Any guesses as to discrepancy in arm length vs. measured arm length?
-
This might go some way towards explaining why you've had trouble with the sensors also.
It seems this particular printer's geometry might need to be investigated. What do SeeMeCNC and Trick Laser have to say about it?
-
I calibrate at a 0.030 consistently at the moment. The best I could get on stock arms (with a 713maker effector) was 0.040. The best I ever got from Trick laser arms with the stock springs was 0.950 (usually in the 0.120 range, though). When I swapped in weaker springs I could bring it down to a 0.080, but no lower.
My current configuration is a magball setup, with Haydn magball arms, and a custom effector. Trick laser carriages, but custom arm adapters (I've posted about them before, 713maker made them for me, and will sell them).
I've calibrated down to a 0.010 with this setup, but not consistently. A lot of that can be altered with belt tension, and I suspect the carriages shift slightly in the grooves based on how high up I tension the belts. I have some ideas I want to test, but I'm waiting on some new parts before I can do so.
In addition to the deviation, though, check for the "Taco Bowl" shape I've posted about before. Run a calibration at S6 a few times, then do an S-1. It will report the offset at each probed point - you'll find it's high at the towers, low between them.
I've improved on that with my current config, but the oddity persists. I have not yet determined exactly why, but hopefully the new parts I'm going to test will tell me.
-
Its possible to get down to calibration deviation of 0.004 as I am doing so, this is with linear rails, 2040 kossel frame and ally corners, aluminium carriages, aluminium effector, Haydn's mag ball arms/carbon fibre/delrin cups, tooling plate bed. This is if I probe the inner 200mm, if I probe to 300mm it goes up to 0.008.
-
This oddity is specific to a rostock max. I've gone back and forth with dc42 and several others about what's going on, and I still have some suspicions.
I do notice the calibration improves if you only probe the inner 200mm of the bed as well, but still haven't gotten it down to those levels with the actual arm lengths.
If I let it do an S7 calibration, I can get to 0.01 or less, but I know those results are false, as the scaling of objects is incorrect.
-
I will switch out my older 325mm Rostock V2 arms to my new printer and see if that changes anything.
I know the arm lengths have varied on the Rostock Max design from 269mm, 290mm, 300mm, and 325mm across different modifications of it. I'll check my old parts collection and see what sort of arm configurations I can swap out for testing.
What length are the Haydn's mag ball arms? Are they exhibiting similar issues at all to the ball cup design?
I did actually loosen a set of springs with pliers at one point and that didn't seem to change much, but I did not do sufficient testing, only a quick visual check with a print. I also ran a pair of arms though with small rubber bands from hair elastics, and those did not seem to improve anything that I could notice on the edge calibration, but I could take a closer look.
I wonder if the software is suggesting that the arm length needs to be longer? Especially since the errors only come into play when the arms are reaching the farthest points from the middle of the bed.
My latest probe deviation is about 0.02 or less, mostly with taking probe points at 140mm on the three towers, and then doing 6 to 12 points at a 75mm radius around the middle of the bed to rule out any sort of bed tilt.
I'll continue to post any results I come up with and try to spreadsheet calibration information and record specific information for future reference.
-
saffi- Haydns arms are IMO the best delta arms currently available he will make them to length you specify, if I was going to be very picky, but then thats delta owners after all, (mine are 360mm) they are a little too bendy in 6mm carbon rod, 8mm or 10mm carbon tubes would be better.
kraegar - if with your best calibration you then use M556 axis compensation for x and y, can you print correctly scaled objects at any peripheral coordinate as well as centrally? If you can I'd say the calibration is correct. I don't do this, I've tweaked rod lengths manually to give accurately scaled objects centrally printed using 8 factor calibration, and I can print an even-height spiral across the whole 300mm but I haven't tried the peripheral object test as above. Maybe I will. I'm instinctively against using M556 but then everything else on a delta requires computational correction to make them work, its just one more set of calibration parameters to shove in config.g.
-
I didn't see any arms from SeeMeCNC themselves above the current 292mm arms. I thought they used 269 for the old style, and 292mm for the current ball cup ones. Anything else (300, 325) was aftermaket (tricklaser).
My Haydn's arms are 304mm, and do show the effect, but it's lesser. They're magball, not ballcup, and don't require any springs, etc between them.
There was some question if the steps/mm was not truly 80, but several of us tested that and found it was spot on, or at least close enough to know that it's not causing the issue. If you only probe near the center, and not the edges, the issue also vanishes, or if you tell it your arms are longer than they are.
I started a discussion here: http://forum.seemecnc.com/viewtopic.php?f=111&t=11441 but it fizzled out.
-
I'd like to test it with some longer and shorter magball arms, as well, but can't afford to buy them just for testing this.
-
I think the 80 steps versus 80.5 depends on belt tension if tight its not an issue.
I wonder if a suitably rigid mechanism could be devised like a turnbuckle to lengthen/shorten rods as needed?
-
I tested belt tension from what I'd consider fairly loose, to as tight as I'd normally go, and saw far less that a .5mm discrepency in steps/mm. less than 80.1 consistently. Perhaps I'm not going as loose or tight as others might in their tests, but it seemed a good range, and didn't account for the problem. (also, I could measure it, see I was under 80.1, but only fix it by going up to 80.75 or so, which threw my height of objects way off)
-
I also did some more testing yesterday and used a 20x60 extrusion and the fsr sensors to be able to measure two different heights as proposed by dc42.
I also have the issue of crazy wrong rod length proposed by s7 calibration and for me, the measured height difference between measuring the 20mm vs. the 60mm side of the extrusion is 39.98, so I guess I can say that my steps/mm are close to perfect.
What I can do furthermore is replace the duet with a smoothieboard to see if this changes anything and I will also print a larger effector to see if effector size plays a role in this game. Glad that tightening the belts is already out of the equation, one step I do not need to do now. ..
-
I've changed to a much larger effector, didn't make much difference.
I highly suspect that if I went to longer arms (around 325mm) with the magballs, the issue would vanish, or nearly so, even when probing to the edge of the bed.
First I want to rule out the carriage system entirely, which I'll be doing in the not too distant future I hope.
-
Hmm… I have quite long arms, 360mm for a 300mm bed, sorry to tell you I see the issue
-
That's helpful! This is a rostock max? What effector and carriages? What arms?
-
It's a homebrew printer, but I use the same rostock max effector (I liked the design) and 360mm tricklaser rods with ball cups. The carriages run on MGN-12 rails.
The distance between the middles of the white bar-bells on the carriages is approx. 330-335mm
-
I'd be more inclined to rule it out if the arms didn't have springs between them for tension. That's why I want to test with longer magball arms.
I'd be curious to see the output of a S-1 calibration on your printer (after you've done a few rounds of S6).
-
Btw, the reason I'm against the springs between arms (at least as far as diagnosing / testing this issue goes) is simple - they change your calibration. If you put the "heavy" springs on, and calibrated your printer, look at the outputs of M665 / M666 and an S-1 probe… then swap to rubber bands or lighter springs, repeat an S6 (or several, so it "settles), and compare the new outputs of M665 / M666 and an S-1, you'll find things change quite a bit. That means there's no way the springs themselves aren't having a measurable effect on your calibration values.
-
I had another look at data I collected yesterday and collected some more data today.
Yes, I agree, there is an influence of the springs, but I guess it is not very relevant.
What I did yesterday was checking probing speed / motor power vs. deviation of testresults.
So I did a log of G32 runs with S-1 to collect the data. I varied probing speed and current for the motors, I saw a big influence of the motor current with the default probing speed of F120 when choosing 1.5 or 2A current, but with the default of 1A there was not much to see at different probing speeds.
What is relevant in this context is that I measured the deviation of the results of each probing point, if springs have a huge impact that deviation in the points should be higher than the measured 5-7.5 Micrometers, shouldn't they? unless of course that because of the quite repetitive movement pattern the effect of the springs is always the same, but I kind of doubt that.
Today I also did some more measurements with moving the head arround and then probing x0 y0, here the numbers are also quite close:
Probe with moves on Bed Probe with moves on 60mm Probe without Moves 60mm Probe without moves 20mm 0.07 59.914 59.939 19.97 0.077 59.914 59.927 19.977 0.07 59.914 59.933 19.977 0.07 59.895 59.92 19.97 0.064 59.914 59.914 19.97 Avg 0.07 59.91 59.93 19.97 Dev 0.00 0.01 0.01 0.00 -
The effect of heavier sprints is to pull up on one corner of the effector at the point when it's between towers. This is easily observed - if you take a set of barbells off your printer and put your arms on them (with springs in place) you'll find the springs pull things into a square naturally. If you push it into a parallelogram, the springs will pull it back square. The weaker the springs, the less the effect. When your effector is beween towers the arms are in this parallelogram position, and the springs are fighting it. Something has to give. It flexes the outer arm - the longer your arms are, the more they flex, and it pulled up on that corner of your effector.
The effect actually is very repeatable and measurable.
If you build one of the effector tilt measurement probes I made, you can observe this effect in real time, and how much it tilts the effector on your delta. Info on that here: https://www.duet3d.com/forum/thread.php?id=1306
That said, the springs are NOT the cause of this issue, they just muddy it up and amplify it some. I'm running magball arms which don't need linkages or springs, and still see the "taco bowl" shape to my probe output.
Here's a brief list of things I've tried to diagnose this:
Seemecnc stock effector, 713Maker aluminum effector, custom aluminum magball effector
Seemecnc 292mm stock ball cup arms, Trick Laser 300mm CF ball cup arms, Haydn Magball arms (304mm)
Seemecnc Accelerometer probe, dc42 IR probe, Piezo Probe
Seemecnc injection molded carriages, Trick laser aluminum carriages (current version)And then I've also made the electronic tilt measurement, so I could measure and eliminate tilt in my effector to rule that out. (I have no measurable tilt at this point, either digitally with the accel/gryo, or with a bubble level).
I have loads of spreadsheets of data gathered, but have yet to find anything other than increasing the arm length that gets rid of the "taco bowl" shape.
Big clarification on that last statement: Several of these have reduced the issue. Anything that adds tilt to your effector seems to really amplify the incorrect measurement of the arm length. The biggest improvement so far was going to magball instead of arms with springs between them)
-
All, please bear in mind that I do not advise the use of 7 or 9 factor calibration. If you can't resolve the problem by using weaker springs or whatever, then I suggest you use mesh bed compensation to handle the residual deviation.
FWIW, my large Kossel with PCB effector and carriage adapters and Haydn's magnetic arms calibrates to a deviation of about 0.025mm using 8-factor calibration and AFAIR 13 probe points.
