Linear Encoder, PID and Backlash

I'm integrating encoders into and existing linuxCNC mill with stepper motors.
The BACKLASH setting (0.4mm) has been carefully tuned to give reliable positioning without encoder use.

Now I'm modifying the system to use a linear encoder and PID.
The linear encoder is measuring table travel.
Everything is functioning as expected and I'm in the process of tuning the PID controller.
My questions are prompted by this observation:
With a high P-term, on every direction change I get a big oscillation on PID output which takes a fews of cycles to settle.
This oscillation only occurs on change of direction.
I fully understand why the PID controller produces a high output/overshoot during the backlash take-up period as the motor is speeding up but velocity feedback stays at zero.

My question relates to the interaction between PID and BACKLASH:
What part (if any) does BACKLASH play when using a PID controller (with velocity command output)?
Should the BACKLASH setting be set to 0 when using a PID controller?
Is it possible to "limit" the pid output during the backlash take-up period?
What is considered "best practice" in this scenario?

Thanks for looking at my question.

Thanks tommylight, I have everything configured and it is working but I'm not 100% satisfied, even with BACKLASH=0.

Without direction change the position feedback follows the velocity command very closely.
On a direction change the backlash causes the PID to do this:

The purple trace (velocity feedback) clearly shows the delay between the motor starting to turn (i.e. green trace) and (after backlash) the table finally starting to move.

I realise the solution to the problem is to use ball screws (on order) but I was wondering if there is any way to overcome the pid-backlash problem?

To complete the picture, here is the same trace but without direction change:

Note: Following error trace is on high zoom level.

why is there a horrible jitter in the feedback
isent the cable shielded
and or the connectors gronded

Hi, the combination of backlash and scale is difficult. On real cnc machines there are parameters around the backlash control loop. But the first rule is: before the parameters are tuned, the mechanics must be 100% okay. And 4/10 backlash is far from a good machanic. I also have a rack as a drive with a similar backlash and scale but with servo motors. they can be resolved higher and regulate more precisely with a small PID. To do this, the rack and gear must also be 100% parallel. my mechanics are also not perfectly aligned and each axis plays differently even though the parameters are the same. But when I mill a circle it is round to 0.1mm and that's okay for me.
A smaller pitch could improve your problem. So you would have more pulses for the control loop. And there is a lot of noise on your scale feedack. Vibrations or something wrong with the cable.

bbsr_5a wrote:

why is there a horrible jitter in the feedback
isn't the cable shielded
and or the connectors grounded

You have to remember the velocity feedback is coming from a linear encoder attached to the table, not the motor. The encoder provides pulses, not an analog speed signal like a tacho. As the velocity signal is calculated from the pulses the "noise" or jitter is unlikely to come from interference. If there was any interference generating additional pulses it would also have to affect the position feedback. Here is a trace of the position signal together with the velocity:

As you can see, the position signal is perfect whereas the velocity signal is noisy, so I conclude the noise is coming from the calculation.

I suspect the noise on the velocity signal is a combination of using a 5um resolution encoder on a low quality [Chinese] ACME lead screw. At the maximum speed of 14mm/sec the encoder signal produces only 2800 pulses per second, that's one pulse every 357us.
I'm not sure how often the velocity calc is performed, but, the low resolution encoder will produce very few pulses for a fast measuring system and therefore a 1 digit variation (as is usual in all digital measurement) is of significant magnitude and shows up as "noise" on the trace.

The encoder cable is shielded all the way to the Mesa 7i85S card and the shield is grounded on the Mesa card end.

Noise on hm2 encoder velocity signals at a constant speed is more likely caused
by actual motion variation and quadrature errors ( lack of perfect 90 degree offset between A and B )
than calculation. Velocity estimation is done by a delta_counts/deltaT calculation where deltaT is not the sample
period but rather the time between count changes. This eliminates the count sample time uncertainty in
velocity estimation but is still susceptible to quadrature errors. Sometimes you can improve
the quadrature accuracy by mechanical or electrical encoder adjustments.

PCW wrote:

Noise on hm2 encoder velocity signals at a constant speed is more likely caused
by actual motion variation and quadrature errors ( lack of perfect 90 degree offset between A and B )
than calculation.

Many thanks for your authoritative answer.
My first suspicion is motion variation on the cheap Chinese made ACME screws as I have observed up to 80u variation in the lead screw pitch.
The linear encoders are no-brand cheapies, so quadrature errors are certainly possible.

The "noisy" velocity signal does not appear to be causing any problems as far as I can see so I'll leave it as is.
However, I will re-evaluate once I have fitted ball screws.

For what it's worth....
In my experience ( 9 different machines), there is no substitute for good mechanical components and near zero backlash.