[SO3XXL] Still working on calibration, big headache

(Luigi Carnevale) #1

Just finished a calibration session, still in a limbo.
This time used this method: Tutorial: Calibrating axis belt stretch on SO3 with a caliper
which for me is the more logical (and reliable) thank you mikep for the tutorial.

Test distance = 120mm
Pre-calibration reading was 120.02mm

Calibrated (via software) for 120.00 exact millimeters with a mitutoyo caliper. X and Y axis.
Tried to move forward an back several time, the reading was always 120.00 mm dead on.

But, if I move in 3 x 40mm steps I have 3 different readings, something like: 30.98/40.03/40.04 etc, don’t remember exactly, anyway the point is that calibration seems related to a fixed distance. The first drag is alway shorter of about -0.03mm similar to a backlash issue.

Same issues for X and Y axis.

After calibration, tried a 50x50 mm square profile and a circular pocket.
The square profile was worst than the pre calibration square it was off by 0.25mm, the pre calibration was off by 0.15 at max.
The circle is an ellipse with about 0.2 mm deformations.

Tomorrow I will try to reset the values to the defaults and make a new and probably last session before giving up.
The problem is the values are not consistent. If I calibrated for 120.00 mm, why I read 40.98 for a 50 mm command?
The V Wheels are just tight enough to avoid any unwanted movement of the carriages, I can freely move the carriages with a decent effort.
The belts are tight, I should have problems the get them more tight than this.
The machin is squared enough, anyway for a single axis measure, squareness is irrelevant I think.

Thank you


(William Adams) #2

For backlash, be sure to check the pulley set screws.

Beyond that, some folks have actually changed the CAD so as to adjust it to the dimensions of a cut piece so as to correct for belt irregularities.

Another option, if your belts are not Gates branded might be to buy Gates branded belts from SDP/SI or BB/Man. Given how much more that they cost, they ought to be better in some way, and more precise / accurate might be the case.

1 Like

(mikep) #3

@WillAdams Do you know what the specific GT2/3 part numbers are?


(Luigi Carnevale) #4

Hi Will, could you please elaborate this? What you mean for CAD?

I have a new machine and a new complete belt set, so at least for now, I don’t want to spend more money on this.

Thank you


(Phil Thien) #5

My procedure is to calibrate over the widest distance possible. I have a 12" digital readout I use, this one:

Once I’ve calibrated the belts over that 12", I then check every 1-2" (25-50mm) and make sure I’m to within .001 or .002" or so. At 2" I might be +.002, at 4" I might be only +.001". Sometimes my variance is below the threshold of my ability to measure.

In MM, we’re looking for values .05mm or better.

Those are very respectable values. I also have a unit with Acme screws and I can tell you that with those values you’re seeing, you’re approaching what a decent machine with Acme screws can achieve when you combine the screw accuracy with nut accuracy with backlash, etc.

I was cutting 1/4" acrylic the other day, and my 4" wide part was +.0025" over. Not bad at all.

1 Like

(William Adams) #6

Draw up a part — cut it, measure it, adjust the CAD by how much the cut part is off, cut a second — iterate until the part is exactly correct.

1 Like

(Luigi Carnevale) #7

So, it seems I already reached the max precision of this machine regarding belt/linear calibration.
I have to check for better a better squaring… Thank you cgallery.

This is the phase 2 Will. What I’m trying to do is to get the max precision from the machine to adapt the projects design to it. I have to know what are the tolerances I can rely on when I make a design.

Thank you for your help


(Chris) #8

the offset command in vcarve pro makes this amazingly easy to add “cutter” comp to the parts


(Richard Cournoyer) #9

Once I was satisfied that my belts were properly tightened, and after chasing belt calibration for months (some areas of machine were great, other’s got worse, some sizes were great, some larger or smaller got worse, etc.), I finally got smart and mapped out the entire cutting area (S3, using a set of calibrated Gauge Blocks) and determined that the BELTS and PULLEYS vary in their timing. Some areas of the belt (inch to inch) are long (meaning 1.000" measure 1.004) while other areas were small (e.g. 1.000" = 0.998").

This confirmed my hypostasis that adjusting the steps per revolution was fruitless, and at that point I stopped chasing my tail. As Will said above, if I am doing something extra precise (Bearing surfaces or holes) I cut the feature under (or over) size and then adjust the program (AND Shape) to cut what I need. Using this method I have cut bearing bores to within 0.0002" (and in a few large holes, they are round too by programming the machine to cut an ellipses, since the Y error was a little more than the Y Axis).

Let me know if I didn’t make myself perfectly clear, or if you have any questions.


Scratching my head
(Phil Thien) #10

Rich is crazy-talented, and if you need to machine to within tenths (that is ten-thousandths of an inch), he is “the man.”

But I’m just going to point out that if you search typical CNC router kit packages, you’ll find (if they say) accuracy indications of +/- .005" or so, and repeatability of maybe half of that.

Here is one such statement:

"That being said, we can make some generalizations based on our own tests and on data from our customers. In general, accuracy of +/-0.005” can be achieved on cuts without too much trouble in a localized area. Repeatability is typically better, and on well-tuned systems, repeatability of 0.002” or better has been reported.

“One specific area of concern related to accuracy is backlash. Backlash is positioning error caused when an axis switches direction. If the direction of rotation changes, any lag in the change of direction in linear motion is not seen by the system and results in positioning error. On both our R&P drives and Acme screw based systems, properly setup systems have very low amounts of backlash, typically less than 0.001” in our PRO setups. To put that in perspective, that’s about the thickness of 1/4 of a sheet of notebook paper.”

Not to knock cncrouterparts.com stuff, but I feel I’m doing better than that by quite a margin.

I recently made some more router jig templates. The parts are about 4" square, and I get four (2x2) out of a 12" piece of acrylic. They are accurate to within +.002" or so, and my variance from one to the next is .001" or better. I’m actually pushing my ability to measure these with my inexpensive digital gear, which differ from one another (I have a couple of them) by .001".

I think sometimes we get hung-up on accuracy. If you do what I tell you, your machine will be pushing the envelope for accuracy for desktop CNC routers.

If that isn’t good enough and you want to tear-up the envelope, Rich will get you there, his method doesn’t amount to that much different than how they map high-end industrial equipment, Rich just has the #'s jotted down somewhere (or maybe in his head).


(Richard Cournoyer) #11

Thanks Phil. The results are in a nice Excel Spreadsheet, and I posted a condensed version on here back in January (ish)…as well as on my IG (which is easier to find things then here).

PS. These numbers have been checked and re-checked, and are quite repeatable.

PS2. I stand by this statement: Out of the box, a properly set up Shapeoko will consistently produce +/- 0.003 accuracy and will repeat within 0.001".

Here are my numbers again:


Scratching my head
(Luigi Carnevale) #12

Hi Rich, I perfectly agree with you. In fact, yesterday, after a calibrating session, I ended up resetting the x/y values to the default value of 40 because the measurement was not consistent. I had variation in linear testing on a single axis with no load, and test pieces, no matter how accurate was the corrections.

Also, I made my testing with scrap wood, most MDF, so measurements was not so accurate. I need some HDPE, Delrin, etc to get more accurate results.

Anyway, at least, now I have a rough idea of the capabilities of the machine.

Regarding your spreadsheet, I do not understand how to read it…

Thank you all for your help


(William Adams) #13

One consideration w/ calibration is that it is like to the problem of layer height and # of steps / mm on 3D printers.

The belts have a nice even # of steps, until one does the calibration, then, each step synchs up with an increasingly odd decimal as one moves away from machine origin — one can see something of that in the representation of zero — it’s perfectly 0.00 at the machine origin, but elsewhere, it’s frequently -0.00.

Or, let’s turn this around — let’s imagine that rather than calibrating the steps / mm we instead try to determine where each possibly location is.

Consider an XL which when using the rapid coordinates has an SW origin of 0,0 and a NE coordinate of 810,350

Using the measured number from: http://docs.carbide3d.com/shapeoko-faq/how-to-calibrate-the-machine-for-belt-stretch/ a requested movement of 75mm results in 75.03mm measured movement

With 40 steps / mm, when we have the machine move from the SW origin to the NE coordinate it takes 810 * 40 steps along the X-axis and 350 * 40 steps along the Y axis to get to 810 * 350

Each of these steps rather than moving 1mm, actually moves 1.0004 mm (75mm (desired movement) ÷ 75.03mm (actual movement) = 0.999600159936026 which when divided into 1 yields 1.0004mm)

810 * 1.0004 == 810.324 

So uncalibrated, the machine won’t be off by more than a third of a millimeter along the longest axis.

In making that move the machine takes 4,050 (810 * 5) full steps and 32,400 (810 * 40) microsteps resulting in a grid of 0.025mm of possible movements at 1/8 stepping.

Things which you can do:

  • align your part so that all critical dimensions are positioned at this grid (as offset by the effective endmill diameter) — note that anything not aligned against the grid will be off by how much each measured coordinate is offset from it
  • get 400 steps / rev. motors — that will double the precision of the movement and allow you to set 80 steps / mm
  • get larger pulleys (which may require more powerful motors which may require replacement electronics)
  • implement a pulley reduction system — there’s a plan for one on the wiki

Please note that any linear movement system using stepper motors will have such an underlying grid system — it’s just smaller on things which use screws and so forth. These are digital systems and one is subject to the vagaries of a binary on/off grid fitting.

1 Like

(Phil Thien) #14

One thing I learned early-on was to not calibrate the machine by cutting test pieces. There are just too many variables, from deflection to bit variance from specification to RPMs. And those variables will change depending on material, depth of cut, cutting speed, etc.

Thus I measure my belt stretch over the widest area possible with my 12" DRO, and then I’m done. If I need to make further adjustments, it is to my CAD. I use (mostly) Vectric software, which allows me to perform final compensation fairly easily.


(Richard Cournoyer) #15

Well the spreadsheet data wasn’t made for general public distribution; moreover , it’s was for me to better understand the machine’s overall movement, but I’ll give you a 50,000 foot level overview.

Position column: Position where I am measuring relative to home (X0 Y0)

Front, Center Back (Left Center Right) position looking at the front of the machine

Data example: X-1.600", Front, -0.0008" Means that from 0.600" to 1.600" (one inch of travel) MY machine was off eight ten thousands (Yes, less than 0.001") ((Negative means small, Otherwise it measures long))

Light Green: Excess error (Mainly a reminder to me to go and recheck this large error)

Yellow: All the error is in the same direction (whether long OR short)

NOTE: This also helped me understand why changing the steps per inch was a waste of time. For instance, from 11.600" to 12.600" ((one inch along the X Axis), my machine (average) measured LONG 0.0026", BUT the very NEXT inch, 12.600" to 13.600" my machine measures SHORT -0.0031" ((Clear??)

I hope this helps.

PS SOMEDAY, Shapeoko will send me a set of “high-end” belts (Probable USA made Gates) for me to compare against the OEM belts…Will they be better or worse???

1 Like

(William Adams) #16

As I’ve noted in other threads, I bought Gates-branded (presumably made in the U.S.) belts and pulleys from SDP/SI back when I did my upgrade to 9mm belts (and had extra so that I was able to use them on my X-axis when I did my XL upgrade) — I hope that they’re better given the cost difference.


(Phil Thien) #17

One of the things I’ve pondered recently is whether my belts have stretched, or whether my cogged wheels are slightly oversized.

One of the first things I cut on the Shapeoko was a hole for a subwoofer project I was working on, and I was almost (not quite and I don’t remember the actual measurement) .01" oversized for an approx. 10" hole.

That, and some reading here, is what caused me to start measuring.

Either my Y or X-axis needed quite a bit more adjusting, I can’t remember which right now.

I wonder what the tolerances are on the cogged wheels.


(Phil Thien) #18

BTW, I hope my posting here isn’t seen as a disagreement with your perspective and expertise. It isn’t intended as such, only sharing my own observations.

We’re lucky to have you around here.

1 Like

(mikep) #19

The belts definitely stretch over time. I’ve had to retension my belts three times in the past 18 months.

1 Like

(Luigi Carnevale) #20

Thank you all mates for your help, lot of information in this thread!

After few attempts in calibrating for belt stretching I decided to live with the default values, I am happy my conclusion is the same of other people (Richard).
I think my machine is enough squared and calibrated, now it’s a matter of skill to get a higher precision (when needed).

Regarding Belt Stretching, a few questions arise:
Does Belt stretching is constant across the whole length of the belt? I mean: if we make a sign every 25mm on a belt, then put the belt under tension and measure the signs, will the distance between each sign will be the same?
If the stretching is variable across the length of the belt, the software correction makes a lot less sense to me.

Belts don’t last forever, so when you know it’s time to replace a stretched belt?

Thank you