Backlash compensation - is it non cumulative?

If I have a linear axis driven by a ball screw there will likely be a measurable amount of backlash which the controller can offset.

If the actual amount of backlash compensation programmed is say .05mm, whereas the actual backlash is only say .04mm is this .01mm difference cumulative or not over time and multiple axis reversals?

My guess would be not, given that the axis only has two directions of movement, and every time direction is reversed then the same amount of compensation is applied in the opposite direction thus cancelling itself. Thus I would say that if a backlash compensation figure is slightly incorrect it does not accrue and thus cannot send the send the axis majorly out of position over time. Obviously though, the axis position itself will be out by (half?) the difference between actual and programmed backlash, but again, this is constant and does not accrue over time after multiple axis reversals and applications of backlash correction, all be it that the error will lead or lag the axis depending on direction.

Am I therefore correct in saying that even though a backlash compensation figure may be slightly off the mark in real terms, that this positional error will not compound itself over the time the axis is run with all other things being equal?

Thanks.

Correct, backlash error is not cumulative, therefore the correction for it is the same.

If you are having trouble with a cumulative error, the problem is not backlash related, It would most likely be lost/missed steps or a slipping coupling.