Accuracy: Expectation vs Reality

I have been following the posts at the bottom very closely trying to dial in my Maslow. I would like to get some feedback on similar issues but I do not want to move the conversation away from solving @Dustcloud 's issues.

Measurements:
Manual:
Outside motor housing to outside motor housing = 3045 - 40.4 = 3004.6 (under tension)
Rotational Radius = 143.175

Maslow:
3000.46 left and right chain
Rotational Radius after Calibration = 135.8
Chain Sag = 21.05713
Chain Sag from previous .ini = 18.965783

Benchmark Test results with Maslow’s measurement for distance between motors: 3000.46:

Questions:

  1. @bar I saw in an old post you were measuring perfectly round circles Fully Calibrated but Oblong Circles - #43 by bar . Do you have your benchmark numbers I can compare to?

  2. Is anyone using the Chain Tolerance?

    • Does Chain Tolerance help with accuracy on cuts or is this just to get the Maslow to accurately measure distance between motors(match the manual measurement)?
  3. Where do I go from here to shoot for better accuracy?

    • Adjust the motor to motor to my manual measurement?
    • If I do this, do I need to run the Triangulation test again to update chain sag and rotation radius? It looks like I get a different chain sag/rotational radius each time I run the calibration.
    • If I have a unrealistic expectation on accuracy and my numbers look good, just tell me to shut up :grinning:

Thanks!

2 Likes

When we came up with the test pattern it was intended to be a way to give us a single number (well really two numbers) so that we could test if a change was making things better or worse.

The idea was to average the error on each of the long cuts and each of the short cuts to give a score. For this pattern I’m computing a score of 5.16-0.32. The best score I’ve gotten is right around 1.0-1.0 with a couple folks doing better if I remember right.

We came up with the test pattern while testing the chain sag code and did a lot of test patterns right around then, but I can’t find them now. Does anyone know where those went? I think they would be a good reference.

I think the answer to this is that not many people are using it yet because we’re still exploring how to use it, but I think it is an important thing to explore.

3 Likes

This needs a proper: https://github.com/MaslowCNC/Mechanics/wiki/Calibration-Guide that brings in everything learned so far (from Z-axis slop, to offsets for triangulation kits, and chain pitch)

3 Likes

@arnoldcp Can you explain to me how you are calculating your chain pitch to be 6.359? Or how I can calculate my chain pitch?

1 Like

First Benchmark test after calibration with Maslow motor to motor measurement 3000.46

Second Benchmark test with Manual motor to motor measurement 3004.6 (no calibration)

Benchmark test after triangulation test with Manual motor to motor measurement 3004.6 (With Calibration)

Reloaded ini from “First Benchmark test” and added chain tolerance of .1377887

With the last test, the chain tolerance test, I tested an 18" circle and it came out round at 17 15/16
However, when I closed and reopened Ground Control, my chain tolerances were not saved.

5 Likes
  1. @bar I saw in an old post you were measuring perfectly round circles Fully Calibrated but Oblong Circles . Do you have your benchmark numbers I can compare to?

  2. Is anyone using the Chain Tolerance?

    • Does Chain Tolerance help with accuracy on cuts or is this just to get the Maslow to accurately measure distance between motors(match the manual measurement)?

nobody is using chain tolerance much yet as it hasn’t been released, but it will
affect accuracy as well as the measurement of the distance between the motors.

if the distance between the motors is off by 10’s of mm, then the length of the
chain from motor to the sled can be off just as much.

  1. Where do I go from here to shoot for better accuracy?
    • Adjust the motor to motor to my manual measurement?

yes, and set the chain tolerance so that the chain measured value matches the
real distance.

* If I do this, do I need to run the Triangulation test again to update
chain sag and rotation radius? It looks like I get a different chain
sag/rotational radius each time I run the calibration.

Yes, after changing anything significant about the machine, you need to re-run
the calibration.

since the calibration requires imprecise manual measurements, it’s not
unexpected to get slightly different numbers each time. how much do they vary

* If I have a unrealistic expectation on accuracy and my numbers look good, just tell me to shut up :grinning:

we are still working to figure out how accurate we can make the machine.

As we solve one problem, we run into the next one. We don’t know how accurate
the machine can be with the chain tolerance factored in, so we don’t know where
we will be after that.

pretty much all the problems (and their fixes) have been the result of people
trying to be accurate and running into problems and our exploration about what
is odd about their situation and how to account for the variability.

3 Likes

Is this the correct way to measure the benchmark test?
benchmark%20measurement

1 Like

Yes, that looks correct. All measurements should be from inside edge to outside edge so that the width of the bit is not measured…and because that is an easier measurement with a tape measure

2 Likes

Here are what I’m getting for the scores for these tests:

5.16-.32 so there is a lot of improvement over long distances, but you are looking really good over short distances which means your machine is nice and stiff

5.00-1.11 So not really any noticeable improvement there. More distortion in the far corners where the squares are cut

6.16-0.99

and finally

2.66-0.70 So a stand out improvement there.

That looks like a really comprehensive test to me which stays that we need to focus on improving the chain tolerance code. It has known bugs right now and it needs to be computed independently for each chain, but it’s already showing the most improvement over any other change tested.

Excellent testing!! :slight_smile: :+1: :+1:

4 Likes

Just for comparison, here are results on my Maslow: dist. beetwen motors - 2962.55mm; motor offset hight 406.4mm; center of gravity 65.00mm
I run benchmark g-code ( 1905x900mmm; squares 100x100mm) and result is as follow: all in mm
(measured using tape)
width: top -1905
middle - 1904
bottom - 1904
hight: left - 898
middle - 896
right - 897
SQUARES
(measured using calliper)
top left
vertical sides
left - 98.2
right - 98.2
horizontal sides
top - 99.1
bottom - 98.6
bottom left
vertical side
left - 100.0
right - 100.0
horizontal side
top - 98.4
bottom - 98.0
top right
vertical sides
left - 99.0
right - 99.0
horizontal sides
top - 100.0
bottom - 99.5
bottom right
vertical sides
left - 98.7
right - 98.5
horizontal sides
top - 99.6
bottom - 99.1

It looks that my machine is consistently on minus side. My sled is “temporary” so my next build should be better. I think that biggest improvement can be made once axis of ring is align with cutter axis. That look lake a challenge - when rotating sled on chains by hand I see circular movement of the bit. If I can only eliminate such movement, improvement will be significant. Another story is frame - after having experience with “new” frame I think that original frame is much better design.
Here is photo of my Maslow
IMG_5927

1 Like

If I am reading this correctly, you got a score of 1.83 x .875

Thank you but how score is calculated and what is average value and acceptable value
Tomasz

A good thread to follow

BenchmarkAccuracyTest.xlsx (10.1 KB)

1 Like

Thanks a lot for xls - now I have a tool to try-and-error for improvement. Looks that higher the number the poorer performance.
How about “Maslow Challenge” - a score board where community may publish their best score (with description how they achieve it)

First prise - a bit
second prise - set of stake knifes :wink:
third - nothing ;-o

1 Like

Third place: an infinity stone. The one that gets me money back (or a coupon on the sequel at least). :slight_smile:

1 Like

what is it that makes you think the original frame was better than the current one?

Two things: top beam and slack chain.

  1. Chain connected to sled pulls motor down (vertical component of sled weight). Horizontal beam attached to frame a good 2 feet from motor is not strong enough and yields under force. Put you thumb near motor and push it down: relatively small force causes the beam to move (2x4 looked like strong enough solution, but it is not). That movement adds to error at the cutting bit.
    In original frame, motor support arm is oriented ALONG pulling force. This is compression force (not bending force) and 2x4 will not move. (True is that motor is not attached directly to support arm but shifted on the plate - this will cause torsion moment and misalignment of sprocket but it is another story).
  2. Weight of slack part of chain pulls chain into vertical plane (not cutting plane) and in current design working AGAINST tenson of the elastic cord. It causes all sort of problems with rubbing top beam, tangling with sled and elastic cord.
    In original design, the weight of slack chain working WITH cord tension. The slack part of chain is away of any part of the frame and elastic cord just nudge loop into cutting plane. Top feed is much better more elegant solution.
    Sorry for too many words ;-( but I hope to make my point. Here are pictures for illustration.
    IMG_5898
    IMG_5897
    I solved slack problem with additional pulleys, rope and weights - not elegant and unnecessary complication. Original frame do not have such problems by design.
    I also had to move forward a cutting plane to minimise misalignment of sprocket and chain connected to sled.
    IMG_5905
1 Like

Two things: top beam and slack chain.

  1. Chain connected to sled pulls motor down (vertical component of sled
    weight). Horizontal beam attached to frame a good 2 feet from motor is not
    strong enough and yields under force. Put you thumb near motor and push it
    down: relatively small force causes the beam to move (2x4 looked like strong
    enough solution, but it is not). That movement adds to error at the cutting
    bit. In original frame, motor support arm is oriented ALONG pulling force.
    This is compression force (not bending force) and 2x4 will not move. (True is
    that motor is not attached directly to support arm but shifted on the plate -
    this will cause torsion moment and misalignment of sprocket but it is another
    story).

In the original frame, we were seeing the entire frame warp and move the motors
closer together (there is far more force horizontally between the motors when
you are near the top of the workspace than there is vertically at any location).
This was causing very significant errors.

Are you sure that the 2x4 is bending under the force? I would guess that the
horizontal arms that support the top beam are moving. Are they glued or just
screwed to the vertical legs?

If the top beam is bending, then you have spotted a new problem.

  1. Weight of slack part of chain pulls chain into vertical plane (not cutting
    plane) and in current design working AGAINST tenson of the elastic cord. It
    causes all sort of problems with rubbing top beam, tangling with sled and
    elastic cord. In original design, the weight of slack chain working WITH cord
    tension. The slack part of chain is away of any part of the frame and elastic
    cord just nudge loop into cutting plane. Top feed is much better more elegant
    solution.

in the original version, the weight of the slack part of the chain had a lot
more leverage to twist it to vertical (even with the elastic cord), and that was
causing a the chain to become misaligned with the sprocket. Putting the chains
along the top beam vastly reduces the leverage (but the software supports both

Sorry for too many words ;-( but I hope to make my point. Here are pictures for illustration.

reading via e-mail, I’ll check the pictures out later

IMG_5898
IMG_5897
I solved slack problem with additional pulleys, rope and weights - not elegant and unnecessary complication.

actually, elastic has the problem that it’s producing the most force when you
need it the least, and the least force when you need it the most. weights
produce a constant force, so are much better.

what do you mean by this?

Look at my last picture - I’ve added a pieces of 2x4 on the frame so cutting plane is closer to motor sprockets plane