Calibration Process Brainstorming

I am curious to see if this would work. At some point I may try to modify the calibration algorithm to use short cuts that are measurable by a caliper on the work area and see if the results end up similar to our current method. I think it’s worth trying!

That is not true. You would know the absolute length of the parts, to the same precision as with any other calibration procedure. However, there is additional information of a highly precise measurement of the difference between the lengths.

Good point here. So my question then becomes, what are you trying to achieve overall? If your calibration routine is based on the absolute lengths then the caliper measurements would have limited benefit, as the initial length measurement of the first piece would likely still be made with some lower accuracy device, typically a tape measure. So while the second piece would be measured very accurately in relation to the first, it would still have all the error of measurement from the first piece being measured as well.

If your goal is to make the measurement off the machine, say on a workbench, then cutting out the calibration pieces seems to be the critical step.

If your goal is to be able to make short measurements with just a caliper and no tape measure, then the calibration algorithm would need to be modified and the math determined to see if it could work.

The goal is to cuts parts the same length, regardless of the position from which they are cut. The absolute scale is less important.

Imagine cutting parts out to build a boat that is supposed to be 18’ long. If the absolute scale of your machine is off slightly, but all the parts are off by the same amount, you might end up with a boat that is 18’ 1" long, off by 1 inch. That is probably OK.

Imagine, on the other hand, that you cut the parts for your boat. The parts for the left half are cut a little small, and the parts for the right half are cut a little long. The left half ends up being 17’ 11.5", and the right half ends up being 18’ 0.5". It would likely require quite a bit of sanding to make it look good.

There are similar use cases in which tabs are cut at the edge of plywood panels. If all tabs are are scaled slightly, it is ok. They still fit together tightly. However, if one tab is a little big and the other is a little small, it doesn’t work.

Imagine trying to quantify chain sag. To do so, you cut two pieces the same length oriented diagonally across the plywood sheet, such that the top piece extends from the top right corner to the middle, and the bottom piece extends from the middle to the bottom left. Two pieces oriented straight across the diagonal, each spanning a little less than half the length of the diagonal. One would expect the bottom piece to be slightly shorter than the top, due to chain sag. The magnitude of chain sag can be known by measuring how much shorter the bottom piece is. By laying the two pieces side-by-side, this difference can be measured very precisely.

I understand your reasoning, but I think it is context dependent. Your example of a boat is a great one - if all pieces are slightly short or long by the same amount, it likely would work. But I think the goal should be to achieve the highest possible absolute accuracy. Say I was trying to cut an insert to adapt my 2.5" shop vac hose to my router 1.25" (I think) port. If the machine cuts both holes slightly large then the relative ratio would still be maintained, but the part wouldn’t work. The other concern is that the errors on that Maslow are not linear.

I think I get the essence of what you are going for though. If I’m off, please let me know!

The current calibration routine is based on absolute length, rather than relative length. So let’s say we have those two pieces. You have two measuring tools: a tape measure with an accuracy of +/- 0.5mm, and a caliper with an accuracy of say +/0.01mm. You want to know both piece’s absolute length to the greatest accuracy possible. So you measure the first piece, and have your measurement with accuracy +/- 0.5mm. You have two options for the second measurement: use the calipers to compare to the first piece, or use the tape measure. If you use the calipers, you will add the error of the calipers to the length and have a resulting accuracy of +/- 0.51mm. If you use the tape measure, you’ll get an accuracy of +/- 0.5mm. So in this case, it theoretically is better to use the tape measure, but in reality makes little difference.

Now, if the calibration algorithm is changed to rely on relative lengths rather than absolute lengths, this all changes completely. In this case, your calipers would give an accuracy of +/- 0.01mm, whereas if you used a tape measure to measure both pieces then found the difference the accuracy would be +/- 1.0mm. Huge win here for the calipers.

I get the impression this is what you are looking at, and indeed it would provide a much better measurement. But the current algorithm doesn’t use that data. It may be worthwhile changing the algorithm so it does.

1 Like

I’m sort of new to this discussion. My thought and I may be off here. Run calibration then- why not mark a 20 inch square using inside measurement and re-enter the real value to find the needed off set. 12 inches should be a large enough sample. I chose 20 inches to allow the center to cut a sled if it’s accurate enough.

my 2 cents

Thank you

It’s worth trying. Can you try to lay out the math that would show the actual
lengths?

what is the advantage of making these diagnals instead of just straight across
the top/bottom?

1 Like

This! It’s a great discussion, and I definitely think it has merit. I’m new to this project as well, and have already had several occasions of something I think should work, but it doesn’t, and something I think shouldn’t work, does. So I’ve found sometimes the best way is just try it out!

The existing calibration is doing much larger distances.

what would you be marking off and trying to measure on the 20 inch square?

Thanks. That is all I was hoping for.

@Bee, I don’t understand.

I can try. What/where is the most consolidated version of the math, its most current form?

On the topic of not wasting good pieces of wood, could the calibration benchmark test be made so that it actually results in cutting out something useful?

It would need to be something that didn’t need to be too accurate but could be easily measured to work out the errors. Being built into GC it could also function as a great first project allowing those of us new to CNC to make something before we have even got the hang of CAM software.

2 Likes

you aren’t the first to suggest this, but nobody has a good suggesion on what
would be a good project.

1 Like

What about the Maslow logo or some derivative of it? Another idea would be some of the small pieces needed for the final frame (those triangular pieces don’t need to be too exact as they are just to weld the 2x4s together).

2 Saw Horses is what every woodworker needs and also makes a good flatpack table :grin:
Simple saw horse
Edit: What started as a joke did not leave me. Playing with the layout it might have ‘good enough’ reference points.

6 Likes

I think it would be fun to have something that’s a playful jab at the 3d printer “benchy” - like a big shark.

How about we include, as part of the purchase list, some bailing twine and duct tape. After the parts are cut, use the aforementioned hardware to creatively assemble something to the user’s liking. Take a picture and post it in the forum. Every year, we hole a vote to elect the best.

1 Like

I did a bit of brainstorming to further this idea. I put some code together (in Matlab, not Python) to prove the concept. It works using ideal inputs.

These are the most fundamental changes:

Use a numerical technique, rather than analytical techniques, to solve for the unknowns. The reason is; with analytical techniques, you need one measurement per unknown, and there is no mechanism to use any more information than that. With a numerical technique, more measurements means more accuracy.

Cut long, skinny parts from the work-surface. Ideally, they would all be the same length, but the small differences between their lengths is the foundation for the calibration. Below is a proposal.

Cut numbers into each part, to avoid confusion.

Measure the first part’s length. Call this L_1.
Measure the difference in length between L_1 and the other parts. Call these D_(2-9)
When the cutting head is located in the middle of the end of each part, record the pre-calibrated Left and Right chain lengths. There will be 4 lengths recorded for each part, one for each end, for both the left and right chains.

The difference between measured and calculated numbers (L_1 and D_(2-9)) are used as errors in an optimizer. Use the optimizer to minimize the measured to calculated errors. The D_(2-9) errors are weighted more heavily, because they are known with more certainty.
To calculate the measured-calculated error, use a forward translation that translates the chain lengths and unknown variables into X and Y coordinates. The calculated part lengths are the geometric distances between the points, and the D_(2-9) differences are calculated as difference from L_1.

There are optimizer libraries available in Python, so this shouldn’t be a limitation. The Levenberg-Marquardt algorithm is well suited for this problem.

I went through a simple , digital, example of the above process. In the example, the optimized parameters were a length offset for each chain, and the distance between the two motors. The process was capable of calculating the unknowns within a very tight numerical precision.

To be continued…

2 Likes

yes, cut part(s) and take a measurement and compare the theoretical size to the real size is exactly what I was talking about. From that you should be able to calculate an offset, rinse repeat until you get same results for real and computed lengths. I’ve seen this done on many 3D printers.

I like where you are going.

Thank you

1 Like

I know this is a concept and the picture may not be truly representative of your concept, but we need to be considerate of the material needed to calibrate. Is it a one-step process (i.e., cut out the shapes, measure, and the machine calibrates from that and you are good to go) or is it iterative (cut out shapes, measure, calibrate, repeat until delta is 0?) If iterative, I suggest finding some way to use one sheet of plywood at the most. The current calibration makes small cuts in the corners and the sheet can be reused for many other purposes. I generally avoid the relative extremes where the it does the cuts.

1 Like

@madgrizzle, thanks for the thought. I agree. Cutting an entire piece of plywood is excessive, and could get expensive in a hurry.

Here is the solution I propose. If the cut schedule is solidified, cutt recesses into the plywood face of the Maslow. Small pieces of solid lumber or plywood, say 1x4, could be cut slightly oversize and screwed into the recesses. This way, the calibration only cuts the small pieces of lumber, rather than destroying an entire sheet of plywood. In other words, purposefully design recesses into the face of the maslow These recesses hold the small pieces of lumber that are cut in the calibration process.

I hope this makes sense.

I understand this complicates things. It is worth discussing, because it could completely invalidate the idea.