emc backlash compensation

[Jon Elson <jmelson-at-artsci.wustl.edu>]

I just started experimenting with the backlash compensation
in EMC, and it seems that there is something wrong, there.
i am running a servo system, with encoder resolution of
.00005" per count.  I have a backlash of about .0013"
on the X axis.  I was able to get the backlash down to
practically nothing by setting the backlash value to .0045
I assumed that must be .0045 mm, but couldn't make sense
of that, either.

But, other than that, I got some very strange movement when
making .0001" incremental jogs with this backlash setting.
It did, in fact, return to the position I wanted, with the
machine's backlash almost completely supressed, down to
the ten-thousandth of an inch.  But, it went 15 thousandths
(.015") past where I requested, and then pulled back, at
every move.  Maybe this is due to the FF1 value I have, but
it seemed to be such a slow move, I rather doubt it.
It appeared that it moves some amount past the requested
coordinate, and then backed up, in an attempt to 'pull in'
the backlash.  This is fine for a positioning-only system, but
entirely unacceptable for a 3-axis machining system.
It would cause .015" gouges all over the workpiece.

The successful backlash compensation systems I have heard
of only insert the compensation amount on a REVERSAL
of motion of an axis.  IF you move in the same direction,
they don't insert any compensation, as this moves PAST
where you have commanded it.

Any comments on this?

Oh, yes, I'm still running the 20-Dec-1999 code.  I don't
know if there have been any changes that would affect servo
machines in this area.

Jon