@dc42 I've forked and updated the github repo. You can see the code changes at: https://github.com/MZachmann/RepRapFirmware/tree/v3.2-screwmap

and it forks from the 3.02-dev branch (3.2 retail).

Let me know where you'd like to go from here. Implement (pull request)? Move to M-codes? Change syntax?

The only thing I couldn't yet figure out from the code base was how to correctly have the ui update the user position when the screw map definition changes.

Mark

This isn't really a bug but an annoyance that has now hit me a few times.

I have error messages not timing out, so I can leave, do stuff, then come back and see what happened. What this means is that I have to manually remove error messages and when I do that by clicking the close X button on my touchscreen I miss the X sometimes and instead it runs whatever random macro is underneath my finger because the macros display in the same screen area as the error messages.

Is there any way something there could be changed? Running a random macro when I want to close an error message isn't great.

Thanks. Mark

I'm going to keep working on this, and here's why... I'm including 4 photos.

Two of them show my current CNC X axis and the camera, to understand the context. The other two show a scan of the camera looking at a ruler as I move the CNC manually in inch increments.

Looking at the problems, what is happening is that the X rails are deformed slightly in the center. The X dimension error increases then decreases (and then increases near the right edge). And .. at maximum X error there's also a Y error that's significant.

Using my last controller (WinCNC) I automatically created a multi-axis screwmap using the camera (which takes under 15 minutes) and the positioning error dropped from a max of .03" to .003".

The mapping seems to be working quite well.

Here's a video clip showing it in use : https://youtu.be/uTqUBrUHinU

I'll write it up in the next week or so. Note that with minor mods this could do backlash compensation.
Here's the input file:

Here's the script I run to set the screwmap

Here's the result:

Mark

What seems like a teaching moment...

I reworked my IR sensor carrier to move the sensor closer to the hot-end and while I was at it I made the sensor board edges more orthogonal to the bed. This seems to produce far more consistent results and noticeably more range than the last less-careful mounting. (The first iteration had oval holes since I wasn't sure where it would end up)

I'm thinking about replacing my current CNC controller board (wincnc) with a BTT Skr Pro running RRF. The CNC has high-voltage external stepper drivers that I want to retain.

Questions--

1. Does the SKR Pro support external stepper drivers (i.e. step/dir pin output) ?
2. Does RRF support this kind of usage?
3. Is there a better board I should look into. I'm using a Duet3 in my 3d printer and love it but it doesn't support external steppers.

Thanks,
Mark

I've posted a video showing a calibration run (in X). It took about 1/2 hour to gather X and Y error terms and improved the CNC accuracy from .033" to .002".

here: https://youtu.be/wpUOaYwDrvI

Postscript:

I tried pasting four small squares of 25lb paper glued down with Elmer's glue-all (easy to remove with water) at the corners. They worked great... very reliable and consistent.

The problem was that they also read 0.8mm because they rested on top of the transparent PEI sheet -- one day I'd forget something, do a G30 at a non-paper spot, and the nozzle would then launch 0.8mm into the glass bed. It's possible I could put them under the PEI but at the moment - not.

Instead I cleaned those 4 square corner areas on my PEI sheet. They had never been used for printing and I avoid them so they're flat and transparent - nearly as good as a paper square for a 4-point calibration. I determined this by running a mesh, flipping the glass plate over to solid glass and running another then comparing the results.

@william3d Feel free to ignore this if too basic but here's my 2c ->

When I was having strange issues with my CoreXY the first thing I did was check simple movement. [I did try to use homing as an early test, immediately drove the nozzle into the bed and thought hmmm...]

Homing is an extreme test because it causes parts to hit other parts and limit out.

Instead, I placed the head in the center of the bed and turned off homing requirements then used the dashboard to move the carriage left and right and back and forth 50mm at a shot. Make sure only the dimension you expect to move actually changes when you do that.

Once moving straight was true everything else worked.

Mark

Here's a doc file describing the current implementation of screw mapping.

Screw Mapping in RRF.docx

I've cleaned up the code so setting a screw map entry changes the DRO readout in real-time.

Here are a couple of plots of my CNC error curves with increment=25.4mm (1 inch). Note the 2d-ishness of the whole thing.

To put this in a worst case if I were just drilling holes the largest X error pre-compensation would be about 1.5mm (or 1/16th of an inch for US). Post-compensation error is 2*repeatability + test measurement error (~.03mm).

@bot Yeah, I wish I had a metric one - if you watch the video I spend a lot of time converting inches. Kinda dumb.

backstory - when I first realized about 10 years ago my CNC had so much error I did this measurement with a normal yardstick, couldn't believe the results (especially the Y error), so in order to root-cause I had no choice.

@bot Yeah, Starrett would probably not be happy me calling it that.

Sorry, I should have posted that. You're absolutely right.

The yardstick is one of the most expensive hand tools I own since it's so important.

https://www.mscdirect.com/product/details/58700246

I've posted a video showing a calibration run (in X). It took about 1/2 hour to gather X and Y error terms and improved the CNC accuracy from .033" to .002".

here: https://youtu.be/wpUOaYwDrvI

Here's a doc file describing the current implementation of screw mapping.

Screw Mapping in RRF.docx

@dc42 I've forked and updated the github repo. You can see the code changes at: https://github.com/MZachmann/RepRapFirmware/tree/v3.2-screwmap

and it forks from the 3.02-dev branch (3.2 retail).

Let me know where you'd like to go from here. Implement (pull request)? Move to M-codes? Change syntax?

The only thing I couldn't yet figure out from the code base was how to correctly have the ui update the user position when the screw map definition changes.

Mark

@dc42 Thanks for all the interest. I just ran it quickly doing the following...

The worst axis is X so I copied the old lookup map from WinCNC to excel, converted to mm, then created a script file... which was too big. So, I added the O option and redid it. This worked.

I set the cnc to 116,0 (left of table), set down my yardstick horizontally.. then I disabled screw map, went to 15",0 - a known bad area) took a screen capture of the microscope view, return to 0,0 and enabled screw map then went back to (15",0) and took another screen capture.

I hadn't tried it before - and - first try with no edits. Here's the script and the comparison photo. I know it's a lot of digits, but I'm used to inches. Probably could shrink it a little.

M801 R"X" A"XY" S116 I25.4 N36
M802 R"X" X0.00000:-0.04064:0.00000:0.06350:-0.07112:-0.10922:0.02286:-0.04318:0.00508:0.10414:-0.02794:-0.09652:-0.22352:-0.40894:-0.37846:-0.47752
M802 R"X" Y0.00000:0.01778:0.06858:0.11176:0.14732:0.18034:0.22606:0.29210:0.34544:0.36576:0.40640:0.43942:0.46228:0.51308:0.53594:0.57658
M802 R"X" O16 X-0.49784:-0.27178:-0.25908:-0.29718:-0.15240:-0.24638:-0.26162:-0.05080:-0.08636:0.01524:0.14224:-0.00508:0.05588:0.09144:0.07366:0.24130
M802 R"X" O16 Y0.60960:0.59944:0.55372:0.48514:0.40640:0.30480:0.22606:0.19304:0.15748:0.09144:-0.21082:-0.48768:-0.15748:-0.12954:-0.10160:-0.08890
M802 R"X" O32 X0.25654:0.25146:0.41656:0.78994
M802 R"X" O32 Y-0.12446:-0.09652:-0.04572:0.01270
M800 S1

Photo

Ok, even this M802 line is too long. There seems to be a 200 character upper limit and mine are over 300. I guess I should allow an offset in the M802 command.

Sure, here's the tentative syntax (from the code comment). I randomly picked the M80x's just to be easy to remember/type during debug.

--Enable/disable all screw mapping. Does not clear tables.
-- if no S prints report
M800 S[0|1]

-- create a screw table for srcaxis with one row per destaxis
-- if no arguments it prints a summary report
M801 R"srcaxis" A"destaxes" Sstart Iinterval Ncount

-- set screw table entries
may be called multiple times (once per row) for shorter gcode lines
-- if no data (rows) it prints contents of table
-- each entry is an offset (delta) value
M802 R"srcaxis" Xf0:f1:f2:f3... Yf0:f1:f2...
-- run the screwmap selftest (ifdef'ed just for debug testing)
M810

I considered condensing the Mcode but the destaxes is a list of possible axes that are adjusted by the source axis. So, the M802 syntax can be very very long.

On my machine the X axis has 3 rows of 36 floats and the Y axis has 3 rows of 48 floats. So the M802 is a very long list of numbers. I'm not sure what the gcode limit is but this would be about 3000 characters.

I let the M802 to be called repeatedly to set rows individually for clarity.

--Mark

postscript: I'm obviously not wedded to syntax. I'd just like the script file to be readable.

There are two sets of 'upgrades' possible to the functionality for later syntax thoughts. I don't use that functionality on the CNC.

This can support directional tables (which I don't use but some milling folks do) to support directional leadscrew wear. It can also support backlash as a 1 column directional screw map.

postscript 2: in practice this works well in a script. Create a table, fill the values, create a table, fill ... The only odd error checking is to ensure the table data list is <= the defined length and the result values are monotonic. It would be easy enough to have M801 do both tasks depending on parameters but the interdependencies are worse and I think it's less clear.