Closed loop with linear encoders but still backlash?

Gnevko wrote: Of course mechanical, but a closed loop system with linear encoders should solve exactly this problem or not?

I would think adding BACKLASH = .03 in the ini would remove it electronically.unles there is a tuning problem

I don't think so.
The backlash setting in the ini file will make the motor "catch up" a certain amount of steps or angle on each reversal. Since your feedback is a true position you don't want that.

It will work when your feedback comes from a motor encoder.

If you have mechanical backlash you will get a small overshoot because the motor is commanded to reverse but the joint will experience a small delay

Best way to fix this is to eliminate the backlash. Easier said than done, I know.

Henk

PCW wrote: What is the following error when you read the 0.03 mm position error?
(that is, what is joint.0.f-error?)

A 0.03 mm error that HAL / LinuxCNC sees should result in a motion of
2.7 mm/s to correct the position error with your current tuning parameters
( 0.03 * 90 = 2.7 mm/s )

That is the exactly the strange sings - in linuxcnc it seams to be all ok: the following error is something about 0.0003 and no position error at all. But if I check the result with a dial gauge (digital with resolution 0.001 or manual with resolution 0.01) or with HAIMER 3D finder (resolution 0.01) - I see the backlash.

Steps to reproduce are simple: move the axis in "+" direction, set zero in linuxcnc and dial gauge for the axis, then move the axis again in + direction on +1mm and -1 mm. After that in Linuxcnc I read 0.0000 mm (as expected) but on dial gauges ca +0.03mm.

Try making the Deadband=0 and add some I gain. That should force it back to 0.

Theoretically, I term is not needed, The reason for this is that the stepgen has a very wide
velocity dynamic range (the 7I95 stepgens minimum settable frequency is ~.025 Hz)

With the OPs 1200 steps/mm settings and P term of 90, the minimum
correctable error is ~.2 nanometers, way below the scale resolution

I term may be useful for analog or low resolution velocity mode systems

I would check that

1. Your DRO is reading actual rather than commanded position

2. You don't have lost motion between where you are measuring and the scale

Because if you are reading actual position and it has no error, the scale is in the correct position but
the rest of your mechanics are not. This might be the result of loose gibs on a dovetail way system.

What I see myself is that there is after all some time needed to catch up on the backlash, or the following error.
When the slides go really slow there is virtually no following error, as the speed increases there is more and more.
It is not realistic to expect no following error at full speed. But at low speed one should expect really low error.
Set Backlash=0 in the ini file, let the feedback handle that instead.