Measuring backlash more accurately?

I have been fighting a problem on the gantry mill. The Y axis seems to
"drift" so that over the course of macining a contoured part (say 10 depth
passes of outline and internal pockets) it has moved 20 thou or so. For a
part that has 30-thou walls, this has created great frustration.

After completely rewiring the controller box, fully separating signal from
power grounds, minimizing wire runs, twisting pairs, etc. I made another
trial run this morning, with no help.

I realized that the backlash settings might not be right. (That's about
the only thing left...) I had measured the backlash on each axis by the
conventional method (move + increment, zero indicator, move + increment
again and then - increment and write down the reading) but after a program
that has maybe 100 Y moves, it seems to me that I need a more accurate way
of measuring the backlash.

Is there a way to cycle the axis so the backlash accumulates? I could then
run the cycle 10 or 50 or 100 times and divide the resulting accumulated
backlash by the number of cycles. I can't think of a cycle though, that
just doesn't run back and forth in the backlash window.

The only other thing I could think of is to input a backlash value that I
know is wrong, and run the machine back and forth and see how it "drifts",
tweak the backlash value, run it some more and repeat until it doesn't
drift.

I have a Starett "Last Word" indicator with half-thou graduations. I can
estimate to maybe one-tenth but maybe more reliably to 2-tenths.

Any advice?

Thanks!

Randy

Randy Gordon-Gilmore ,----.___________ ______________ _________________
ProtoTrains // = = === == || == == == = || == == == = == =|
Rio Vista, CA, USA /-O==O------------o==o------------o==o-----------o==o-'
zephyrus@... http://www.prototrains.com