Re: MaxNC modifications

hi Jon,

here's another slant on the situation...

since you report some part patterns can be machined with no errors, yet
other patterns show the error, let's consider a method that could produce
the errors even if the steppers never lose a a step.

taking a hypothetical machine that has steps at .001 inch, let's move
X 1.000 inch, but move back with three equal moves. no matter how far
we carry out the precision of the calculation, each of the three moves
will be .3333333333.... inches. our math will have any degree of
precision we want, but the machine will be forced to round off our
calculation. we move plus 1000 steps and move minus 333, 333, and finally
333 steps. we see an error of .001 but our stepper didn't lose a step.
now move 2.000 inches and back in thirds. we move plus 2000 steps and
then move minus 667, 667, and 667 steps. now our error is -.001 inches.

this is just a condition of G91 programming. it happens with servo machines
and stepper machines (although some servo machines can keep internal
register precision higher than the positioning precision)

-ron

On Tue, 15 Jun 1999, Jon Anderson wrote:

> My dad machines little resistive elements for a customer of his. These
> are generally under 1" dia and we run multiples of up to 8 at a time on
> a custom pallet setup. We've tried using G92 to set offsets for each
> nest, no subroutines; G92 with subroutines; G91 with and without
> subroutines; and if a particular job is going to show lost steps, it
> doesn't seem to matter how we approach it. Some parts do not lose steps.
> Each part will show a different pattern of lost steps. The pallet base
> has a .500 dia hole that we indicate from to establish datum, all
> pallets and all programming is referenced from this hole. It's real easy
> to drop an indicator in and check for lost steps.
> For example, one particular part run in multiples of 8 parts per pallet
> might show .0013 lost in X and .0003 lost in Y every for every pallet
> load. With the tolerances on this job, Dad can let this accumulate over
> two pallets before having to rezero. This is so consistent that he
> stopped checking after every two pallets and simply would jog X.0026
> Y.0006 and rezero. After running several hundred parts, he'd check with
> the indicator after the last two pallet loads and be off exactly
> .0026/.0006. (I don't recall exactly the sign (+/-) of the error or
> corrections, but the numbers are real close to what he runs into)
> This is with a program that does not use G92, it's done with geometry in
> a subroutine, we move to a set entry point, enter the subroutine, enter
> G91 and machine features, then back to G90 and exit the subroutine.
> We lose NO steps in the Z axis.