What is precision and accuracy?

[KAS]

So I'm new to the 3D printing world but I'm really stuck on this precision kick and not sure where to look.

I thought my layers were to close together so I printed a 20x20x20 cube to check the height. These are my results:

Am I missing the point of the 3D printer and shouldn't worry about it or can the accuracy be reached with more software/hardware calibration?

[img]http://i482.photobucket.com/albums/rr189/Onuxis/20141230_203510.jpg[/img]

[img]http://i482.photobucket.com/albums/rr189/Onuxis/20141230_203403.jpg[/img]

[img]http://i482.photobucket.com/albums/rr189/Onuxis/20141230_203313.jpg[/img]


This picture doesnt do it justice. It literally has 3-4 layers between the notches of a quarter.
[img]http://i482.photobucket.com/albums/rr189/Onuxis/20141230_203838.jpg[/img]

[JFettig]

Reduce your delta arm length 1-5%

[lordbinky]

That will widen the bowl of the movement. The effect of which will change behavior most noticeably near the center where small test cubes are made, the difference will still show back up on larger prints (how large is unknown without getting a probe and mapping the bed). The core issue is likely to be a geometry issue (either a tower is off even a tenth of a degree, or more likely the towers themselves aren't the exact same distance from center. It may be easier to stretch your models by the offending direction to correct this

[dunginhawk]

or you can look at it from the perspective that out of 20mm you are off by a tiny percent and be happy, as i would be

[Khalid Khattak]

OMG..you are only ~0.1xmm off ... just think one of layer extended a little... if i were you i would be the happier person in the world with this accuracy

[mhackney]

Accuracy is how close the measured value is to actual. Precision is how close a set of measurements are to each other. You have provided one set of measurements for width, length and height. From this we can calculate accuracy in % as (error / actual) * 100. You will need to print multiple cubes to calculate precision.

For your first photo, error = 20.06 - 20.00 = 0.06
Accuracy = (0.06/20.00) * 100 = 0.3%

if you printed 5 cubes and all 5 were 20.06 then your precision is very high. If they varied from 19.05 to 21.09 your precision is much lower. There is a formula for that too but it's late!

Cheers
Michael

[lordbinky]

Depends what you're using the parts for though. Some test prints lend themselves to functional items that need to be accurate if you don't want to do little post print effort. For example http://www.thingiverse.com/thing:11261 the nickle calibration lends itself to working with other physical objects, such that it's easy to picture how you'd want a printer calibrated you're best if you wanted a print snugly fit on an object your designing it for.

[KAS]

Thank you all for the replies. I definitely understand the difference between accuracy and precision just lost on how to obtain it. I'm just looking for ways to improve on something I know very little about.

Measurements I took while the nozzle and bed was heated. Not sure how to find true implied center but these were taken with the nozzle touching paper using the macro's made in the assembly guide.


Center to Z 89.39mm
Center to Y 89.36mm
Center to X 89.58mm

The height of the objects made, I'm fine with. It's the other dimensions that I feel will cause issues if I try to design something with screw holes for example.



Took a while to find two nickels at the correct size of 21.21
[img]http://i482.photobucket.com/albums/rr189/Onuxis/20141231_101435.jpg[/img]

[mhackney]

Great! At this point I do think you want to establish that you are getting good precision by printing a few of the cubes and comparing measurements. If you are not, this likely indicates a mechanical issue - slop in the arm joints or carriages are primary culprits. Don't overlook loose tower screws! After a year of heavy use I discovered that the screws attaching the towers to the base were loose on my X. Test precision and then go from there.

If precision is good, then I suspect either a geometry or delta configuration issue. lordblinky describes the geometry issue, JFettig describes a firmware issue. Making quick tweaks to firmware and testing is fast and easy so I start there. Then test parts of different sizes from small to large and measure. You may find the variation is acceptable - it's a compromise. If not, then you need to spend more time measuring and tweaking the mechanical structure to eliminate any issues. For my original Rostock 1, I have it accurate in a build volume of about 90mm x 90mm x 90mm - where I do most of my precision work (fly fishing reels). Larger parts do "shrink" outside of this ideal envelope. In those cases I tweak the STL X and Y scale by a fudge factor (I keep a list mapping X and Y positions to a fudge factor in 2 cm increments). If I am making a lot of the same part, I'll measure the first one and update my scaling factors as needed. I concluded that this was easier than trying to chase down structure gremlins!

cheers,
Michael

[KAS]

Awesome, I'll definitely work the machine calibration when this print is over. I'm noticing some cheapskate wheels that appear loose or slipping a bit.

Right now it's 75% into an 9 hour print.


[img]http://i482.photobucket.com/albums/rr189/Onuxis/20141231_172454.jpg[/img]

[KAS]

Spoke to soon. Started to get an odd artifact climbing up the side at an angle and notice the cable loom was catching on arm during the same nozzle rotation.
Also started seeing the edge start to step off intended path towards the top.

Time to go over everything again.

[img]http://i482.photobucket.com/albums/rr189/Onuxis/20141231_190843.jpg[/img]

[lordbinky]

You can try lowering your jerk setting or try lowering the speed and see if that clears up the edge shifting. There's a few threads about chasing that one down and correcting it, but it usually comes down to missed steps. The artifact looks like a seam where a layer ends/stops

[KAS]

I've been working on this all day yesterday and cant figure it out. I've made some extreme changes to X&Y AXIS_STEPS_PER_MM and DELTA_DIAGONAL_ROD. For the life of me cant see the difference in measurements.

I noticed this in the firmware Configuration.h "To override EEPROM settings with config settings, set EEPROM_MODE 0"
So now I understand why the changes wouldn't work, but is it the EEPROM settings in the firmware (Eeprom.h) or the settings in Mattercontrol?

Or does MatterControl read the firmware every-time it connects?



I

[KAS]

haha think I figured that one out. don't change the EEPROM in MatterControl to 90 from 80 steps per mm and click the bed center macro....

[KAS]

I like to say it was easy and fast to correct the accuracy, that was not the case...

But the research and posts from others have helped greatly.


The two 21.21 nickels now fit snugly within the slots. The circle on the other hand appears to have polygon lines forming the circle and I'm not that worried as its very close to fitting.

Ended up buying a new square that was actually square. Applied a bit more pressure on the drive belts. Corrected the two loose end stop switches. Adjusted one stepper motor gear that was offset. Adjusted the steps per mm to match my measurements. And finally offset the X axis by -.052 (not proud of that one) I just couldn't get the towers aligned enough with that plastic screw holding the top plate down.

and needless to say way to many calibrations....

[img]http://i482.photobucket.com/albums/rr189/Onuxis/20150102_142433.jpg[/img]