Need to discard the first probe - help me find the root cause

Developers
22

I do not have this issue on either my heavily modified CR-10S Pro or my 300mm XY 350mm Z spec Voron 2.4.

Initially I had severe issues with first layer squish repeatability on the Voron and I conducted a small study to investigate it: Z Homing Accuracy & Repeatability. It is also worth noting that I use the “stock” Omron inductive probe (genuine non-discounted unit) that exhibits similar lever of repeatability at a fixed chamber temperature.

The Voron probing is an order of magnitude more repeatable in Z axis than my CR-10S Pro that uses genuine BL-Touch, 0.9° Moons’ steppers, LDO 4mm lead leadscrews with radial anti-backlash nuts, and is as precisely tuned and adjusted as possible within the constraints of its design.

1 Like
23

On a Voron 2.4 here, so I have been trying to track down the cause of this since day 1. My solution so far has been to increase samples to 5, and use an average instead of median in the calculations. It slows things down, but its more accurate. I would love to see an option to ignore the first sample.

24

TBO increasing samples to 5, but staying with median is probably more effective. As median will throw out the high and the low and then average the middle 3. This should help reduce the effect of the first probe issue.
There are other ways to decrease the effect as well by adding dwell time between xy moves and first probe. I have discovered that if I use [smart_effector] instead of [probe] it allows adding two additional parameters: probe acceleration and recovery_time that seem to help as well…

probe_accel: 10
recovery_time: 0.4
#   A delay between the travel moves and the probing moves in seconds.

You can also run two different probe accuracy tests that can gleen valuable information.

this one tests probe_accuracy in one spot, with or without temps (heat soaking)

e.g. you can see pi and controller temperatures on mine
Screenshot 2023-04-08 at 14 46 57

and this one that tests multiple corners and center

probe_accuracy_test_suite.py is a collection of tests to help checking probe accuracy, precision and drift under different conditions

edit: just sharing some useful info till the “ignore first probe” can be addressed.

Here is a simple test, sloooowd down. :wink:

SET_SMART_EFFECTOR accel=1 recovery_time=0.4

probe_accuracy samples=10 probe_speed=1 sample_retract_dist=1 lift_speed=1

13:26:27 
// SmartEffector:
// probing accelartion: 1.000
// probe recovery time: 0.4
13:26:27 
// PROBE_ACCURACY at X:208.000 Y:205.000 Z:1.306 (samples=10 retract=1.000 speed=1.0 lift_speed=1.0)
13:26:29 
// probe at 208.000,205.000 is z=0.307031
13:26:32 
// probe at 208.000,205.000 is z=0.306250
13:26:35 
// probe at 208.000,205.000 is z=0.306250
13:26:38 
// probe at 208.000,205.000 is z=0.306250
13:26:40 
// probe at 208.000,205.000 is z=0.306250
13:26:43 
// probe at 208.000,205.000 is z=0.307031
13:26:46 
// probe at 208.000,205.000 is z=0.307031
13:26:49 
// probe at 208.000,205.000 is z=0.305859
13:26:52 
// probe at 208.000,205.000 is z=0.307031
13:26:55 
// probe at 208.000,205.000 is z=0.306250
13:26:55
// probe accuracy results: maximum 0.307031, minimum 0.305859, range 0.001172, average 0.306523, median 0.306250, standard deviation 0.000430
2 Likes
25

I too have this problem on a Voron v2.4. The printer has had all manor of probes (Omron-Chinese, Omron real, Generic Inductor, Klicky, Euclid, and Now Tap. I can confidently say the phenomenon is not related to the type or style of probe. Since there have been three different toolheads on this printer, it is likely not caused by the toolhead. I will be following this closely as I have seen others do a quad-gantry level in one pass (which I have never achieved.)

I have pondered whether the belted gearing of the z-axis has some bearing on the problem. In that vein I was thinking of capturing a dozen or so QGL outputs to quantify the distance differences for each corner and seeing if that matches z-motor steps/microsteps or something similar. This might point to motor currents used, loose Z-belts, or even loose motor belts. -nevermind- The original PR goes through this in great detail.

26

First time poster here. I had a pretty stock Voron 2.4R2 and never had an issue with an inductive probe - was great every single time. I recently upgraded to umbilical, Tap and PEI bed and then started getting very regular probing/QGL errors.
If I do a probe accuracy test (100 samples) the standard deviation is 0.00811, but interestingly the first 3 probes are -1.38875, -1.32375, -1.320. All subsequent readings are between -1.31 and -1.32. Ignoring just the first reading alone, the reading are very consistent indicates that the issue isn’t mechanical.
When I turn the spring steel sheet over and probe on the metal side, the probing is consistent right from the start. Both drift slightly over time, but the std dev is very good.
My steel sheet is 0.5mm and the PEI is an additional 0.5mm.
I believe my issue has to do with the “give” in the PEI layer on my bed, so the ability to ignore the first x readings would be invaluable.

Voron TAP Probing
Voron TAP Probing790×661 51.1 KB

27

Does the PEI flexplate have the same dimensions like the non-PEI?
Of course the thickness of the PEI layer might be compensated but the results should be equal in terms of precision.

What I had that made me drive crazy was that my inductive probe was not fully above the PEI steelplate on the left side and measured nonsense. After correcting the offsets all was fine.

28

Well, that would be mechanical.

29

Technically yes, but not in the way most people are talking about it here. E.g. belt tension, frame stiffness etc.

30

@LifeOfBrian It’s exactly the same plate, just flipped over. I’m also using TAP and I probe well inside the corners, so I don’t believe that is affecting it.

31

Sorry to resurrect this thread but I am experiencing this as well on my newly converted Enderwire. I am using the same CRTouch and PEI bed I had when it was an Ender 3V2. The two biggest changes are marlin → klipper and lead screw → Core XZ.

My testing revealed it to be worse if the Z travel before homing is a larger distance. I think it might have something to do with the Keyback that holds up the gantry stretching/binding a little. That’s why the first couple of measurements are off. They are also consistently a smaller Z distance then subsequent measurements leading to the binding theory.

Anyone know if you can operate a Switchwire without the Keyback so I can test it? I’m afraid the head will just crash if I remove it.

32

Could you create a new “General Discussion” thread with all the requested information provided?

It’s not helpful piggybacking on a thread that’s more than a year old.

1 Like