More information about the calibration benchmark test

The calibration benchmark test is a tool to help us measure how well calibrated a machine is. We came up with the test as a “before and after” test to check if a change to the calibration process made the machine more accurate or less accurate. We wanted a test which would let us compare apples to apples instead of cutting different shapes in different places.

The test can also be a good way to test how well calibrated your machine is. Right now a good calibration score is something in the neighborhood 1.0-1.0. More on how that number is computed below.

You can easily load the calibration benchmark test by clicking Actions → Advanced → Load Calibration Benchmark Test

The test is a gcode file which looks like this:

The test cuts shapes near the edges of the sheet because calibration inaccuracies are most pronounced there. The test cuts four rectangles and four lines.

When measuring the test pattern all measurements should be taken from one side of a cut to the same side of the opposite cut so that the size of the bit is canceled out. For example from the right edge of one cut to the right edge of the other cut.

The squares should be 100mm which is small enough to be measured by a standard set of calipers.

The vertical cut should be 900mm apart

The horizontal cuts should be 1905mm apart. 1905mm is a legacy number which we used to use as part of the calibration process. I don’t even remember why we used that number.

The results of measuring these distances is boiled down into a two number score like 5.43-2.10 where the first number reflects accuracy in the long measurements (1905mm and 900mm) and the second number reflects the machine’s accuracy when cutting small parts in the far corners of the sheet.

Each number is the average error in the measurements. For example if we had the measurements (902, 900, 899, 1903, 1906, 1904) for our long cuts we would compute.

Measured Expected Error
902 900 2.0
900 900 0.0
899 900 1.0
1903 1905 2.0
1906 1905 1.0
1904 1905 1.0

for an average error of (2+0+1+2+1+1)/6 = 1.16

I just wanted to put all that into writing somewhere we can reference it. Let me know if I made any arithmetic mistakes in my example :upside_down_face:


Slight thread necromancy, but 1905mm converts directly to 75in. That’s probably why it was chosen :stuck_out_tongue:

1 Like

So I ran through calibration for the first time today with 1.13 and was pleasantly surprised that the calibration shape was different, just 4 vertical lines, and 1 horizontal line. Is this a different than the calibration benchmark test listed here, or is this the new version?

The four vertial marks are used by the machine to calibrate the machine. It uses those four marks to refine the measurement of the rotation radius and to determine how much the sled weighs.

The calibration test pattern is something that we came up with to put a number on how accurate a machine is once calibrated.

They are too different things, but closely related :grinning:


So the calibration test pattern is just informational and doesn’t further refine the positional accuracy of the machine? But the “4 marks test” does update machine parameters?


Exactly! You got it @LakeWorthB

1 Like

I am getting an error message that the Calibration benchmark test has been moved or deleted when I click on the Load Calibration benchmark test. where is it locate and how do I find it. I just downloaded 1.16 this morning.

1 Like

I just downloaded, tested and it works.
It should be in the folder -> gcodeforTesting

Which operating system are you using? It might be the case that the macOS or Windows package is at fault.

1 Like

using a Mac. I suspect that is the difference. I did not see a folder in the contents of the application package with that name. anyone know the Mac path?

1 Like

I looked at an older GC on the mac and the files are not there.
Calibration Benchmark (567 Bytes)

You can get more attention opening a new topic under the troubleshooting category with a speaking title like : “Benchmark .nc not found on mac” or so :wink:

1 Like

Thanks for the detective work, @Gero, I’ll find the cause and get a corrected version ready.

1 Like

Just made sure it’s the same with 1.16 :slight_smile:

I opened issue #736 and PR#737 for this. The package-building software needed to be told to include the ‘.nc’ files of the gcodeForTesting directory.


I remeasured. Average of 6mm off. Mostly in center and top left Y being 7mm and 6mm everything else is within 2mm. :crazy_face:
From upper left its 6mm, 7mm, 1mm in Y (vertical) and
0mm in X (horizontal).
Any ideas why?
FYI the sled is staying level for all cuts.

@bar, I feel silly but I’m missing something here. I get the 1.16 average shown in the example above but how should the two different numbers be calculated?

The first number is calculated like this?


…meaning the 100 mm squares? Is the second number calculated the same way but using measured - expected width and height for each square?

Is this right? Note, the “Big Sq2” and “Big Crl2” are for a test cut that I made (concentric squares and circles that are centered just further out from the 100 mm squares).


You got it! The second number is measured exactly the same way except that all of the expected values are 100.

1 Like

What are you guys using to measure these distances? I feel like a standard tape measure is too inaccurate, or rather the human eye

the small squares are able to be measured with a cheap digital caliper (similar
to )

longer distances are with a tape measure.

take a look at the ‘in search of accuracy’ thread
In search of accurate measurements where I
am working on a 3d printed part to pair up with a tape measure to get sub-mm
accuracy. I used a set a couple days ago and it worked well and was
substantially faster than just using a tape measure (or so I’m told by someone
who has calibrated more machines manually than I have)

David Lang


I have an interesting idea about measuring longer distances accurately. Assuming you can accurately place the ends, and have enough people or clamps to hold everything in place. Take two tape measures, one metric and one imperial. Extend them from the endpoints so the meet in the middle. Then carefully look at the marks on both measures to find a pair of marks that align precisely. Record the two lengths, convert the “wrong” one to your desired units, and add them up.

The idea is based on a vernier scale, which is a very precise way of measuring distances used in things like calipers. Well, it was until dial and digital calipers came along. This is not a vernier scale, but I think one would still be able to find a pair of marks that line up and get a very accurate measurement.