Unsolved Mystery. Weird Z0 behavior around build perimeter.

[626Pilot]

I printed a statue (about 5x6" base by 7" tall) with my firmware at an 80 micron layer height. It came out fine. I had to print from the SD card because if I use Repetier Host, it tends to reset the connection after a few hours. Total print time was ~22 hours. I printed the statue twice and several other things that take awhlie, and it never failed, so long as I stuck to using the SD card.

I also did one of the Onyx calibration prints (big circle with spokes). The print is at 80% of its normal size because the calibration was done with a probe that couldn't be mounted at effector center. (My firmware supports probe offsets, and they do work mechanically, but it introduces error to do it this way, and you don't get nearly the same quality of calibration.) It's calibrated thinking the probe is perfectly centered, but actually the probe was out at the edge of the effector. This does throw it off a little, but I still get far better 1st layer performance. Without the calibration, what you see here would be lifting off in some places, and dug in to the point of blocking the nozzle in other places.

[bot]

626, have you altered your steps/mm settings for your individual towers? That's what I was asking to see a print of.

[626Pilot]

No, I haven't tried that. Even with the probe offset issue, the calibration I have now is more than adequate.

[0110-m-p]

0110-m-p, have you printed anything tall? I'd like to see if there are any weird artifacts from the steps/mm adjustment.

I haven't yet. Hoping to tonight.

[0110-m-p]

It looks like I may have halfway jumped the gun on the motor steps/mm fixing bed flatness. Well technically it does...but after a lot of testing and measuring it does change the size of parts. It is a very small amount on small parts, but large parts (specifically tall parts) see it more significantly.

Testing with a 150mm tall single wall, 0-infill cylinder yielded a 151.5mm tall object with 81 steps/mm (So approx. 1% too tall). Changing back to 80 steps/mm yielded a 149.7mm tall object. I am retesting with 80.1875 steps/mm and expect the part to be just a hair over 150mm (interpolation between my two data points says it should be 150.04mm tall).

I will have to retest bed flatness and tune printer radius after this, but based on my all my data I would expect my bed to still be low by 0.004-0.005" between the towers .

[geneb]

You really don't want to mess with the steps per mm. It's not something that's ever really variable. Given a specific diameter pulley, that figure is going to be constant.

g.

[0110-m-p]

You really don't want to mess with the steps per mm. It's not something that's ever really variable. Given a specific diameter pulley, that figure is going to be constant.

g.

Agreed generally, but nothing is perfect. If the pulley's are machined slightly oversized the steps/mm would be forced to be slightly larger. That is why it looks like I'm still going to have to change steps/mm a bit past the standard 80 that is recommended.

The problem is I couldn't think of any other parameter that I hadn't already changed to get the bed level between the towers. When I changed this parameter and all of a sudden I could level my bed to 25-micron all around I was blown away. Sometimes things are just too good to be true though.

[rpress]

The pulley does not really affect the steps/mm, unless it has a different number of teeth. It's the belt, but I'd guess that is made okay. So I'd say leave that alone.

Changing the steps/mm had the same effect on flatness as changing both the delta radius and arm length the same percent (but opposite direction to your steps/mm change).

[0110-m-p]

Changing the steps/mm had the same effect on flatness as changing both the delta radius and arm length the same percent (but opposite direction to your steps/mm change).

Interesting. I would get the delta radius to be perfect (zero at center and X,Y,Z) and changing arm length didn't really improve my flatness much at A,B, C. How much did you have to change arm length to achieve this?

[rpress]

Changing the steps/mm had the same effect on flatness as changing both the delta radius and arm length the same percent (but opposite direction to your steps/mm change).

Interesting. I would get the delta radius to be perfect (zero at center and X,Y,Z) and changing arm length didn't really improve my flatness much at A,B, C. How much did you have to change arm length to achieve this?

I increased the arm length about 4mm from what I measured using my cheap calipers.

If you think about what the printer is doing, every movement is relative to the steps/mm. Essentially the belt is what makes 80 steps == 1mm. So rather than change that, you change everything else.

It's still possible the belt is wrong. If you have a precision ruler/calipers you can count the teeth of the belt over some long distance.

[0110-m-p]

I increased the arm length about 4mm from what I measured using my cheap calipers.

Did you verify that your printer is still printing objects with the correct X/Y dimensions after increasing the arm length so much more than what you measured?

From my experiences and what I have read is that if you increase the arm length in the firmware/EEPROM then the controller won't move the motor as far for the same movement which would result in objects with undersized dimensions.

[rpress]

I increased the arm length about 4mm from what I measured using my cheap calipers.

Did you verify that your printer is still printing objects with the correct X/Y dimensions after increasing the arm length so much more than what you measured?

From my experiences and what I have read is that if you increase the arm length in the firmware/EEPROM then the controller won't move the motor as far for the same movement which would result in objects with undersized dimensions.

Arm length does affect the dimensions of the printed part, but obviously so does steps/mm. Whereas steps/mm changes it equally everywhere, changing arm length has a more convoluted effect.

My printed part dimensions are within 1%, when I account for shrinkage. I find that different filaments shrink a bit differently, so I usually have to scale things anyway to get perfect dimensions. I'm printing parts right now to build another printer, and the parts are fitting real-world items quite nicely.

[JFettig]

I just did this calibration with tower angles and got 0 at center and 3 towers and -.003 opposite the towers with using the radius of 90mm from the bed center as the manual describes with the original calibration - I'll move this out and measure again. Also, cold bed and PEI sheet on top of glass.

A corresponds to X, B to Y, C to Z. I call the opposite measurement for Z: Z-, and for X: X-, Y: Y-

Lets look at the Z tower, its at 90 degrees stock, modifying this value has no change on the measurement directly across from Z(Z-) but it modifies the measurements next to Z on both sides(X- and Y-). If you add angle to Z(C), nozzle gets farther away from Y- and closer to X-.

My numbers came in at A:210.05, B:329.95, C:89.75

Modifying steps/min will scale Z and X and Y, Z by a percentage, X and Y by a distance. Pulley size doesn't effect scale with timing belt setups.

[bot]

This is some interesting stuff. I believe I need to experiment with some of the tower rotations. On my machine, each tower has a weird phenomenon where the portion immediately to the left of the tower is squished into the bed a lot, and immediately to the right, it's not nearly as squished. You could draw a centerline on each tower, the effect is so pronounced.

[JFettig]

I just ran the code JJ posted here with slight modifications http://forum.seemecnc.com/viewtopic.php ... 100#p42808
I get -.005" on Y- and Z- and -.003" on X-. I think that is plenty good. This is at 115mm from center vs 95mm from center. I think modifying the carriage offset would allow us to take up that last little bit.

G28
G1 Z10 F15000
G1 Z0 F3500
G4 S5
G1 Z10
G1 X0.000 Y115.000
G1 Z0
G4 S5
G1 Z10
G1 X-99.593 Y57.500
G1 Z0
G4 S5
G1 Z10
G1 X-99.593 Y-57.500
G1 Z0
G4 S5
G1 Z10
G1 X0.000 Y-115.000
G1 Z0
G4 S5
G1 Z10
G1 X99.593 Y-57.500
G1 Z0
G4 S5
G1 Z10
G1 99.593 Y57.500
G1 Z0
G4 S5
G1 Z10
G1 X0 Y0
G1 Z0 F3500

[JFettig]

Changing the carriage offset up to +2mm has not changed anything. The delta corrections also did not help. I am satisfied with 3-5 thou deviation for a printer made of wood and 80/20

[0110-m-p]

It looks like my 0.005-0.006" deviation (nozzle gap increase between towers) is going to have be the best I can get as well (which is extremely frustrating). Here's a run down of what I did and the results.

1) Keep "stock" settings for delta arm length and tweak steps/mm until bed is completely flat....resulted in parts that were dimensionally good in X and Y, but 1.3% too tall. (Example: 150mm tall tower measured 152mm)

2) Keep "stock" settings for steps/mm and tweak delta arm length until bed is completely flat...resulted in parts that were dimensionally good in Z, but approx 1% too narrow/small in X and Y. (Example: 45mm x 110mm plate measured 44.45mm x 109.1mm)

3) Keep all "stock" settings for steps/mm and delta arm length and bed is not flat, but parts are dimensionally good in X, Y, and Z.

I say "stock" in quotations because there is some calibration off true factory settings, but just slightly.

[lordbinky]

All I'm taking away from this last page is that I have too few Cthulu statues. I knew he was getting his inspiration somewhere...

[duvdev]

Hi all,
I am having the same calibration issues on spots ABC as you all, but I can see some starnge movment:

If the hot end moves counterclock I will get to high or too low on ABC spots but it move colock side it will be OK.
also I can see some problem if the hotend is in front of Z tower and it moved to the center it will touch the glass and I will get no plastic layer at all but it is all OK arounf the Z tower.


can any one please advise?

Thanks

[dclunie]

So this weekend I've read through this entire thread and I can say I'm throughly confused and not sure exactly where to go next. I see lots of thing to try but none of them seem to actually FIX the issue, just a bit of speculation mostly it seems.

The things I have tried have been the official method as was posted by seemecnc to fix this, as in updating the firmware to use the changed Carriage Horizontal Offset from 35 to 37.5, which after downloading the most recent firmware for the max v2 the amount isn't 37.5 but 38+ which someone mentioned and I never saw any response to which is the version that I'm supposed to be using. I assumed the most recent firmware is the one that I'm supposed to be using.

anywho here are some pics of my issue in particular. As you can see I have a very odd shape when printing, The very center seems fine, the location where Z and Y are printing seem ok, the letter X is half squished into almost nothing (a high spot), as is almost the entire arc directly across from X between the Z and Y towers. The middle of the arc between the X and Y and X and Z are low spots and the filament doesn't even adhear, depsite the glue and blue tape (I'm using ABS) And lastly the inner circle seems ok except for where it meets at the X conjunction where is squish (high point) on either edge. Also something to note about my printer, I didn't build it. I purchased it from the seemecnc fellas at the makerfaire back in May of last year, they didn't want to haul it back and I purchased printing and running fine as it was at the show. Only recently have I really been printing larger objects (past the inner circle) and then decided to try and print out to the edges and found this entire thread and the exact issue i'm also facing. Its a rostockmaxv2 (black).

[img]https://lh3.googleusercontent.com/-Mu_K ... 182017.jpg[/img]
[img]https://lh6.googleusercontent.com/-ov4E ... 181952.jpg[/img]
[img]https://lh6.googleusercontent.com/-3yT8 ... 181919.jpg[/img]

thanks in advance - david.
http://www.dbclunie.com

[Jimustanguitar]

the official method as was posted by seemecnc to fix this, as in updating the firmware to use the changed Carriage Horizontal Offset from 35 to 37.5, which after downloading the most recent firmware for the max v2 the amount isn't 37.5 but 38+ which someone mentioned and I never saw any response to which is the version that I'm supposed to be using. I assumed the most recent firmware is the one that I'm supposed to be using.

I asked about this too, and the official answer was that the theoretical value and the measured actual value varied slightly, so they changed the firmware to the correct value after writing the blog/forum posts.

[nitewatchman]

You might look at the squareness of the x tower to the bed.

I suspect that it may be leaning away from the center.

[IFinnimore]

I have been reading through this thread and agree with the numerous posts about precision of the machine assembly being a key factor to the accuracy. Tower tilt/lean/rotation and position from ideal 120degree all will have effects on planar XY translation that becomes more pronounced the further the head is from the tower. However, many of the issues raised here are not just "opposite the particular tower" but are "anywhere outside the triangle" type of errors.

The IK and FK Mathlabs sims that have been done do highlight some of the bowling and potato chipping that can occur with the buildup of structure tolerances. These can and do affect the XY position of the arms (and therefore the Z). For example a tower leaning in by 1degree is displaced in by 3.49mm and is 0.76mm lower than the software thinks it is at 200mm height above plate. At 100mm above the plate the same tilt in causes a displacement in of 1.75mm and down of 0.02mm . Calculated as Sin(1)*H and H-Cos(1)*H respectively, for those interested

One item i haven't seen discussed here is the differences in forces exerted on the rods within and outside of the "golden triangle" of the towers and how that relates to motor stepping and accumulated moving parts slop (U-Joints/Rod ends/Skate/Belt tension). In the physics of a 3 arm position system, when the extruder platform is within the golden triangle all the rods are under tension of the weight of the platform distributed based upon the rod angles. Highest for the rods closest to their own tower and therefore lowest for those further away and all equal at the 0 center position. However when the platform moves outside of the triangle the lowest angle rods (most extended) now go into compression with the other two rod sets taking a much higher tension load.

The easiest way to visualize this is by doing the following thought experiment: Imagine replacing the rods with cables. With that setup, you would still be able to position the platform anywhere inside the triangle by tightening or loosening one or more cable sets. You would not be able to position outside this triangle however, because max loosening of one cable set only positions the platform between the other two towers. From that point you have to push.

My thoughts here are for those people who are experiencing the trouble of not a gradual Z position error but an abrupt change from laying down easily and cleanly to air print or nozzle dive outside the triangle should look closely at the moving parts slop. This is where the "zip tie compressing" the U-Joints or rod ends, or the stiffer rods with less play from tricklaser.

The problem has been stated here my multiple posts that many errors manifest in similar ways. All deviations from model information regarding the physical geometry will manifest as Z (and X/Y) errors. The XY errors are small enough that it is hard to see them through the noise of the plastic thermal contraction, but the Z error means failed first layer printing, so it is glaringly obvious.

I am working up my own Delta style printer (will be a 2m tall monster) and in doing the engineering am working up a spreadsheet to calculate variable effects. Given a few days to polish it up, and I will happily post it to anyone interested. For example, it already contains 10200 rows of calculated Z Tower positions required to reach points on a 101x101 grid including the positional errors that must occur due to rounding or truncating to whole microsteps. This is shown in a pivot graph to make it most readable, Now I am working on adding all the variables in a user definable manner and calculating the impact of each.


Regarding aligning of the components, the problem always comes down to the reference plane. What I have found useful for my other printers is a small piece of granite countertop that I picked up. I use this as the alignment platform, and reference all my squares to it when checking angles.


As for mapping the position of the bed, has anyone considered using a spring loaded PCB test probe or a bent bit of wire, and covering the bed with AL foil (or a machined true AL machinist plate for those who have one)? Simple circuit into the controller just like an end stop switch, with one lead to the foil and the other to the probe. Then the calibration can be just like end stop calibration, 100% automatic with "find contact, back off, step slow in again" until contact. Of course this means the foil has to be flat on the plate (no wrinkles please). PCB test probes have more than enough travel to allow time for the motors to stop and are held in a very fixed XYZ position when extended. I'm considering a variant of this for my machine, but may use the hot end directly onto a "home tool" spring loaded plate. Yes, I know that the residue build up on the tip will be a problem, but that is the next thing to work on. What would then be really handy is if the result of the "calibration" is the machine saying "Tower Z is tilting in by X degrees and there is excessive play in tower Y mech", as I always prefer systems that don't just tell me what is wrong, but suggest how to correct it.

[626Pilot]

So this weekend I've read through this entire thread and I can say I'm throughly confused and not sure exactly where to go next. I see lots of thing to try but none of them seem to actually FIX the issue, just a bit of speculation mostly it seems.

My firmware has been tested by three people (myself included) and it solved the problems for all three of us. You do need a Smoothie-compatible controller, which doesn't ship with any SeeMeCNC machine. (And a Z probe - ditto.) It can't fix leaning towers, but it is pretty good at correcting for tower rotations, tower radii (like delta radius, but specific to each tower), arm length, endstops, and surface normal. It could probably correct for tower lean if I added those variables to the IK/FK calculations and to the simulated annealing heuristics. On these printers in particular, it is very hard to get really square towers because they are inadequately constrained by the structure on both the top AND bottom of the printer. I used a digital angle gauge (the kind you'd use to square up a table saw to exactly 90 degrees) and managed to get each one to within about a degree, but a degree is still going to throw things off the further you get from <0, 0, 0> in any direction. A good fix, one I contemplated but never started work on, would be to print some replacement parts for the top and bottom plates that constrain the towers axially, so they can't lean in or out. Ideally, they would be constrained at the ends, so you wouldn't have to screw around with a yardstick and those little laser-cut lines to make sure they're all "sort of" the same height.

The easiest way to visualize this is by doing the following thought experiment: Imagine replacing the rods with cables. With that setup, you would still be able to position the platform anywhere inside the triangle by tightening or loosening one or more cable sets. You would not be able to position outside this triangle however, because max loosening of one cable set only positions the platform between the other two towers. From that point you have to push.

That is a pretty good way to visualize it. I believe that exact system is used in some sports stadium "flying cameras." I also noticed how ridiculously fast the carriages have to accelerate when the effector is furthest from them - they have to move several times faster than when the carriage is close. I wonder if the math doesn't get a little weird close to the very edge.

As for mapping the position of the bed, has anyone considered using a spring loaded PCB test probe or a bent bit of wire, and covering the bed with AL foil (or a machined true AL machinist plate for those who have one)?

This would not conform to the surface of the glass. Foil would be practically impossible to apply smoothly - remember, we are dealing with tolerances that have to be within 50 microns or so. Ideally 20 or less. A machined plate might be perfectly flat, but it will mask any surface undulations on the glass. My firmware uses only a Z probe, and that works reasonably well, but it doesn't report X or Y. Correcting Z (by adjusting geometry) tends to also correct X and Y somewhat, since all the carriage axes (in "Rostock coordinates") are completely interdependent. I had the idea to use three FSRs (one at each tower) at precisely known locations and using their reported "weight" to triangulate the probe position. That would produce X and Y coordinates, so the calibration would be completely closed-loop on all axes rather than just Z. However, I don't know if FSRs have enough resolution or repeatability.

[IFinnimore]

I believe that exact system is used in some sports stadium "flying cameras."

Yep it is, as well as for movie production work. Brand name of Skycam (which is also generic term) http://en.wikipedia.org/wiki/Skycam Invented by same guy who invented stedycam,

I also noticed how ridiculously fast the carriages have to accelerate when the effector is furthest from them - they have to move several times faster than when the carriage is close. I wonder if the math doesn't get a little weird close to the very edge.

I attached the steps curve for a 265mm pivot - pivot rod moving straight out from a tower. If the software is trying to maintain a fixed XYZ motion speed it would definitely speed up the carriages. Speed should be about 2x when at 200mm as compared to speed at 100mm extension.

Steps curve for Y Motion

BTW, in your smoothie board software configuration, have you considered removing the tower radii as a configuration and replacing it with something that can be directly measured, like Tx - Ty Dist, Ty - Tz Dist, and Tz - Tx Dist? I am a fan of using settings that I can validate with a micrometer or calipers, and radii isn't one of those. Then a calculation can be made like the following:

Given Towers as shown:

...........Z
!Y Side / \ X Side
.........X---Y
........Z Side

Angle cos A = (b2 + c2 − a2) / 2bc
Radii a/sinA = c/sin C

Tower Angle Bisect CW Center Angle Radii
X =ACOS((YSide^2+ZSide^2-XSide^2)/(2*YSide*ZSide)) =XAngle/2 =PI()-(XBisect+ZBisect) =YSide/SIN(XCWAngle)*SIN(ZBisect)
Y =ACOS((XSide^2+ZSide^2-YSide^2)/(2*XSide*ZSide)) =YAngle/2 =PI()-(YBisect+ZBisect) =ZSide/SIN(YCWAngle)*SIN(XBisect)
Z = ACOS((XSide^2+YSide^2-ZSide^2)/(2*XSide*YSide)) =ZAngle/2 =PI()-(ZBisect+XBisect) =XSide/SIN(ZCWAngle)*SIN(YBisect)

For values
X Side = 100
Y Side = 102
Z Side = 98

Tower Angle Bisect CW Center Angle Radii
X 59.6 29.8 120.8 58.210
Y 61.7 30.8 119.8 56.754
Z 58.7 29.3 120.8 59.69


Just a thought. Probably not worth a major code change to implement, especially since you have some testers already.