You can’t guarantee everyone will have a top beam so may as well make sure it works with the initial frame design as well.
2 Likes
I am not sure if you guys have already discussed this, but I am going to ask anyways. Do you know what the distribution of inaccuracy is along the chain’s length? For example, if the total chain length measures 2 mm too long, and you measured half the chain length. Would you find that it is 1 mm too long, or would you find that it is 2 mm short? I am just asking the question of how link-to-link variation manifests its self along the chain length. This might come into play as Maslows get used and worn. I know bike chains stretch as they are used. In the case of the Maslow, the sections of the chain that are used more will stretch more. This could lead to a portion of the chain that is longer than the other.
Further, in reading the post, I thought of an additional measurement, or group thereof, that could be meaningful. This is just an idea. If you attach the left and right chains together with a connecting link, you could feed 12 in from the left motor, and pull the right motor tight. You would find that right motor pulls in 12 in, minus 1 mm, for example. That means first foot of the left chain is 1 mm longer than the last foot of the right chain. You could feed out another foot (2 feet) from the left chain, and pull the right chain tight again; record how much chain the right motor took in. Continue this across the length of the top-beam. Take measurements with the left chain length at 1 ft, 2 ft, 3 ft, 4 ft, all the way across. As I write, I think it is a good idea, for two reasons. First, you get a whole lot of data about how the chain has stretched. Second, it is easier than manually wrapping, and removing each chain from its respective sprocket, just to have to replace them at the end. It makes a whole lot of sense, and is really easy, to, at any point in time, connect the two chains together and run this measurement.
Let me know your thoughts.
1 Like
I would not expect the beam to flex (especially if the motor mounts are directly
on the beam), but there have been cases where the screws don’t hold the mounts
solidly enough and the mounts move slightly under tension.
Do you know what the distribution of inaccuracy is along the chain’s length?
We do not, we are assuming that it’s roughly linear
Further, in reading the post, I thought of an additional measurement, or group
thereof, that could be meaningful. This is just an idea. If you attach the
left and right chains together with a connecting link, you could feed 12 in
from the left motor, and pull the right motor tight. You would find that
right motor pulls in 12 in, minus 1 mm, for example. That means first foot of
the left chain is 1 mm longer than the last foot of the right chain. You
could feed out another foot (2 feet) from the left chain, and pull the right
chain tight again; record how much chain the right motor took in. Continue
this across the length of the top-beam. Take measurements with the left chain
length at 1 ft, 2 ft, 3 ft, 4 ft, all the way across. As I write, I think it
is a good idea, for two reasons. First, you get a whole lot of data about how
the chain has stretched. Second, it is easier than manually wrapping, and
removing each chain from its respective sprocket, just to have to replace them
at the end. It makes a whole lot of sense, and is really easy, to, at any
point in time, connect the two chains together and run this measurement.
there is a fair amount of effort attaching and detaching the master links to
connect the two chains.
We may need to map out the chains in detail like this, but let’s see how good we
can get assuming linear error.
David Lang
2 Likes
Do we need to go that far? Could we connect the chains with a twist tie or a piece of wire?
I’d like to see someone try @Joshua’s suggestion and determine if non-linearity is an issue with a new chain, or if the multiple measurements results in a more accurate calibration. I am unfortunately still a few weeks from starting our build.
Do we need to go that far? Could we connect the chains with a twist tie or a piece of wire?
not if we are going to be pulling on them to see how much they stretch compared
to the theoretical 1/4" per link
I’d like to see someone try @Joshua’s suggestion and determine if
non-linearity is an issue with a new chain, or if the multiple measurements
results in a more accurate calibration. I am unfortunately still a few weeks
from starting our build.
It’s worth someone who has some questionable chain doing, I’m not sure it’s
worth trying to make standard until we see how accurate we can get without
having to have everyone do this (this is also something that will change with
use as the chains wear, so I would not be surprised to find that it needs to be
done eventually)
1 Like
OK, using v1.12 of firmware and GC, I have tried to use the motor spacing routine and the chain tolerance adjustments to get my chains to match the Man measurement of the motor spacing…
My tape measure tells me (Measuring the spacing of the gearboxes and subtracting the spindle offsets) that the spacing is 3005.142 mm. The Maslow measures 3001.8 for the Left, and 2997.15 for the Right.
The Maslow measurement error for the Left is 3005.142 - 3001.8 = 3.342, and I calculated the error percentage by dividing 3.342 by the actual measurement 3005.142, getting .0011209, and entered the number .11209 (because it is a percentage) into the left chain tolerance adjustment. Did the same thing for the Right, and entered .265944 (Note that the Left calibration line needs to be used to measure the Right chain)
Using these figures and running the motor measurement routines, a correction occurred, but did not match my measurement. From there, I used trial and error to get the Maslow to get close to the actual measurement. This took several attempts… Note that the machine never gives you the same value twice, but usually comes close. The best I could do before losing patience was a correction value of .105 on the left, which yielded a Maslow chain length of 3005.59, and a right correction value of .22, which gave a Maslow reading of 3005.34
I then calibrated the machine using these values and my manual measurement for the motor spacing, running the triangular cut sequence, entering those values, and then running t he Benchmark pattern.
Here is the result of the Benchmark…
And here is my latest .ini file…
groundcontrol_4-28.txt (1.5 KB)
No circles were cut, but, based on the squares, they will probably be slightly wide by a little short. What do I do next?
Thanks!
1 Like
That looks like a considerable improvement! Thank you for documenting this.
I would cut something out for enjoyment.
I was stressing out over 1/16 but when you see it cut out one of your designs, you don’t really care about that 1/16…for a few mintues.
Alright, back to benchmarks for me.
6 Likes
I totally agree with this! I’m still using the original (non-top bar) frame design and angle bracket attachment points on the sled (no triangular kinematics here). All the new frame designs and discussions were happening when I was building my machine and the rings were just prototypes. I wanted to get started and I figured that since Bar had built so much cool stuff with the original setup I just need to get started making stuff. I haven’t slowed down enough to build a new frame or install my ring kit. But it’s on the list!
3 Likes
what were were the calibration numbers you ended up with?
looking at your .ini file, you still have values in there for the quadralateral kinematics, we’ve seen cases in the past where erasing the file and starting from scratch cleared problems that were there if the old kinematics values were in there.
What is the best way to be sure that I start with a clean slate and be sure all old stuff is gone?
Thanks David!
remove the .ini file and you are starting from scratch.
set thekinematics to triangular as soon as you are given a choice, and if you
are asked for distance between mounts or something like that, start over
Yesterday I removed the .ini file, worked to get the chain tolerances more precise, and went through the entire calibration through to the Benchmark test, and here are the results. BTW, I bought a Metric tape measure, so there are less decimals in my measurements due to eliminating the inch to metric conversions. Here is the latest chart, 4/29…
and here is the latest .ini file…
groundcontrol_4-29.txt (1.5 KB)
and here is the previous one, 4/28…
Although the scores are better on 4/28, the horizontal numbers are better on 4/29, with the verticals are still not great (re: the topic of this post, out of round circles)… I am positive that I am pretty much in the same place, except with really good Horizontals.
Too much time has been spent messing with this with no real improvement, so I followed @ScrumdyBum 's advice, and started an actual project. If I keep all parts oriented the same way, the stuff should go together, maybe. I will find out later today.
But, if I am going to use this thing to make things like cabinets, this problem needs to be solved.
Thanks for all your attention… the Forum is an amazing resource, and the Maslow is a real marvel.
4 Likes
I’m seeing scores of 3.16-1 and 3.44-.945…I might just be reading the chart wrong, are you putting the score at the top there?
While you are working on projects I’m going to try to get to where my machine can consistently beat those scores then let’s check in again. I think that you had to use a guess and check method to make the chain compensation work the way we want it too is a sign that is a good place for me to start.
No, the scores just happen up there.
FWIW, I noticed that with v1.12 the rotational radius is computing smaller… The value I started with was 139.8, but the final calculation was 131.9, much smaller that I have seen in the past.
Thanks!
1 Like
I’m digging into this so the first thing I did was to establish a baseline for where things stood so we can tell if we’re improving.
I ran a calibration test pattern for my machine just as it was standing without touching anything so it had a pretty dull bit in it and hadn’t been calibrated in 1-2 months.
I got a score of 3.5-1.5 which is similar to what you have been seeing.
Then I wanted to give the current calibration procedure a test. I put a nice sharp bit in and did the calibration process, using all the automatically generated numbers. I found a distance between the motors of 3009.28mm and a rotation radius of 137.9mm.
I did the measure the distance between the motors process twice from start to finish to see how consistent it was and found 3009.71 on the second take.
After finishing the calibration process I immediately ran the test pattern and got a score of 0.96-0.71 which is pretty good. That already seems better than we’ve seen earlier in this thread which is something we’ll need to look into.
Here are all my measurements:
| Long | Short | Long Error | Short Error | Long Score | Short Score |
|---|---|---|---|---|---|
| 901 | 101.25 | 1 | 1.25 | 0.9583333333 | 0.71 |
| 902 | 99.5 | 2 | 0.5 | ||
| 901 | 100.23 | 1 | 0.23 | ||
| 1904.5 | 98.88 | 0.5 | 1.12 | ||
| 1904.5 | 98.97 | 0.5 | 1.03 | ||
| 1905.75 | 98.89 | 0.75 | 1.11 | ||
| 100.01 | 0.01 | ||||
| 100.43 | 0.43 |
Now I am gong to work on computing my chain correction factors and adding them to see how they fit in. I manually measure my motors to be spaced 3057mm edge to edge so 3057-40.4 = 3016.6mm as the true spacing
3 Likes
Wow, thats nice to see. Once you finish with the chain corrections I will run the same process you did and see what I come up with.
Same chain?
1 Like
Yes, I just wanted to see how repeatable the process was given the same starting conditions. I think the biggest source of error is how well I can set one tooth to 12 by eye, but having the numbers be that close makes me think that the next thing we need to focus on is the chain pitch
Edit:
This might take me a day or two. I’m going to try to automate the process as I go
4 Likes