Skewed print any ideas?

Its just represents the relative positions of the endstops on the towers. I haven't positioned them with extreme precision, what would be the point, when the firmware automatically offsets them depending on probing the bed?

The bed is 90 degrees to the towers, so I could manually move the endstops so that they showed 0,0,0 after auto-calibration, but that wouldn't physically change anything. Whilst it might be a problem if they were a huge distance apart surely their positions are arbitrary. I can see that before the auto calibration and least squares calibration came along positioning the endstops precisely made for a good starting point when manually calibrating.

Also I haven't changed anything endstop related, I manually levelled the bed a bit by shimming it on the Y tower bed clamp, but that was after the slanting print problem started, and it didn't make any difference.So far nothing has made any difference, I am going to print a stiffer effector to rule out the wobbly hot end theory.

Even if the bed were very non-level why would the object begin to print straight after z=50mm? When I have set grid-compensation fade out to 10mm not 50mm? And it prints the same with grid off or on, as my bed is flat.

Edited 1 time(s). Last edit at 12/19/2016 03:16PM by DjDemonD.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

If you have a wobbly hot end then the sideways force of the Bowden tube will affect its position. The size and direction of that force depend on the effector position, including its height.

The hot end on my delta recently became wobbly. I hate groove mounting. If I had access to a machine shop, or even a decent drill press, I would drill and tap 3 holes in the top of the heatsink for attaching the hot end to the effector.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

An easy way to stop a groove mount from wobbling, is to drill a 2.5mm hole through the clamp, tap it with an M3 thread, and then put an M3 grubscrew or a longer normal screw into that thread. You can then tighten down the screw and it will jam the hotend to where it needs to be. This is how the RobotDigg effector works, and it seems to work well.

An easier way is to just shim the clamp with a small amount of thin aluminium like a small piece of beer can. This gives you more adjustment if the clamp is bottoming out. This is discussed in "Zen and the art of Motorcycle Maintenance"

I'm keen to hear how DJ's modifications are going.

Edited 1 time(s). Last edit at 12/19/2016 08:56PM by nebbian.

Yeah my new micro delta has a deltaprintr mini hot end and it attaches with m3 screws tapped into the heatsink. It fits really well compared to groove mount, but the bowden adaptor fouls on the screw heads.

Beer cans make an excellent shim/joiner - I have a motorbike which has both exhausts attached with beer cans and jubilee clips.

Unfortunately the micro delta seems to have a slight skipping steps issue right now so the new effector won't be made until this evening. On the large kossel the hot end is slightly less solid than I'd like, but if I move the umbilical around its doesn't wobble at the nozzle. So who knows?

Edited 3 time(s). Last edit at 12/20/2016 03:53AM by DjDemonD.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

I don't see the conclusion of analysing the gcode. But I assume the gcode is for a straight print.

So the slant starts at a certain height and ends at another? Is this fixed? Meaning is it always the same height where it starts, regardless of where on the bed you are printing or the size of the object your are printing? You ruled out firmware, I think you have ruled out slicer, so it must be mechanical. I know, I am stating the obvious, just thinking aloud.

If you print a square column, which way is the slant pointing, in reference to the bed or rather, columns? Again, is it always in the same direction regardless of the position on the bed? Is it straight towards a column, or away from one?

Gcode is for a straight object.

The lean is the same irrespective of bed position.
It is towards the -x direction not away/towards a tower.
If it's mechanical it's not the frame which is solid, hasn't moved and is very precise.

Slant starts at z=0 and finishes at z=50.

A wobbly hot end being pulled by the cabling is a possibility.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

You've measured steps per mm per carriage right? Have you tried marking an area on the bed, doing a dry print and jogging back to the spot? I wonder if you're losing steps over the course of a run that isn't showing under simple carriage steps/mm runs. My guess is its on the tower opposite the -X direction. You could change your tool and stepper declarations around to use your E2 stepper to drive that (or other towers) and rule out a fritzy stepper driver. If we're testing motors or steppers my thought is this: (keeping in mind I'm new to duets and deltas lol) If you rotate the XYZ plugs on the duet and do the same print I'd be curious to see what happens. If you're positive all of your mechanics are perfect (no belt stretching even) and the skew stays the same way I'd guess that stepper driver is shot. If the skew moves with the rotation, I'd guess the motor is going.

Edited 1 time(s). Last edit at 12/20/2016 08:33PM by clearlynotstefan.

If it is the stepper it is strange that it stops at 50mm.
So basically you will have to print everything with a 50mm tall brim? (sorry)

Does the -X direction slant confirm cabling as the fault? Is the cabling bending in that direction? If it was the wobbly hot end, you would expect it to fade not simply stop at 50.
Also, what happens if you go higher? If it is the wobbly hot end, it should return at a certain height, shouldn't it? Hmm, no that depends on the cable system and how it is mounted.

Steps/mm - check.Belts - check.

I made a new effector, and reinstalled the hot end. No difference. So we can now rule out wobbly hot end and cabling pulling on the hot end, it is now extremely rigid and my effector is now looking good in red abs.

I tried a print with the fans off, I was running them at 20% before. No difference. I wondered if airflow from one side was warping the print, it isn't.

The lean really isn't towards a tower which is extremely puzzling.

So I rotated the motors (and endstops) and printed the same tower again. Whilst the tower is now rotated, the lean is exactly the same towards left of the bed.
So its not software, its nothing to do with the coordinate system, its not motors or drivers, or belts, or pulleys.



It suggests my frame is bent, but only for 50mm as it starts printing straight after this height ???? I've just check the mgn-12h rails, one was slightly notchy around z=0 so I've lubricated it and its running a bit smoother, will try another print identical to the one above now.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

No change at all.

Would having a slightly tilted effector but consistently tilted irrespective of bed position make any difference? According to my little spirit level the effector has a slight tilt but this doesn't vary as you move around the bed, so I can't see how that can explain the effect we're seeing here, especially as the problem seems to disappear after 50mm z height.

I might try rotating the arms maybe that will make a difference.

Edited 2 time(s). Last edit at 12/21/2016 05:32AM by DjDemonD.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

Do you have another Board that you can use to rule our firmware/electronic for good ? maybe your mini delta electronic ... Its a weird problem since its not toward a tower. also you switch slicer alrealy just to test right ?

Hi it's something I could try. I don't think it is the board it's seems too subtle to be the only symptom, but I am running out of ideas. It's a duet wifi board, so switching over to a ramps/mega would be a headache.

I've loaded the gcode and verified it is for a straight object. So I think we can rule out slic3r as the cause.

I could work around it by using RRF axis compensation up to 50mm but I'd need a post processor to add the gcode to turn this off after 50mm. Very unsatisfying but it would get the machine back to functional.

Edited 1 time(s). Last edit at 12/21/2016 11:34AM by DjDemonD.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

When viewed from above does the part skew dead straight toward that edge, does it wobble slightly on x/y too? Is it a consistent lean per layer? Basically the more consistently refined it is the less likely it is mechanical in my opinion. Also I'm not positive that test technically ruled out motor as you rotated the plugs but obviously not the motor, if that motor was shot itself, it's still hooked to the same belt and arms that would push the print that way. You did rule out the driver though as it'd be driving a different motor and the print skew would follow.

Sure I take your point on diagnosing the motor, swapping motors around would be a real test, but whilst I am getting more desperate for ideas I just cant buy it being a faulty motor. If it will position reproducibly at high speed over and over it really can't be a bad motor. But since I'm running out of plausible theories its time to Sherlock it and start looking at the improbable.

Thanks for your input. Perhaps I'm just hoping to find the fault before tearing the whole machine down and rebuilding it. Which is looking increasingly likely.

To answer your specific questions, the part leans directly towards left of the bed (-x) and consistently tilts until z=50mm then prints straight. It is perfectly straight in y, and correct within 0.5% in height.

Edited 1 time(s). Last edit at 12/21/2016 01:57PM by DjDemonD.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

I'm still thinking software or straight up tower lean. Anything like motors or steppers would be more erratic I think.

Wait its going to -X exactly? Just to verify your axes are like mine



.............................Ztower


<---skew exacty this way?



Xtower-----------------------------------------Ytower

Edited 1 time(s). Last edit at 12/21/2016 03:45PM by clearlynotstefan.

If so I suspect software. In my mind hardware problems should usually manifest themselves as errors at angles directly toward or away from towers. To move perfectly straight along X requires coordinated mistakes between multiple carriages, skipped steps etc on one motor wouldn't line up directly towards -X in my photo above, I THINK. Keeping in mind I'm just thinking this through theoretically and am not that experienced with these issues.

No we agree on it probably not being hardware for the same reasons, plus nothing has changed physically that I can find. Yes it might still be something more diffuclt/time consuming to diagnose sure.

Your diagram is correct it skews directly towards -x/left. This is why I thought Nebbian's loose hotend/cable tension theory was most likely, my hot end was not very firm and my cabling is to the left of the effector. But I've fixed these things and literally no measurable change at all.

I might go through the software config again with an even finer toothed comb. Maybe even check the calibration manually.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

Can you post the gcode file?

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

LargeKossel_axesorthogonalitycompensationobject.gcode

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

Thanks, I'll try printing it during the weekend. Meanwhile, two more suggestions:

1. Is the bed clipped to the bed support? If so, could it be moving?

2. Change your slicer end gcode if necessary to not home the printer, just raise the head a little. When the print has finished, jog the head until the nozzle is lined up with the edge of the top of the print and read off the X coordinate. Then home the printer and repeat. How do the two X coordinates compare? Do they match the position of the edge of the print given in the gcode?

Edited 1 time(s). Last edit at 12/21/2016 06:14PM by dc42.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

I just saw you're using 7 factor calibration. Why? You know how long your arms are I presume?

Try running 4 factor calibration instead and see how that turns out.

David,
Thanks its appreciated. I will try the coordinate test you suggested that's a really good idea.

Clearlynotstefan
Yeah I am thinking of switching to 6 factor as I'd like to optimise around rod length to improve my dimensional accuracy in x and y. Often prints come out at around 98%-99% scale in x and y, I sometimes scale them in slic3r. It would be nice to determine the right rod length for 100% scaled parts natively so to speak.

However using 7 factor calibration is not something that I've introduced recently and the rod length derived hasn't varied much over time as I've been writing it down in order to update my M665 and M666 in config.g, now that m500 is here this won't happen anymore. Not complaining about that it's much easier now.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions

Consistent doesn't mean accurate though. My understanding is your bed.g needs to probe points far enough that the rods are nearly horizontal to get accurate readings for rod length. I'd measure them out on a caliper and switch to 6 factor. I get that you didn't change that recently but perhaps the updated firmware treats it differently? We'll hear from dc. To be clear, I'm not proposing this as a solution to your exact problem, just that generally rod length calculations aren't usually as good as measuring so might as well get that sorted.

Wow this is a perplexing problem. I'm dealing with a different perplexing problem at the moment, so I feel your pain. Just another set of eyeballs and observations:

* If you switched motor connections and endstops, and the lean is still towards the same direction, then it is NOT software, or anything in the controller board. It has to be hardware of some description.
* It can't be effector tilt due to the normal causes (balls on carriage not level, balls on carriage not at 90 degrees to tower, different distances between balls), because this would stay the same at different Z heights.
* The fact that it is so reproducible is puzzling to me.
* Something has to be changing with relation to Z, as the print progresses. Note that it can't be firmware or controller, for the reasons mentioned before.
* Are your carriages nice and tight? If you twist your effector, is it rock solid, or is there slop?

This printer uses linear rails, right? So that means it's unlikely to be a bent tower (those rails are STIFF).

If the Z tower's rail is slightly twisted towards the bottom of the bed, then it will introduce this sort of error -- the effector will tilt with relation to Z.


Here's what I'd do:
1) Attach a laser pointer to your effector, aiming in the +X direction.
2) Move the nozzle down to 0,0,0 and mark on the wall (a long way away from the printer) where the laser pointer hits. Some paper taped to the wall will help.
3) Raise the nozzle in 2mm increments, to a total of 20mm, marking the point on the wall where the laser pointer hits each time.
4) Measure the distance between the first and last points and see if this is 20mm.

The further away from the wall your printer is, the better you'll be able to see any inaccuracy.


I used the laser pointer trick last night when trying to see if there was any twist in my frame. It's a good test.

Edited 1 time(s). Last edit at 12/22/2016 01:59AM by nebbian.

The fact that it is moving in exactly a -X direction, aka not towards or away from a tower, and is so consistent and reproducible isn't enough to overcome the doubt raised by swapping motors tho? I can't think of a single non-digital error that would move all three carriages just right to skew a print along the x axis reproducibly. You also raise a solid point about swapping motors not changing the skew being very bad at implicating a digital problem. To clarify, my suspicion firmware wise isn't bad gcode or anything, its that the probing results are off in a reproducible way, that tells the firmware to create that slant. For instance if effector tilt caused a bad probe point over on that corner, resulting in a bad say X offset, rotating plugs would just generate a bad Y offset instead and the results would look the same. I just can't fathom a mechanical issue that so consistently and reproducibly causes all 3 carriages to perfectly move the print along the -X axis without firmware intervening somewhere, except maybe leaning towers which are pretty unlikely here. My only guess is its technically a tiny mechanical error, that is consistently messing with the calibration to generate this problem. Ultimately dropping the 7 factor probing is the way to go anyway. Your arms aren't changing size and the reading under most peoples probing setup is pretty bad anyway.

LOVE your laser pointer trick, great idea!

You are not using the grid leveling system, correct? You had switched back to previous version of RRF if I recall correctly.
As I was thinking if it was an error in the grid system, that tapered off over Z that would explain it.

Can you print a large diameter cylinder wall? As large as will fit. You will have to go above the 50mm just to be sure. Would be interesting to see if the slant is consistent around the whole perimeter.

I was thinking, did you try to print on a cold bed? Or, as cold as needed just to allow your print to stick? This would rule out way-off temp. measurements leading to way overheated bed, leading to the print being softer than normal while printing close to the bed, say up to 50mm
Just an idea...

I also agree with changing the 7-factor authentication to 6 or even 4. No reason to play with your rod length with each calibration run.

Just for fun, have you tried actually running the new auto bed level feature? I'd love to see the height map it generates, it'd be useful for us to know what the duet is seeing when it probe the bed. The newest version will generate a visual display you can post here, peep DCs post in the duet forum!

Nebbian - I agree there must be something physically affecting this first 50mm. The linear rails are still fixed at three points solidly to the frame (2040 verticals) which measures up as square and true. The linear rail carriages themselves have no play, my carriages attached to them have no play either. The effector is fairly firm, there's a small amount of compliance in the system as I am using printed carriages and effector and 360mm arms in 6mm carbon (Haydn's setup), but it is minimal, far less than say my recently sold Kossel Mini with traxxas joints, and its consistent at any given z-height (obviously). I've replaced the effector with a new one.

Clearlynotstephan - I will set an arm length to whatever gives me 100% scaled prints in x and y then calibrate around that.

Lykle - I have tried with grid levelling on and off and firmware 1.16, 1.17dev8, 1.17rc1 exactly the same error every time, I have tried with grid compensation not tapering, and tapering off at 10mm - no change exactly the same error. To be fair my bed is very flat (I will post a grid height map when I am at home tonight - it barely corrects at all) - I think grid levelling is not the cause. I now have hundreds of sloping cylinders lying around everywhere, my 3 year old son thinks they're great for building towers with (they're not obviously).

I can try to print a really big cylinder 300mm! I have kind of ruled out bed position by printing a load of small cylinders at different points on the bed and they all lean the same way, by the same amount. So I will put that idea under the "try if desperate" list - I am getting closer to these options.

Mandrav - I mainly use ABS so cold bed isnt an option, but for the sake of loading some PLA I might try a few runs to see if it makes a difference. So thanks for the suggestion. That being said I've printed a lot of objects onto this bed at 120 deg C (printbite) without problems and whilst I notice slightly collapsed lower walls in the prints, which Nebbian picked up on first, this effect is even and doesn't skew the objects.

---

Simon Khoury

Co-founder of [www.precisionpiezo.co.uk] Accurate, repeatable, versatile Z-Probes
Published:Inventions