David, thanks for the response. I’m not quite sure I’m parsing your response correctly. When you say “if the other block is symmetrical”, do you mean the other calibration block or it’s location relative to the first block or both? I was envisioning an identical known dimension block just oriented to index to the opposing corner.
I was trying to get a sense of what the range of non-symmetrical error might be, as that would seem to constrain the ideal size for the calibration blocks. Obviously it could be enormous but practically speaking I’m guessing it’s within a few inches in practice given the nature of the design? And is likely much smaller for a square and well assembled example of the frame?
By checking both left and right sides you would get known absolute positions being detectable using the blocks, and creating that dataset for calibration, wouldn’t you? And if you land somewhere other than expected, you have some offset information to help explain where you thought you would land and where you actually do [that is assuming you land on the block but just not where expected]. Maybe I’m not understanding the optical calibration discussion but it seems to me like we’re in the same church, maybe I’m just sitting in my own pew.
You design them as right angles of known dimensions and you index them to the edges of the workpiece used for calibration. Ideally a piece factory cut square, or squarely on the home goods panel saw (if available) or squared at home if necessary. So with CNC milled squares in the corners of a square sheet, we should be able to set edge positions at least. Center positions are another matter, of course. But how far can we get with multiple data sets with great accuracy from each corner using this method?
-Jeff
[Edit: I don’t know if Ron Olson had a particular reason for using this dimension, but the block I had milled is 3.5" x 3.5" for use on the 4x8 table CNC, for example]
David, thanks for the response. I’m not quite sure I’m parsing your response
correctly. When you say “if the other block is symmetrical”, do you mean the
other calibration block or it’s location relative to the first block or both?
I was envisioning an identical known dimension block just oriented to index to
the opposing corner.
I was meaning location
I was trying to get a sense of what the range of non-symmetrical error might
be, as that would seem to constrain the ideal size for the calibration blocks.
Obviously it could be enormous but practically speaking I’m guessing it’s
within a few inches in practice given the nature of the design? And is likely
much smaller for a square and well assembled example of the frame?
By checking both left and right sides you would get known absolute positions
being detectable using the blocks
only absolute to the accuracy of the location of the blocks
finding the corners is not good enough, you need to find many locations
around the workpiece to map out what the curves are, and since the size and
direction of the curve depend on your particular machine, the system cannot know
how to move and find the blocks.
now, you could manually move it to find the next block and have it measure where
it thinks it is. But how accurately can you position these blocks?
it doesn’t matter if you identify locations by a camera or by detecting a touch
(if you look for posts by me, I describe a way to do this pretty early on)
The issue is having enough points to measure, and knowing where those points are
accurately enough.
your idea of a block at each corner isn’t enough points and they aren’t
positioned accurately enough (because the dimensions of the sheet are not that
accurate)
This works for the shopbot because it has very precise mechanics that allow you
to move in the X direction independently from the Y direction, so you don’t need
to figure out how to move in a straight line, you only need to detect where to
start.
since X and Y are reliable, you can even put blocks on each corner and detect
that you don’t have the board straight and have software compensate for it.
but if your movement down the X direction isn’t a straight line, but is instead
a curve, you need to map that curve, and since that curve may miss a block that
you put beteen it, this isn’t something that can be automated
You design them as right angles of known dimensions and you index them to the
edges of the workpiece used for calibration. Ideally a piece factory cut
square, or squarely on the home goods panel saw (if available) or squared at
home if necessary.
first off, factory cut parts are not known dimensions.
cuts done at the home center or by you may or may not actually be square and are
even less likely to be of known dimensions at this distance.
So with CNC milled squares in the corners of a square
sheet, we should be able to set edge positions at least.
yes you could, but not without you guiding the machine to find the blocks. The
system doesn’t know how to find them until it’s calibrated.
Center positions are
another matter, of course. But how far can we get with multiple data sets
with great accuracy from each corner using this method?
True, but you can measure the material you purchase which is no better or worse than the current measuring method. And while materials can be any size, I have found, quite often, that they are actually dead accurate when I check them in the big box store. No guarantee, I know, but it’s not quite the problem it sometimes is made out to be.
Ok so what if instead of sending it off from block A to block B, we move it by hand (via repositioning through the software) over the second block and have it calibrate against that new block, but at the new corner? And repeat for all four? so now we know where they are in the coordinates of the work area, and depending on how we execute the relocation, assuming the human makes an attempt to get it there as straight horizontally as possible (block A to block B) then we can also learn something about the directional compensation the operator has to do to reach block B, which would help us understand sage in that direction?
True, but you can measure the material you purchase which is no better or
worse than the current measuring method. And while materials can be any size,
I have found, quite often, that they are actually dead accurate when I check
them in the big box store. No guarantee, I know, but it’s not quite the
problem it sometimes is made out to be.
when you are doing press fits of slots and tabs, you need to measure the
thickness of each board, because they vary enough to not fit tightly otherwise.
I just spent over $3k on plywood sheets and they varied enough to affect the fit
in all dimensions.
Ok so what if instead of sending it off from block A to block B, we move it by
hand (via repositioning through the software) over the second block and have
it calibrate against that new block, but at the new corner? And repeat for
all four? so now we know where they are in the coordinates of the work area,
and depending on how we execute the relocation, assuming the human makes an
attempt to get it there as straight horizontally as possible (block A to block
B) then we can also learn something about the directional compensation the
operator has to do to reach block B, which would help us understand sage in
that direction?
the user is going to tell the machine to move X and then correct in Y. Also, the
user isn’t going to care if the actual movement of the router is a straight line
of a curve.
I might be wrong, but if you use the directional buttons to get to the top left block, the machine probably would be able to find the top right block on its own because of the symmetry. It will always take a curved path to get there (the amount of curve being minimized if its well calibrated). Finding the bottom block right block from the top right block, however, likely won’t happen because its not symmetrical in that direction. The user would have to use directional buttons to get it there.
I know cost is always a consideration, and touch sensors are another way to go, but with regards to a more traditional setup block, my first article samples from my hobby project to test out on-demand manufacturing just arrived and they look pretty good. I posted a photo over here: CNC Milling a ShopBot XYZ Finder via Xometry Service - Review and will update with more details in the next day or so.
I think some CNC milled calibration blocks or tools might be closer to being viable than I certainly would have thought a few months ago. I know having a known perfect reference of some type is a challenge we’ve kicked around lots…
