Incorrect dimensions on printed objects

[thedoble]

Hi folks, I'm having some problems with dimensions on objects that I'm printing.

I noticed this first when printing a raspberry pi case.
First I tried this sleeve case: http://www.thingiverse.com/thing:604915" onclick="window.open(this.href);return false;

I found that I couldn't fit the Pi in there - I figured I'd just try a different design.
Second two-halved case: http://www.thingiverse.com/thing:665042" onclick="window.open(this.href);return false;

This had the problem of the two case halves not fitting together. Both of the cases I had printed at 0.3mm layer depth in PLA. I then tried 0.2mm layer depth and had the same problem. So I got my calipers out and started measuring! I found that my dimensions were off by between 0.15mm and 0.35mm depending on which part of the object I measured.

I did some searches on the forum and found that this can be caused by incorrect diagonal length settings. I have the U-shaped arms, and I checked my EEPROM and it is at 269mm which is apparently correct.

I downloaded this calibration circle - good news, my angles are perfect, but I found the following when measuring the markers. Each marker should be 60mm.

X: 59.76
Y: 59.93
Z: 59.97

So I started tweaking the diagonal length parameter in my EEPROM, printing a 100x2x2mm object to check after each change. I printed it in line with the X axis, as this was the one that was the furthest out. After getting exactly 100mm from my calibration object at a diagonal length of 258.350, I re-printed the above calibration circle. These were the new results:

X: 60.02 (+0.26)
Y: 60.19 (+0.26)
Z: 60.26 (+0.29)

So while I seem to have corrected the X axis, I've just moved the problem elsewhere. I set the diagonal rod length back to 269mm and did some more reading into how to correct problems with individual axes, and couldn't really find any concrete explanation. Thankfully 'repetier' commented on the thingiverse calibration circle with some valuable insight:

[img]http://i.imgur.com/CD9tIGD.jpg[/img]

The explanation is .. brief, and references 'Corr. diagonal A/B/C' when the EEPROM values are actually called 'Delta Radius A/B/C'. I know this because I found this forum post which references them, and thankfully found this handy 'map' of the A/B/C towers:

Code: Select all

/*  =========== Parameter essential for delta calibration ===================

            C, Y-Axis
            |                                   |___| CARRIAGE_HORIZONTAL_OFFSET
            |                                   |   \
            |_________ X-axis         |    \
           / \                                  |     \  DELTA_DIAGONAL_ROD
          /   \                                         \
         /     \                                         \    Carriage is at printer center!
         A      B                                        \_____/
                                                            |--| END_EFFECTOR_HORIZONTAL_OFFSET
                                                    |----| DELTA_RADIUS
                                                |-----------| PRINTER_RADIUS

    Column angles are measured from X-axis counterclockwise
    "Standard" positions: alpha_A = 210, alpha_B = 330, alpha_C = 90

So I input the values into my EEPROM and re-printed the calibration disc. Got it pretty close. I then adjusted and printed again, and got 60mm (+/-0.02mm) for all the axes. Great! I re-printed my raspberry pi case ... and had exactly the same problem the wall thickness is too high for the pieces to fit together. The overall dimensions of the object are bang on, but the inside walls are off by as much as 0.6mm.

Am I missing something here? Am I trying to correct this problem in the wrong way? Am I expecting too much of the printer?

Here's some photos of the raspi case compared to the STL file - the case was printed at 0.2mm layer depth. PLA. 0.4mm nozzle. Minimum 2 perimeters. Minimum 3 top / bottom layers. Perimeters 45mm/s. External Perimeters 33mm/s. Small perimeters 15mm/s. Infill 65mm/s. Hot end 195C, bed 60C. I've also attached the GCODE.

[img]http://i.imgur.com/OcwdGiY.jpg[/img]

[img]http://i.imgur.com/ht1AE63.jpg[/img]

[img]http://i.imgur.com/GV29B7M.jpg[/img]

[img]http://i.imgur.com/3uyiUVo.jpg[/img]

[img]http://i.imgur.com/abkZRyn.jpg[/img]

[img]http://i.imgur.com/WJLJ5mp.jpg[/img]

[img]http://i.imgur.com/wjMyq8e.jpg[/img]



Any guidance would be appreciated!
Thanks!

[mhackney]

Delta calibration is a "bulk" operation, meaning that it sets the parameters to get accurate movement of the nozzle when commanded by gcode. You want to calibrate your scaling using a largish calibration object to make sure you have reasonable resolution. Most folks use the 20mm calibration cube and that will give satisfactory results. Bigger is better though. I use a 50x50x20 cube since Z is not calibrated with arm length.

So, you've done a good job of calibrating arm length above. Now our issues are most likely related to either 1) the model design - was it designed with sufficient clearance and/or 2) extruder calibration - is it putting out too much filament when requested and/or 3) slicing configuration - print speeds, temperature, layer height, extrusion width, perimeter #, infill density and type, etc can all have an impact on dimensionality.

I'm not sure what the actual issue is - does the Pi not fit in the case? If not, simply scale both parts up by a few %.

[thedoble]

Thanks for your reply

In terms of the model design - the problem is twofold:

- the raspi board doesn't fit into the case, even though the outer case dimensions are correct, the inner dimensions are not, because the walls are thicker than they should be.

- the two halves of the case don't fit together. There are four 'pins' that are supposed to slot into the base. The STL is such that the pins are exactly aligned with the base (I.E. there is no 'tolerance' built in). I expect this is by design, the idea being a small tolerance offset in the printer itself (0.05mm for example) is enough to create a solid frictional bond. In my case however the pins are too wide, AND the base walls are too thick, so they won't fit together.

[thedoble]

I've done some more calibration today. First I re-printed the calibration circle another 5 times until I was getting extremely consistent horizontal rod lengths. I am confident that there is no more tweaking to be had in this department.

Next I turned my attention to the slicer. I designed and printed a 50x50mm calibration cube which had multiple interior wall thicknesses - 0.5mm, 1mm, 2mm, 4mm. When measuring these walls, here are the widths I was getting:

Outer X: 50.06mm
Outer Y: 50.13mm
0.5mm wall - 0.65mm
1mm wall - 1.13mm
2mm wall - 2.23mm
4mm wall - 4.25mm

[img]http://i.imgur.com/muO1PLA.jpg[/img]

Next, I tried printing a 50x50mm cube with only 1 perimeter layer, specifically set to 0.40mm extrusion width. When measuring the walls, it shows 0.39-0.41mm. This makes me think the extrusion / filament settings are correct.

[img]http://i.imgur.com/15oSbIc.jpg[/img]

Next I tried printing small 2x4mm pylons to see if the printer would produce them at the correct width. The pylons were all 2.26mm - 2.40mm in width, and 4.04mm - 4.23mm in depth.
I also tried printing at 0.1mm layer depth and found that the pylons were too wide by around 0.2mm as before.

[img]http://i.imgur.com/PetCBPZ.jpg[/img]

Interestingly, there seems to be some 'billowing' however I have measured them at their thinnest point and they are still too wide by 0.2mm+.

[img]http://i.imgur.com/0PZfvns.jpg[/img]

NOTE: Please ignore the curve in the first layer, this is simply because I roughly peeled it off the print bed once it was done.

Finally, I tried going back to mattercontrol instead of Repetier host / slic3r and found exactly the same results.

Any ideas folks?

[mhackney]

Are you using a part cooling fan? And using it appropriately (see my sticky post on fans)?

I think your calibration all looks good at this point. The billowed pylons are indicative of inefficient part cooling.

[thedoble]

I do have a layer fan, it's set to run at 30-100% depending on layer time.
However, it's not angled very well at the print itself, I have been waiting on some new fans to arrive to remedy this.

Would this have such a significant effect, especially when printing with PLA?
If so, is there a way to test for this, EG printing at a very slow speed to account for a lack of a layer fan?

Again, thanks for your help

[mhackney]

Please read my post on fans: http://forum.seemecnc.com/viewtopic.php ... 1584#p9768" onclick="window.open(this.href);return false;

[thedoble]

I've re-read your post, and will try adding additional cooling to see if this remedies my dimensions issue.

Thanks

[mhackney]

It's more about getting the air at the correct location rather than blasting a lot of air on it. If you blast air on the pylons above with such a thin first layer I guarantee you will peal the print. A thin well directed beam of air right around the nozzle will be much better. You can even hand hold a squirrel cage fan with a soda straw rigged up to it and direct it when you get to these critical areas. That's a quick and dirty approach. Even a slow even breath through a straw (or a balloon) will do it.

(click image to see the high resolution version)

The 3 pillars on the red part are 5mm in diameter. They are perfectly smooth and dimensioned. If I used ineffective cooling, they would look like your pylon test.

[mhackney]

Actually, I pointed you to the wrong post. NOTE - this has been superseded by tip 12 here: viewtopic.php?f=36&t=7361

[thedoble]

OK so I re-printed at 1mm/s speed and the pylons are still 2.35mm when they should be 2mm wide.

[img]http://i.imgur.com/0Nti3f0.jpg[/img]

[mhackney]

They look a lot better - smoother.

What are you using for infill? How many perimeters and inside out or outside in?

I would print something likes this with 2 perimeters, inside out. With 33% straight infill.

[nebbian]

What's your infill speed? Is it significantly higher than your external perimeter speed?

If so then there's your problem.

I've been chasing similar issues on my delta, and tracked it down to the speeds being too different. A high infill speed results in high pressure in the bowden cable, which leads to overextrusion when doing external perimeters.

I know it sounds strange but try setting all speeds to be the same and then remeasure.

[thedoble]

They look a lot better - smoother.
What are you using for infill? How many perimeters and inside out or outside in?
I would print something likes this with 2 perimeters, inside out. With 33% straight infill.

I've been printing at 20% infill - hexagonal - inside out, 2 perimeters.

What's your infill speed? Is it significantly higher than your external perimeter speed?
I know it sounds strange but try setting all speeds to be the same and then remeasure.

I've tried setting all speeds to 10mm/s, 5mm/s and 1mm/s, and if anything I get worse results at slower speeds.

I've spent another full day testing, printing a variety of walls, pylons, pegs and other items over and over 10+ times for consistency, while tweaking and adjusting various settings. Here's what I've found when trying to consistently produce 2mm wide walls:

• Reducing print speed doesn't improve accuracy, I have tried everywhere from 1mm/s to 15mm/s.
• Increasing layer resolution makes no difference. I have tried 0.3mm, 0.2mm, 0.1mm.
• Disabling 'detect thin walls' makes no difference.
• Adding additional cooling makes no difference. I tried holding a 2nd small fan right next to the print, and directing two 80mm fans at the print from the edges of the platen.
• Switching off infill seems to help - Wall thickness goes from 2.25mm to 2.1mm.
• Returning all my settings to 'normal' and printing at 85% extrusion multiplier results in the walls being the correct thickness of 2mm.
• However, this also results in the overall dimensions of objects being too small. EG a 50x50x20mm cube came out at 49.5mmx49.2mm
• Also the object has gaps in the sidewalls, as if the layers are not bonding correctly, see pic below.

[img]http://i.imgur.com/9kkhJO6.jpg[/img]

I have calibrated my e-steps and averaged my filament width.
Reducing the extrusion multiplier fixes one problem, but introduces another.
My impression is that the problem lies with the infill being too thick and 'pushing' the outer layers out so they are wider than they should be. I'm not sure how to fix this.

Any thoughts?

[mhackney]

Contact the original designer? When I design complicated assemblies I design in the tolerances. Perhaps s/he has some suggestions for getting this to print. Could be that s/he printed in ABS and you are trying PLA. In these early days, slicers just aren't sophisticated enough to deal with these nuances. I design for the material and the tolerances and test, test, test. On my fly reel in the photo above, I provide 4 versions of the white part with a bore diameter increasing by .2mm. This accommodates both printer calibration differences as well as material (PLA vs ABS vs PET) differences. You might have to print one or two to get the right fit, but one of them will work.

[thedoble]

Contact the original designer? When I design complicated assemblies I design in the tolerances

I've been doing all my testing with objects I've designed myself. These objects include:
2x10x10mm wall (most common)
50x50mm cube with various interior wall widths from 0.5mm to 4mm.
2x30x10mm wall with chamfered edges of various depths.

So far I'm struggling to get the correct dimensions. EG a 2mm wall is 2.25mm, a 4mm wall is 4.3mm etc

[mhackney]

Ok, on these things you designed, if the walls come out to 2.25, make them narrower and reprint. Repeat until you get the correct dimensions. It's the nature of FFF and the state of the current slicers. You, as designer, need to be aware of this and accommodate critical dimensions in your designs.

[thedoble]

Ok so this comes back to my earlier question - should I expect accurate dimensions from my printer?

If not, I can deal with that, but so far no-one has told me that I shouldn't expect +/-0.05mm accuracy from my printer.

[thedoble]

Does it also mean that if very popular designs on thingiverse have no tolerances built in, then other printers are capable of producing accurate dimensions? Surely if this is a common slicer / printer issue, all popular designs would have a tolerance built in to be successful.

[mhackney]

There are limitations to the technology. Thin cross sections and small diameter rods and holes are all on the problematic end of the spectrum in terms of tolerances. Part of this is due to the limitation of the CAD model->STL->line segment gcode process. Read http://sublimelayers.blogspot.com/2016/ ... count.html" onclick="window.open(this.href);return false; for some insight into this. Small feature are more difficult from a dimensionality perspective. But even with large items +/- .05 mm is not really feasible on desktop consumer grade printers. Read this: http://machinedesign.com/3d-printing/ac ... ufacturing" onclick="window.open(this.href);return false; for some more info. The high end guys claim +/- .089 mm (see paper).

Frankly, most of the stuff (multipart stuff specifically) on the sharing services is touch and go. "Popularity" has nothing to do with the original designer's abilities and the "goodness" of the design. There are many, many popular models that have all sorts of STL issues (non manifold, shared edges, etc) or too low triangle count, not to mention not really designed to be reproduced on someone else's 3D printer. This leaves folks like you and me to use slicing "tricks" to compensate for these issues. Sometimes you can and sometimes you can't.

[geneb]

Here's another good resource: http://hydraraptor.blogspot.com/2011/02/polyholes.html" onclick="window.open(this.href);return false;

g.

[bot]

While there are definitely limitations in the slicers, and in the math when dealing with faceted curves... the Rostock max is also not the greatest printer at holding tolerances. You should definitely not expect to hold a 0.05mm tolerance. The "thin walls being thick by .3mm" is something I encounter on my rostock max as well. I have tried everything short of ripping my hair (completely) out. You'll just have to learn to live with it. Michael's tips are good. If you can, design for the tolerances; know what to expect, etc. Short of building/buying a new printer, there isn't much you can do. There is too much free play in the system to allow the precision you are expecting.

The gaps in your walls points to under-extrusion, however. This can be resolved.

[mhackney]

I also have a finely tuned Taz 4 Cartesian printer where I've modified the linear guides and motion control elements to get very little backlash. It suffers from exactly the same issues as the Rostock in terms of dimensional accuracy. One thing you can do is take a look at a rendered gcode layer of these thin wall/diameter features and I think you might be surprised at what you find. If you only have the gcode paths, just measure a diameter or width of one of these small features and add one nozzle (or extrusion) diameter width to get what the slicer is attempting to print. Yes, I spend way too much time thinking about and studying this stuff. I can hand-write gcode that will print almost exactly what I want/expect but that isn't practical. And if you think that is bad, don't get me started on what slicers do along the Z dimension in terms of dealing with uniform layer thickness round-offs.

[mhackney]

The Machine Design site has a number of papers (mostly marketing propaganda) that you can download. This is from a Stratisys paper:

The key to FDM’s accuracy and precision is the
coupling of material feed rates and extrusion head
motion. Both are constantly changing to produce
a flat ribbon of material that measures from 0.008
in. to 0.038 in. wide (0.20 mm to 0.97 mm). On the
highest-performance FDM machines, part accuracy
or tolerance reaches as high as 0.003 in. (0.08 mm),
which rivals injection molding.

So they are claiming +/-0.08mm tolerance for their high end FDM machines.

[mhackney]

Here's another thing to try - rotate the part 45 degrees and reslice and print. Sometimes that "tricks" the slicer a bit.