So I've had other work related things keeping from the ability to work on this problem much for a while. I did however during that time manage to print some new belt tensioner options that work very well. I now have consistent belt tension on all three belts and it is easily adjustable. Right now each is set to ~1lbs 3oz measured at 1" deflection from the tower at the halfway point from top to bottom of the tower.

I've checked the alignment of the towers once again and somehow they're farther off now than they were before. At this point I'm guessing attributed to environmental changes. It goes down to 18C in that room overnight, but during the day with people it runs around 22-24C. Also moisture has gone from 70%RH to 47%RH in the last month. So as of now the towers are very close to 90degrees to the bed. As in I can't slip a sheet of paper in between the square and the tower, but I can see a bit of a gap.

Measurements between the towers at the top and bottom were taken with a 6" caliper using the head to tail methods so they are not absolute measure, but simply relative between each tower. The X-Y bottom: 80.97, top 81.04. Y-Z bottom: 80.13, top: 80.93. Y-Z bottom: 80.24, top: 81.11.

(all measuring etc done with bed@60C)
This is this morning G32 with S-1 before adjustments with 0 offset in the bed.g
Bed probe heights: 0.103 -0.007 0.162 -0.007 -0.057 0.151 0.015 -0.022 0.115 0.020 0.010 0.039 0.011 0.117 0.062 -0.020, mean 0.043, deviation from mean 0.065

I just reran through the "Adding trigger height to bed.g" steps and calculated all the offsets and placed them in my bed.g.
Bed probe heights: -0.017 0.161 0.475 0.368 0.028 0.006 -0.258 -0.232 -0.142 -0.032 0.165 0.191 -0.004 -0.035 -0.136 -0.008, mean 0.033, deviation from mean 0.193

Then G32 with S6 x3
Calibrated 6 factors using 16 points, deviation before 0.197 after 0.052
Calibrated 6 factors using 16 points, deviation before 0.055 after 0.055
Calibrated 6 factors using 16 points, deviation before 0.051 after 0.051

Then I switch back to no offsets (because it was better before) and I get worse results for S-1
Bed probe heights: 0.174 0.022 0.355 0.301 0.210 -0.024 -0.090 0.017 0.015 0.004 0.263 0.186 0.048 0.001 -0.097 -0.128, mean 0.079, deviation from mean 0.144
S6
Calibrated 6 factors using 16 points, deviation before 0.168 after 0.072

Cycled power, now no offsets S-1
Bed probe heights: 0.099 -0.003 0.342 0.339 0.235 0.176 -0.002 -0.070 0.040 0.004 0.126 0.211 0.123 0.101 -0.010 0.009, mean 0.107, deviation from mean 0.121

S6 x2
Calibrated 6 factors using 16 points, deviation before 0.160 after 0.054
Calibrated 6 factors using 16 points, deviation before 0.055 after 0.055

M665
Diagonal 311.755, delta radius 146.204, homed height 360.718, bed radius 140.0, X 0.102°, Y -0.139°, Z 0.000°
M666
Endstop adjustments X-0.18 Y-0.20 Z0.38, tilt X0.27% Y0.19%

And printing a 75mm diameter single layer test disk comes out pretty bad looking still.

I have no clue what else to adjust, what else to modify etc. I've locked everything down mechanically that I can possibly think of and is possible with the melamine frame.

@number40fan:

I'll even add, if I can't get it to work, I'll pay to ship it back to you. Or I could load it up with a couple pounds of tannerite and dispose of it properly.

ROFL the tannerite idea has occurred to me SEVERAL times over the last two years!!

@number40fan:

If you live in the states, I'll pay shipping to take the one you have at home off of your hands to give a go at it. If I can get it to work, you can have it back for the price of my shipping cost plus whatever to get it back to you.

Heh, yep I do and that's mighty tempting! The more I read about Rostocks and their seeming random issues the less I have motivation to tackle it. I've been out of work the last few days sick so I haven't been able to get any more measuring done on the one at work unfortunately.

I just read another thread about calibration and I'm getting the feeling that much longer arms are needed to make anything work right. And the spring tension between arms also being an issue up for debate as to whether it works better or worse. It just seems like there isn't any clear answers when dealing with Rostocks. Yet I see quite a few people getting absolutely fantastic prints out of them…makes me confused.

Great….

Yup the older style CF rods used Traxxas ball ends and used rubber tubing bands to keep tension for that reason..I have that setup on my machine at home that I can't get right either and I've given up on basically. So either both ball end styles are amiss or... idk?

The ball ends are the now standard SeeMeCNC injection molded type. They don't have much play, I have to put some decent effort into poking at the effector to get movement.
https://www.seemecnc.com/collections/parts-accessories/products/replacement-ball-joints

And they came with the Trick Laser effector and mount
http://tricklaser.com/Aluminum-ball-joint-platform-with-Groove-Mount-PLAT-ALU-BJ-32GM.htm

I didn't know they had an odd flatted part on them from the pictures or I would have been suspect of their ability from the get go. However, I cannot believe that SeeMeCNC sells SO many printers with this setup and so many people having more than decent success with them if these balls are the problem.

Bed shift?? I really doubt it. NOBODY was around the machine and it's held in place…it could only possibly rotate around center if it moved at all due to the bed fasteners being at each of the towers. These are the holders I am using for the bed. I couldn't stand the clips because they force the bed to curve to fit the bow in the PCB heater. Which ALWAYS humps in the middle and causes problems. I've tried everything I can find to flatten said heater to no avail. These clips with the TR aluminum heat spreader plate on the PCB allow the glass to sort of float at the edges and reduce the overall affect of that curve.
https://www.thingiverse.com/thing:974489

@number40fan:

There you are….took me a bit to find this thread again. Do you have a way to measure the width of the diagonal arms near the carriages and at the effector? Reading above, someone in the "shop" has a large caliper. Might have them check the uprights/tower spacing, if you could. Place a flat bar on the inside from tower X to Z and measure how far away Y is. Do this for all three towers at the bottom, middle and top. There has to be something seriously wrong to have your calibration disk to come out like that.

One last thing, can you post a picture of how you have your Z-probe mounted?

Another fun print has been added that clearly failed horribly! No clue what happened there with the massive shifting.

I can measure diagonal arm width, I have standard 6" calipers. However, that almost seems moot since the width likely changes a bit since the SeeMeCNC ball ends aren't exactly round and looks to me like the arms would shift a bit during operation.

And yup I asked them to check that tower distance also, so they SHOULD have. However I can check that again.

I also posted a few more pics to that same imgur link. The terrible print and the mount for the IR sensor.

Here's images of the 40x40x2 block. And a 75mm diameter 0.2mm high calibration disk.
http://imgur.com/a/1iV34

@number40fan:

Did you save the results before printing?

After running auto calibration and before doing anything else, run the bed mesh compensation and see what your bed map looks like.

Don't forget that you can calibrate the steps of each motor too. Find an accurate way to measure the movement of the X, Y and Z carriages. Command 100mm movement and measure actual movement.

I'm going to post pictures in a bit if I can get a chance. I haven't done any mesh compensation yet, I'll look at that tomorrow I guess.

I'm not horribly concerned about the steps/mm thing. Those measurements are pretty darn close and could just be shrinkage of materials. I won't worry about perfect dimensions until I get the rest sorted out. Simply being able to print a good looking object is the primary concern right now. So far everything I've printed looks horrid with wavy walls and banding.

@number40fan:

Any progress today?

Sort of …it's printing. Which is better than it was before. Sadly I've now also experienced some interesting heat creep type problems which I never had before...so I'm trying to understand what changed that as the only hot end related changes were to a PT100 and the all brass heat block, which I'd have thought would reduce such type of problems.

This morning I run G32 and here's the results, bed @60C, nozzle@ 200C :
Calibrated 8 factors using 16 points, deviation before 0.089 after 0.043
Calibrated 8 factors using 16 points, deviation before 0.047 after 0.043
Calibrated 8 factors using 16 points, deviation before 0.048 after 0.039

So I'm thinking that's looking pretty good now. I fire off an XYZ calibration cube print and it looks terrible!
First layer is still doing exactly the same thing I've been seeing for weeks now. Toward the right side of the machine (between Z and Y tower and around to the front) the first layer prints SUPER thin, just measured 0.07 and the left side between X and Z towers is 0.19 or 0.17 depending on how far forward or aft it is measured.

The 'cube' is supposed to be 40x40x20 and I'm measuring ~39.16x39.14x19.91.

This is still with NO H offsets in the bed.g. Just the 8 factor calibration

@dc42:

You can use a nozzle contact Z probe, such as the one we've just started selling (https://www.duet3d.com/DeltaSmartEffector) or a piezo probe, to avoid the problem of trigger height changes caused by effector tilt. But whatever causes effector tilt causes other geometrical issues too.

I'm frankly not sure I'm willing to spend MORE money on the same machine. Every time there's a new fangled thing that is supposed to make it better and awesome all I ever get is a headache and marginal improvement if any. $500+ extra into it and I'm still no better off than the original machine with manual calibration mixed with some formula and guesswork. If the readings I've posted above are finally right and I can get some prints I will be truly amazed!

What say you about the automatic arm length calculations here? Everywhere I read that I shouldn't, but that 8 and 9 factor FINALLY shows decent results.

@Eddiie:

Grumble. It seems calibrating a delta is never complete.
My effector tilts a little bit too. It sucks but I think the bigger the radius, the worse the tilt on the outter edge. So, smaller radius means no tilt. There is some math out there to calculate the radius based on many angles and variables but no doubt it is a labor of love it seems. My Prusa clone printer is so darn reliable, thank god I got that to use while I toy with my Delta.

I'd LOVE to see 3 IR probes, one for each tower! I think that is the only way…

How is your dimensional accuracy?

Ha no kidding. One of my Rostocks at home is still in the same pile of parts it was almost 2 years ago because I basically gave up on it. I was never able to reliably print anything larger than 100mm circle. And even that was not great. Anything under that and it was flawless beautiful prints. Also note that a Rostock was my very first 3D printer….over four years ago!

Dimensional accuracy on the machine in question is undetermined...I haven't even been able to get a solid print out of it since I started this maddening upgrade process last month. Tomorrow when I get back to the office I can load up filament and test with the results I was able to get yesterday operating it remotely.