I’m rerunning the job that demonstrated the issue before just to see if removing the film helped. I already opened a ticket. It does seem like a pretty extreme skew according to the vision encoder, but when I do a factory reset it prints fine. The only reason I got the vision encoder is the dimensional accuracy is a little off by default. But it prints to the edge perfectly.
I’m thinking the camera might be wonky or maybe the bed is crooked. Causing the vision encoder calibration to be off.
No way to know how much deviation it was unless Bambu provides the numbers or you did a pre and post print calibration test.
This is why they put the note “The deviation in the grid diagram is magnified and does not represent the actual scale”.
To give you an example: This is from my H2C’s result with a correction of up to .02 after doing a calilantern before and after.
If you haven’t yet, you may want to calibrate filament shrinkage too. That’ll get you a lot closer.
Let me run a calibration tomorrow so I can share my image. The deviation you see is relative to non compensated points.
In the older firmware version of the H2D you got a maximum and average value of your compensation and no image.
I used that value and extrapolated that.
Still, the exaggerated scale of the deviation might also vary and therefore my educated guess might be off also.
But the fact that it hits the end stops after calibration means his deviation is quite large.
That’s the part I apparently read right over and didn’t notice in the OP’s original post. Yeah, that’s a problem in itself beyond the vision encoder.
Well the cover wasn’t the issue for me. After rerunning the calibration with the cover removed I got the layer shift again. Doing yet another factory reset right now. 
It might be interesting to print the califlower before and after the Vision encoder run to measure the skew before and after.
If you want a simpler model, you might use the calibration object from klipper.
I assume the skew is eliminated after the vision encoder.
But unfortunately you then run into the end limits.
I guess the end limits are software limits, and the slight shift to correct itself is repeated because the software limits are shifting with the corrective translation.
I don’t think there is a skew. Everything I print looks fine. The only reason I even bought the vision encoder is because I printed a magnet, with a press fit hole, and couldn’t get the magnet in. Same model on my A1 worked perfectly. Then I saw that Uncle Jessie video where he had issues with that nerf gun and the vision encoder fixed it, so I figured it might fix this minor dimensional inaccuracy for me. But instead it exposed some issue.
My best guess is the camera is messed up. I bought the blade cutting kit a while back and had alignment issues with it too. I assumed it was because the software sucks and I was doing something wrong so I got frustrated and gave up, planning to figure it out in the future when there was a software update. But if my camera is messed up then it explains both problems.
A lot of people think the Vision Encoder solves issues on small dimensions, but that is definitely not the case.
It measures a larger grid and uses the deviations to compensate for the positioning of the motion system.
Even uncalibrated, a hole of 6mm diameter, won’t be 0.1mm off, because on the entire 300mm, that would mean 5mm deviation.
It only calibrates the motion system, but things like flow calibration still need to be done.
And then particularly the small holes. They are by default always smaller because of the way FDM works.
They seem to have put some nice hidden offsets in the slicer, but that doesn’t seem to work for filaments other than PLA or PETG.
The holes on the H2 Series seem to have more deviations than other Bambu printers, but that is not related to the inaccuracy of the motion.
This still needs some good optimization in the slicer if you ask me.
I always model my 6mm magnet holes between 6.2 and 6.3.
When holes are getting smallen, the offset needs to get even bigger.
I am still curious if the skew model differs before (after factory reset) and after calibration with the Vision Encoder.
Mine (H2D) is definitely not skewed after calibration.
I ran that calibration thing. It says my skew is 0.12 degrees. And I’m using a pair of cheap chinese calipers so I’m not even sure how accurate they are.
I definitely don’t think that’s enough for the vision encoder to skew as much as it showed in the little diagram or enough to skip a step when corrected.
I also used a square to test the edges of the build plate to the machine and they’re square, so I don’t think the vision encoder plate is crooked.
Which means either the vision encoder plate itself is defective, which it doesn’t look like it is to my eye, or the camera is messed up.
My guess based on the results is the camera. Like I said I had a similar issue when using the blade attachment, which also uses the bottom camera. I assumed I was doing something wrong, because the software sucks and is super confusing to use, but if the camera is messed up then it would explain that issue as well.
Maybe it’s not pointing straight down. If it was cocked just a little then that would explain the perfect skew from the vision encoder.
Maybe check that the screws that hold on the toolhead camera are tight. If they are loose, I can see that causing all kinds of issues.
Also check that the calibration sticker at the back of the heatbed looks okay and is clean. Might as well clean the toolhead camera too. HMS_0C00-0100-0001-0004: Toolhead Camera lens seems to be dirty. Please clean the lens. | Bambu Lab Wiki
Camera is clean and doesn’t feel loose, that calibration sticker in the back looks ok to me. If the camera is wonky then it’s gotta be a factory defect, not just loose or wobbly.
Bambu support thinks the vision encoder plate is bad so they’re sending me a new one.
I think it’s still probably something to do with your camera like you think, but at least it’s a start 
I agree, but at least I’ll know for sure if it happens again with a different plate.
Now they’re back peddling. They want logs and for me to try some belt tension adjustment, which seems unlikely to be the problem since the machine prints fine once I do the factory reset.
That looks different than my issue. Mine slowly drifts. Every layer is shifted a little to the right until finally it’s printing in mid air and fails. That looks like an actual layer shift.
One was too short and one was too long on my H2D and both were too long on my H2C. It seems the acceptable length range they’re using for those tubes is too wide. They need to tighten that tolerance.
Belt tension can affect skew in a core-x-y so it’s reasonable to ask you to do it.
Did it help?
