a standard deviation of 0.01 or better. The problem is my Z offset changes from print to print. If I am lucky I can print 3 or 4 times in a row without having to mess with my Z offset.

I am pretty certain that my brackets are stiff and are not going to flex much with respect to the nozzle tip. I run arduino/ramps/marlin combo on all my machines. They are good machines, just having a hard time getting ABL Z offset to remain reliable.

Everything else works fine, the probe is very repeatable and sensitive, the M48 command confirms this. What else should I look into?

I just noticed something strange. I uploaded my firmware with a z offset of -4.5mm. When I run the machine, sometimes it requires me to go up to -6.5mm because the nozzle is floating in the air. I use the LCD screen to do this and I retry the print until I hit the right offset.

Then for some reason it jumps back to 4.5mm or the whereabouts, usually its off by a couple 0.01mm's. Point is, this machine has a mind of its own. It decides what the new offset should be and I have to play along. Nothing has changed mechanically, everything is the same. The offset almost seems arbitrary at this point because I cannot find a consistent value within a reasonable range. Where talking millimeters apart!!

If anyone can help me with this, it would be really appreciated. This is my single biggest problem I am encountering.

When something like this happens, it's usually mechanical. Check that everything is tight. In particular check the Z-coupler... I had over a mm of play in mine. Lubricate rods.

There's no funny noises from the steppers is there?

The problem is quite simple, you are using a linear device with a polar one (switch, servo) increase by servo glitches and less than exact positioning.

You think the servo gets allways at the same position, but in reality is not. Its too small of a difference to see, but enough to create triggering inaccuracies.

Get a proximity sensor or built a switch with a circular trigger.

I ran M48 command multiple times and it shows a standard deviation of 0.007. If it was mechanical then my deviation would be big but it isnt... I can check for slop but everything is tight.

If you ran your M48 commands one after the other without doing anything in between (e.g. lifting a print off the bed), it won't show up a mechanical change caused by whatever that action was.