5i25 + 7i85s +....

Phew phew....

Problem solved, if you tell my my idea is ok... So:

I disabled FERROR.
As i understand: MIN_FERROR calculates the Maximum Error in combination with TRAJ - ACCELERATION.
Goog or bad Idea?

To your Questions:

Is LinuxCNC using the encoder or a step-generator to drive joint.N.motor-pos-fb ?

Atm i use the Encoder to gain experiance, but not joint.Nmotor-pos-fb. I use:

Is your stepgen_maxaccel bigger than the joint accel limit? (it needs to be, to give a bit of headroom)

MAX_ACCELERATION = 800
STEPGEN_MAXACCEL = 1000

So, not just a more sensible PID tuning?

Disabling ferror is not good idea in general (its really a safety feature)

You are now using encoder feedback to close the position loop
this requires very different tuning than closing the local steogen loop

At the minimum P must be lower since there will now be a significant delay
between setting a stepgen velocity and the resultant position change

I would suggest that you plot the ferror vs velocity on halscope to be able to
tune the control loop

Yeah, i was just reading wrong.

MIN-FERROR calculates FERROR with TRAJ - ACCELERATION.

And i found something more. If i give more "load" on the Servo it doesnt happen that often.
So maybe overshoot?

And i found something more. If i give more "load" on the Servo it doesnt happen that often.
So maybe overshoot?

This isn't something you need to speculate about, but rather something you can measure (with HALScope)

But yes if you still have your 1000 P term from the local stepgen tuning, its pretty much guaranteed to oscillate/overshoot

I see, youre worryed about my health.

All fine, im still in test mode, with the servo mounted on a Table
Even on the Machine my endswitches are direktly hooked to the drives main power plus brake resistor.

Youre right, its Velocity, not Acceleration, to work with FERROR im sorry, was just reading to quick!

Ok, before i waste more of your time, ill better dive deeper into Hallscope and tuning.
Otherwise ill go from one problem into the next one...
See:... G00 X1000 leads to 999.970. Typing x1000 againt, it wont move the last 0.030.... (stepgen also as encoder)

My Version is 2.7.0
Kinematics should be standard XYZ trivkins. Well i didnt set anything special atleast.

Update:
COMMANDED also as ACTUAL position moves from 0 to 999.97x and 1000 to 0.02x

If it also has something to do with PID i guess ill figure it out.

[edit]
Im just stupid... -> unconected pid.x.feedback accidently. All fine now.

I understand now, why the software PID is needed.

The steps at a MESA setup are not precalculatet from position to position like in a stepper configuration.
Its a totally different way of movement.

I dont know, why i didnt realized that before.

Actually I'm pretty sure all of LinuxCNCs step generators work in a similar fashion, its just that the feedback mechanism
is hidden in the driver when you use position mode.

Thats what i see when i track the signals on my USB Oscylloscope.

It would be the same, if Axis sends Coordinates and MESA calculates the Step/Dir sequence.

The MESA card isnt able to know how many Steps are needed.
Axis just sends velocity and acceleration informations and recalculates it (PID) by the Feedback from the MESA card.

In Steppermode there is a perfeckt stable set of Step/Dir signals for a calculated move.
20 repeats and 20 times the same.

Im in a hard decision now... I compared my Servos in
- steppermode on LPT with 8.000steps/rev.
- and Servomode on MESA

-> Steppermode is way more accurat. And the Hardware PID inside the Drive does the rest.
-> Servomode gives me about 10 times more Speed.

Its like: digital is digital ... analog is analog.