Page MenuHomeAleph Objects Inc

Dimensional Accuracy - Calibration Print
Closed, ResolvedPublic

Description

While trying to nail down the calibration process and gcodes (T4129) I am unable to get accurate dimensions, especially on circles.
I have tried:
Outer before inner walls
lower temps
higher temps
slowing it down
backlash comp on
backlash comp off
backlash comp 50%
multiple smoothing distance values
Belts are tight
no binding occurs

None of the above made a difference on the circles and this occurs on every Quiver machine produced.

Event Timeline

logan created this task.Feb 27 2019, 10:34 AM
logan triaged this task as Unbreak Now! priority.
logan created this object with edit policy "LulzBot Hardware Products (Project)".
oliver added a subscriber: oliver.EditedMar 5 2019, 11:32 AM

@logan From the data that I gathered, I noticed that setting the backlash compensation on the machine to 100% seemed to go along way towards making the part round but would still come out undersized.

In the past I have seen this issue due to too much drag in one of the axis causing micro skips in the motors. Though this doesn't seem to be the case here.

@west tested the calibration print with Redgum which mechanically has the same x/y axes as quiver but uses RAMBO instead of Archim. The print came out sized properly, which points to an issue with the Archim

karrad added a subscriber: karrad.Mar 5 2019, 1:07 PM

@tutley What version of FW did you flash to redgum?

karrad added a comment.Mar 5 2019, 1:09 PM

@west The SE quiver firmware?

karrad added a comment.Mar 5 2019, 1:21 PM

@west Ahh is this one in marlin build yet? Not seeing it in software http://devel.lulzbot.com/software/Marlin/2.0.0.96/

west added a subscriber: marcio.EditedMar 5 2019, 1:23 PM

because its not, this is preliminary fw that @marcio sent me so I could compare the dimension test

oliver added a subscriber: Steven.Mar 5 2019, 2:59 PM

@Steven Updated the spreadsheet with the last printer that is down in MER except for the OHAI printer.

karrad added a comment.Mar 5 2019, 3:29 PM

On beta 0006 I am getting 9.6 in X direction on the cylinder, and 9.85 in the Y direction. I will run some tests with increased motor current to see if it helps

karrad added a comment.Mar 5 2019, 4:04 PM

Bumping X and Y from 975ma to 1050ma showed no/minor improvement.

9.65 in X and 9.83 in Y

marcio added a comment.Mar 5 2019, 4:09 PM

Could you test with .97 firmware? I made one small modification to the backlash compensation code that may affect dimensional accuracy.

oliver added a comment.Mar 6 2019, 7:07 AM

@karrad Are you using 100% backlash correction in the settings? Also would it be possible to share your belt tensions? How does the drag on the x and y axis feel?

oliver added a comment.EditedMar 6 2019, 7:07 AM

@karrad Also what profile or gcode are you using? Are you using the gcode from R&D?

If increasing the current of the motor is helping it maintain dimensional accuracy that tells me that the current setting right now is not providing enough torque to overcome the forces on the axis. Your getting slips in the motor itself.

karrad added a comment.Mar 6 2019, 7:47 AM

@oliver I am using the sliced gcode in the calibration folder: http://devel.lulzbot.com/TAZ/quiver/sample_prints/calibration/quiver_vernier_PLA.gcode if there is a different one we should be using, can you point me to the ticket/branch it is on right now?

@marcio Will give another test on .97 with stock current

logan added a subscriber: anolen.Mar 6 2019, 8:47 AM

@karrad See T4129, @anolen had posted there some gcodes she was using with better results dimensionally for the circle. Those prints did affect the readability of the vernier however so perhaps not a good all around calibration gcode; I don't think there is one currently.

logan added a comment.Mar 6 2019, 8:50 AM

To clarify; the gcodes posted in T4129 should be run with backlash compensation set to 100% with 0 smoothing from the printer's LCD menu

karrad added a comment.Mar 6 2019, 9:16 AM

Using gcode in current taz-quiver directory with .97 FW i got these results:

X = 9.69
Y = 9.87

I then ran an automatic calibration and received these results:

X = 9.6
Y = 10.03

oliver added a comment.Mar 6 2019, 9:27 AM

@karrad And is that with turning smoothing off and 100% correction in the back lash settings of the printer?

karrad added a comment.Mar 6 2019, 9:29 AM

@oliver At the moment, I am just using pre-sliced Gcode, stock firmware, and stock automatic calibration. Currently running tests on most recent gcode posted to T4129

logan added a comment.Mar 6 2019, 10:29 AM

results from 2.0.0.97 on QB7:
factory defaults restored; z offset restored to -1mm
used backlash/offset values from auto calibration
backlash compensation 100% & smoothing @ 0
noz 1 circle X: 9.02mm
noz 1 circle Y: 8.69mm
noz 2 circle X: 9.84mm
noz 2 circle Y: 9.66mm
gcode generated from fresh local build CuraLE 3.6.5 with freshly cleared cache
polylite PLA Standard
0 bottom layers nozzle 2
0 top layers nozzle 1
infill aligned to make X in middle of cylinder (like Mini 2 QuickCircle)
backlash fading distance 0 (to use machine defaults)





tutley added a comment.EditedMar 6 2019, 10:36 AM

Using 100% correction, 0 smoothing, autocal backlash values, and 100.5 xy steps, I was able to get a print that was in spec.

Measuring 9.82-9.91

logan added a comment.Mar 6 2019, 11:15 AM

above gcode reprinted after setting X/Y steps to 100.5
results:
noz 1 circle X: 9.07mm
noz 1 circle Y: 8.72mm
noz 2 circle X: 9.86mm
noz 2 circle Y: 9.72mm

Orias added a subscriber: Orias.Mar 6 2019, 1:17 PM
oliver added a comment.Mar 6 2019, 3:15 PM

Just to see what would happen, I went took a more extreme approach and set my x/y steps to 105. You will not be able to do a calibration and it it did make my each nozzle be off but the circle came out to;

noz 1 circle X: 10.04mm
noz 1 circle Y: 10.06mm
noz 2 circle X: 10.14mm
noz 2 circle Y: 10.33mm

The max diameter on the circle for noz 1 was 10.26mm. This was with 100% back lash.

karrad renamed this task from Poor dimensional accuracy X/Y to Dimensional Accuracy - Calibration Print.Mar 6 2019, 4:01 PM
tutley added a comment.Mar 8 2019, 7:43 AM

@Steven and i looked at carriage drag on a few machines, his machine has about 18N drag in the y and ~23N in X,
i also looked at 2 of @anolen's machines. both had about 20N y and 24N in X

Looking at the collected data, S/N 0012 passed without backlash being turned on. I tracked down this machine and both X and Y were at about 23-24N

Based on this it seems that carriage drag is likely not the cause

We wanted to get some data for different materials and a different wall order to determine if Horizontal Expansion and/or Differential Shrinkage could be factors here.

Here are some tests to try with your machine. Please print, measure and post your results for comparison if you are able to print 1 or more sets of these.

PLA (Polymaker)

PLA (Verbatim)

ABS (Chroma)

PETG (IC3D)

logan added a comment.Apr 4 2019, 8:37 AM

On T5928 @tutley said:

backlash compensation is turned off by default. Hence the M425 S0. This is enabled through Cura settings just like with Mini2.

Is there supposed to be an M425 S0 in the start gcode? There isn't one in the Quiver start gcode most recently provided on T5928

logan added a subscriber: paulette.Apr 5 2019, 12:07 PM

I provided @paulette with a vernier gcode with backlash fading distance set to 99999mm
We have 2 passing prints from QB22, the first one measuring 9.84 to 10.13mm. The second print measures 9.81 to 10.03mm
@Steven @west @tutley Would you like to take a closer look at this machine? Perhaps take some measurements from it to help determine why this machine passes while others do not?

tutley added a comment.Apr 5 2019, 2:19 PM

@logan yes there should be the M425 S0 in the start gcode, you are correct, Im not seeing it in there either.

paulette added a comment.EditedApr 8 2019, 2:06 PM

Ran 7 different quiver machines so far with the new vernier with backlash gcode. Only 2 have passed specs of 9.8-10.2 on the circle (0022 and 0029), the rest are not passing. there are 3 more that still need testing.
Tagging @kent

paulette added a subscriber: kent.Apr 8 2019, 2:13 PM
kent claimed this task.Apr 9 2019, 11:10 AM
kent added a comment.Apr 9 2019, 11:16 AM

The gcodes with the backlash fading distance set to 99999mm in cura seem to be working a lot better in terms of being able to tune in the dimensional accuracy of circles we are getting prints on the betas now that are looking a lot more circular and are coming in closer to an average of 10 rather than being in the 9.4 to 10 range.

I've got a sample set of taz 6 prints, a sample set of pre-backlash-fading-distance-set-to-99999, and a set of post-backlash-fading-distance-set-to-99999 which I am going to measure and create a control chart for. If the control chart looks like what I expect it to based on what I've been seeing, I think this issue will be resolved (at last).

@anolen for our users to see the same dimensional accuracy that we are seeing on the line, the profiles will need to have backlash fading distance set to 99999, which they are not currently afaik.

@kent @anolen @marcio @Steven Updating the backlash fading distance for all profiles should be very quick once we get consensus. Please re-assign this one to me once that decision has been made.

logan added a comment.Apr 9 2019, 11:50 AM

Is there a more efficient way to accomplish this?
Would adding M425 F1 S0 to the start gcode accomplish similar results?
Would that require hiding the backlash fading distance setting by default for Quiver?

logan added a comment.Apr 9 2019, 11:53 AM

Also, to be clear on concerns for implementing on all profiles for quiver; this has not been tested on part geometries other than the calibration print. I would advise caution in implementing a change such as this for all profiles without further testing.
It may be that we do this solely for the calibration print, and advise customers with dimensional concerns on the path towards alleviating the issue via documentation.

oliver added a comment.EditedApr 9 2019, 2:46 PM

Here is the html files for the current test prints that I have received. I was seeing an improvement of approximately 20% in fallout when backlash was enabled.




oliver added a comment.Apr 9 2019, 3:10 PM

Quick update: I had a small mistake on one of the files and forgot one


kent added a comment.Apr 11 2019, 8:43 AM

One thing to note about that data set (especially the post-backlash enabled data) is it does not include some of the passing tests, since they got shipped out with printers before the data was recorded. That said, it's still not as conclusive as I would like.

kent added a comment.Apr 17 2019, 9:46 AM

So, we have been having printers pass by using gcode that keeps backlash enabled and tuning the settings for backlash and smoothing distance through the touchscreen.

The issue that we were having when this ticket was created is solved (we can get passing calibration prints), and there is nothing in the hardware source that needs to change so I'm calling this resolved.

kent closed this task as Resolved.Apr 17 2019, 10:17 AM