Strange Behavior During Cut

[wcs39204]

Ok so I was fooling around trying to get the MetalMaslow Sled & Z-axis upgrade working for HOURS yesterday. After about 4 hours I gave up and put my original sled back together, and recalibrated. This time I gave it an exact measurement between motors and started cutting my piece. It was on 4x8 3/4" MDF, and it was going beautifully. I did a quick measurement of the part, and it was as accurate as the model. 24 13/16 on the model, 24 13/16 on the part being cut. Great.

It finishes that cut, and goes to the upper left side to cut 2 smaller pieces, and it cuts the first pass, then shifts down about an inch. Cuts a 2nd pass and shifts down another inch. cuts a 3rd pass, and you guessed it. It shifted down another inch. So I stopped the cut as to not ruin the entire piece of wood I was working with. What happened? Did I jump a tooth 3 times? Is this because I was in the upper corner working? Anyone ever have this happen?

Edit: I checked all the obvious things. Sprockets were tight, bearings were on the ring correctly, etc.

[wcs39204]

Forgot to attach photos. Just getting back to my shop now.

See how pretty and clean that cut is? No jagged edges, no glitches during cut, nothing. Just a pretty cut.

And then we have this. Just absolute opposite of the first cut. I’m pretty sure I jumped 3 teeth, looking back. smh

[JWoody18]

I can’t comment on what might have happened to the later cuts but I did want to say - WOW that first cut is clean. And nice photo of it too.

It certainly looks like a perfectly square cut in the photo, is that true if you throw a precison square on it? This is a really great example of what the Maslow does when it is operating optimally.

[bar]

I think you diagnosed the issue correctly as a skipped tooth. There are some diy guides to stop that from happening in the Community Garden. The other key is to make sure the chains stay in plane. If the attachment point is far above or below where they join the motors you will get chain skips.

[wcs39204]

@JWoody18 I put my triangle square on the top right corner (the only one that makes a 90 degree) and it’s dead on. The rest of that cut wasn’t square, because of the slopes and such it was designed with. Maslow is a great tool, you said it, when it’s operating optimally. My problem is, I can’t ever seem to get it to stay that way lol I redid the calibration today, and the ‘move to center’ doesn’t put the sled in the center any more, and the reticule in GC that represents the sled, is way off too. So I guess another couple hours calibrating again…

@bar I’ll look into some of the DIY Guides. The plane is off, that’s for sure. I measured it after reading your reply. It’s 1/2" difference at the motor, than at the sled. I don’t know how to remedy that unless I add a piece of 1/2" plywood in addition to the 3/4" wasteboard I have now. Or modifying my frame, which I honestly was hoping to get around a job of that caliber for the time being.

[JWoody18]

Hmm, I just finished writing a long thread about my journey to get an XYZ zeroing block made for the Shopbot at my makerspace. That has me wondering if such an approach could be a way to speed up and improve calibration so your situation is less likely to happen and easier to recover from.

What if metal four blocks of known dimensions were placed in the corners of the work area and a similar calibration for XYZ was done for each? And then distances could be calculated or interpolated accurately between them? Would that give us a way to get consistent measurements and calibration data so less user input and manual measuring was required? [Edit: hmm since we can both measure and also calculate, would this allow us to, using differences, detect sag areas and auto adjust for chain slack etc?]

Just whiteboarding on-the-fly here but lets say I put a 2"x2" square (stepped for the work piece), 1" thick machined aluminum block (like in my post linked above) in each corner of the frame and squared to the edges of the wasteboard/workpiece… I can do a standard Z depth test to Z at top of work piece, and I can get X and Y by bump testing like in the Ron Olson video in my post. And then with that data triangulate the underlying corner of the work piece or wasteboard. Repeat in each corner and also bump measure between the two inside edges of each block pair, giving me the length between corners

Hmm I think we might only need 1 block or 1 pair of blocks. Having 1 pair allows you to do multiple measures between left and right without moving the block from one corner to the other. Then move the pair to the lower corners and repeat.

Would that give us enough calibration data to at least complete the basic calibration that is done by hand today? I realize more datapoints like in holey calibration should yield even more accuracy, but I think this would get us to a more stable starting baseline than todays manual method, would it not?

It seems it would scale to different workspace configurations as well, as I know not everyone made full 4x8 frames.

Taking it a step further, if you had two pairs of blocks and were working with a square work piece you could potentially calibrate to each work piece before starting, if there is any value in that? I’m just thinking scaling down to the smaller workpiece within the larger frame if that yields any advantage, as long as it is reasonably quick and automated. Or even automatically confirming calibration after a bit change.

Or, taking it to an extreme, and again assuming it was reasonably quick and automated, recalibrating between each depth of cut change - which would also allow throwing a warning or error if something was different from the previous calibration run on the job, to give the opportunity to fix skipped sprockets etc when detected.

@bar am I out to lunch here or do you think this is something that could be a benefit?

-Jeff

[JWoody18]

Adding to the idea pile, instead of having indexing edges for the workpiece, perhaps it makes more sense to design blocks at have a nice true 90 that indexes to the corners of the wasteboard so the principle human actions required are to carefully align the blocks to each corner, secure them with pistol clamps and hit start on the calibration routine.

Hmm, maybe that is also the same thing, or we can design it to support both workpieces (internal corners for 0,0,0) and wasteboards (external corners for 0,0,0).

And following on my block/pair of blocks idea… I think one block can still work if a human can pause the routine to measure between corners and move the block, but it adds a lot of human error potential and interaction and it removes the ability to rapidly do a second or third pass measurement to ensure everything is ok. At a reasonable volume and if the block was kept smallish I suspect two pairs, so a block for each corner, would allow for the most automation/least time for each calibration and potentially really fill a usability gap for the Maslow…

@dlang You’re up. Give me the counter point, please.

-Jeff

PS - It occurs to me that a block design could enable the measurement of four known coordinate pairs in each corner of the wasteboard. The principle one being the outside corner location but also the diagonal inside corner location as well as the two known positions (the other pair of opposing corners) along the horizontal and vertical edges adjacent to the outside corner. That would yield 16 X,Y pairs, albeit only in the corners of the board, but doesn’t that get us a long way towards being able to interpolate other key positions for calibration? It feels like there should be something here we can use, my spidey senses are tingling.

[Joshua]

I think the biggest culpret for skipped teeth is a lack of tension on the slack side. I have had that happen on my machine. That is why I am considering the upgrade to weights, rather than the bungee system.

[wcs39204]

@Joshua I have a spring on mine. in the middle, pulling the slack at all times. It literally never has slack.

I think something went arye in that upper corner, I just don’t know what. I’m usually cutting on 4x4 sheets centered on the machine, so I’ve never really ventured out on the edges of a full sheet before. I just assumed it would work out perfectly. Man did my Maslow have other plans tho haha

[dlang]

there are several factors that can cause errors, and they cause different
amounts of error in different places on the workspace.

so if you know that what you think is 1000,2000 is reall 1123x1933 you can
correct from that point, but you won’t have any idea what the correction should
be 4000,-2000, depending on the source of the error, it may even be in the
other direction.

there are folks working on a project they have called ‘optical calibration’
which uses a camera to find points on a ‘known’ grid and establishing an array
of correction points, interpelating between them. That code could be made just a
little more general (and get more people involved to support it) and be used
with touch sensors.

The big problem as it turns out is to get accurate references. Even when you
have commercial shop print a custom ‘banner’ that’s a grid, they found that it
was off by a significant factor in one direction

David Lang

[dlang]

Adding to the idea pile, instead of having indexing edges for the workpiece,
perhaps it makes more sense to design blocks at have a nice true 90 that
indexes to the corners of the wasteboard so the principle human actions
required are to carefully align the blocks to each corner, secure them with
pistol clamps and hit start on the calibration routine.

until there is a reasonable calibration in place, don’t count on being able to
find the blocks and hit them. with a bad calibration you could be off by inches
and not know how you need to move to hit them

@dlang You’re up. Give me the counter point, please.

there is a simulator in ground control that lets you enter an error (say the
chain is 0.1% longer than you think) and graph what shape you would get if you
tried to cut a grid.

Play around with this and see how different errors can produce similar shapes
(but at different amounts) and how other errors produce different shapes.

Just getting a couple points of error isn’t that good (which is why our standard
calibration routine works as poorly as it does, even though it’s a lot better
than the prior one)

The holey calibration routines take 12 measurements between 6 holes and compare
what it things should be with what is. It’s producing results much better than
our standard routine and we are working to integrate it.

If we could have known locations that we can detect by touch (and a routine to
moveproperly to detect the touch from different directions accurately enough),
that would be far easier to work with than the multiple, precise measurements
(and ho precise do you think your tape measure is, did you know that there are
standards for such things and how most tape measures don’t qualify to get graded
by the standard?)

if we had a lot of touch points, we could use the logic from optical calibration
to make a grid of known corrections and interpolate between those points. But
this is only as good as the accuracy of your ‘known locations’

good ideas here, look into optical calibration, holey calibration, ‘in search of
accurate measurements’ to get up to speed on the various issues (catching up on
the work/discussions we have already had to avoid repeating them)

David Lang

[wcs39204]

Well I got her all calibrated, again. This time, I routed my chain/spring a little differently. Thus keeping a constant tightness around the sprockets. I didn’t get a photo, but I will tomorrow when I’m at my shop. I do believe it will take further tweaking tho, as when I was cutting my test cuts in the final stages of the calibration, the chain rubbed the slack side a bit. I think I’ll add a 2x2 block on top of my top beam, and put the ‘nail’ for the end of the chain link up there, so I can bring my ‘guide’ up a bit more to keep it from rubbing. It only did it in the lower right corner, so I probably wouldn’t have to do anything, but the OCD in me, just won’t allow it to go like that lol. I just took both of those white ‘roller/spacer’ things in the kit and ran a screw through it, close to where the sprocket is, to keep a loop of chain on the sprocket at all times. Something like this (Not my photo, but totally stole the idea from it)

Mind you, I don’t have the pulley set up, because the spring keeps more than enough tension on my slack side. But the way the person ran their chain in this photo, is the same way mine is. And it looks like it would keep any skipping from happening because the chain keeps more than half the sprocket covered with chain instead of just 1/4 of it being used. If that makes sense.

We’ll see! Going to try this cabinet cut again tomorrow to test.

[wcs39204]

Ok so. That seemed to fix my issue. Here is my current setup…

I just used those white roller/guide things that came in my kit. Ran a screw with a washer through it to keep it in place, and have something to roll over… Also a test cut (first time using 1/8" bit) figured why not… haha

The first cut (top) was done with the stock sled. 2nd cut was done with my metal maslow sled. There are still some issues with the sled I ordered, but they’re working with me to get me the right parts for my router!

And both together!

[wcs39204]

And I should mention this, in case anyone was curious as to what is on top of my top beam, holding the nail that holds the chain, it just got 2 6" pieces of 2" x 2" and predrilled 3 holes in it, then used some 2 1/2" deck screws to attach it to my top beam. This allowed me to get the chain high enough out of the way, so my rollers could be moved to where the nail used to be, closer to the top of the beam. Now I don’t have to worry about the chains rubbing when it’s in the far corners of the workspace. Any questions, feel free to ask.

[JWoody18]

@dlang

until there is a reasonable calibration in place, don’t count on being able to
find the blocks and hit them. with a bad calibration you could be off by inches
and not know how you need to move to hit them

I hear you. However, if you start on top of a known block (how exactly we get “on top” of it when the sled is level to the work piece is TBD I realize) and we know the size of the block, we can reliably find it’s edges. Once we have that and we transit across the board to the other corner in a straight line, we can reasonably continue until we touch the block, couldn’t we?

That is to say, let’s say we have touch plates in all four corners, as proposed. And we start in contact with the upper left touch plate (you lower the bit down from above to get your Z and then, because the block size is known, you can fairly accurately move out past the maximum edge distance (from the Z test location) and then move back in to touch and confirm edge locations.

Any reason why we couldn’t slowly progress in what we believe is a straight horizontal line near the top edge until we bump the inside of the upper right touch block (whose size is also known)? Then rinse and repeat heading down vertically and then back across horizontally and then back up vertically? Have there been calibration situations where an expected horizontal lines travels materially diagonally across the workpiece over 8’?

If we could do this, it would allow collecting data points at each known dimension block and we would be able to impute the arrival landing position after we grab our X,Y with the sensor on arrival (and therefore assess the accuracy of the real path traveled relative to expected path traveled between the previous and current block)?

Pardon the low res video quality, but I’m thinking something like this video from the link I included above about my XYZ block for the Shopbot. The Youtube video from the originator is here: https://www.youtube.com/watch?v=f4mFNMzBV2k

It just strikes me that as long as the user has accurate dimensions for their wasteboard or workpiece, it would be great if we could have the system measure the movement through sensors and do that portion of the calibration itself.

It also occurs to me that having basic sensor blocks in the design may be able to help in other areas of initial setup and calibration when deployed in alternate locations or orientations. E.g. the block could be sized so combining more than one yields ideal spacer sizes for various things during frame construction, and likewise deployed for initial chain or other setup. I don’t have all the answers, just pulling a thread to see where it might go…

-Jeff

[JWoody18]

@dlang

The big problem as it turns out is to get accurate references. Even when you
have commercial shop print a custom ‘banner’ that’s a grid, they found that it
was off by a significant factor in one direction

Right, this is where CNC milled aluminum calibration blocks can come in. It seems like the missing piece. The blocks would ensure a known reference piece in the end users hands that could be built on programatically to collect the data points and interpolate.

-Jeff

[JWoody18]

@wcs39204

Well I got her all calibrated, again.

Out of curiosity, how long did it take you to run through the calibration after your changes and now that you knew what to expect?

-Jeff

[wcs39204]

If I had to guess… Actual time working on the calibration… 45 minutes give or take. To me the longest part of the calibration is the chains. Not because it’s difficult, but because they feed out so dang slow haha But, I’ve learned, when you’re at the Extend Chain portion, for my machine, I change the value to 3030mm and attach a hammer with a piece of coat hanger to the end of the chain that hooks to the right motor, so I don’t have to babysit it for 3030mm of slow moving action.

[dlang]

no, because you do not know how to move in a straight line. If the other block
is symmetrical, then you have a pretty good shot at finding it, but if it’s
somewhere in between, you may be moving on a curve.

if the chains match, and the top beam is truely horizontal, the movement will be
completely symmetrical and you will get no value from checking both left and
right sides of the machine. If the machine isn’t symmetrical, then you don’t
know how to find the other block because what you think is a horizontal movement
isn’t.

David Lang

[dlang]

the problem is in positioning these blocks.

If you have a block that is 6’ wide, that’s one thing. but if you have two
blocks that are 1" square, how do you position them exactly 6’ apart? (with
sub-mm accuracy)

David Lang