Print layers drifting towards rear

[DDLLC]

I agree that ball screw should not have been turning unless the nut parts were damaged. I expected at least the upper part to be stripped, but it seemed okay. I can't explain how it could be possible. Another run is in progress so I'll see if it happens again in about 1.5 hours.

Run finished and I watch the entire retract. It did not exhibit what I saw last time. The only easy explanation is a ghost in the machine. Honestly I've thought a lot about what I saw and I just have no idea of how it could be possible. I'll be watching the end of each print for awhile now.

[DDLLC]

As a recap all the following comments are relative to printing open nested rectangles (95 x 65 x 5 mm H with a smaller open square centered on the build plate) with Elegoo standard gray resin. The layer thicknesses are 0.020 mm and the base is 25 layers thick. The objective is to have the printed dimensions match the design dimensions as closely as possible.

I've tested the effects of all the resin parameters available in the Chitubox v1.7 (also tested a limited number of repeats of those same variables in the Beta version 1.8) using statistical methods. I will refrain from using statistical jargon to make my comments more readily understandable to all. (For those who do know, I admit some things I may say for the sake of clarity, might not be perfectly accurate.) Nevertheless a measure called standard deviation is important as that is a key measurement to minimize relative to desired outcomes (like printed thickness deviations from the design thickness.)  Standard deviation is a measure of how much the data is scattered about the mean (or average).  The difference is visualized as a small stdev producing a tall, narrow Bell Curve and a large stdev producing a short, broad Bell Curve. The reason that achieving small stdev's is important is that if a large percentage of the parameter data points are clustered close to the mean, I'd then be able to use scaling to dial in any differences between design dimension and printed dimensions.

Now for the quick summary: There is so much scatter in the data that I can't estimate the effect of ANY parameter (and forget interaction effects) with reasonable confidence. In fact the result may be the worst I've ever seen. To me that suggested there are uncontrolled parameters that have an effect (like temperature) or the parameter settings are not accurately executed by the printer. Consequently the following should be considered almost guesses that can't be backed up with mathematics that are not laughable.

I've made some comments in a different thread about my observations that room temperature seems to be critical at least for the base layer in adhering to the build plates. When I first started I was using a base exposure of 40 seconds. The first few stayed on the plate and I think the room temp was in the low 80's but then they started to either peel up from the plate on a corner or stick to the FEP. In one of those trials I believe the room temp was lower 70's. I restarted with a base exposure of 50 seconds which also worked for a while, but then some center squares stuck to the film and corners started pulling away from the plate for the larger rectangles when the temp got down to 68F. At that point I set the heater to 73 F for that room which seemed to have solved that problem. I recommend tracking room temp. If you have peeling from the build plate you likely will see a correlation to lower temperatures.

Also during this experiment it became obvious that the light delays parameters didn't seem to match the input parameters. I will create a separate thread for the associated information.

To get a feel how odd the data is, I've selected some runs which had some of the best (smallest) stdev for the thickness. In the chart below called "poor thickness repeatability" each run number has the same parameter sets (except there was an error in run 14). If there is an M behind the number it means it was performed after some machine modifications that write about later in this post. An "v1.8" means the run parameters were sliced on version 1.8 of Chitubox instead of version 1.7 like all the rest.

In general the average thickness is between 150-500 microns more than the desired 5mm. Typically the front rectangle rails are thicker that the rear rail by 100-500 microns. This was true for all test runs except it reversed on 2 of 3 runs after modifications were made.

For reference the worst run had an average difference to the design of 0.791 mm and the worst stdev in a run was 215 microns.

Even though the data wasn't robust from the experiments, I did run some predictor plots and the runs labeled 21 were based on the parameter directions estimated to give the best (lowest) stdev within the parameter space I tested. As my English and Scottish friends would say: "I was gobsmacked". I wasn't expecting to really get good results and with reasonable repeatability (... but go back and read the bold caveat above). This one single event makes me believe that there are main effects that can be optimized but the data is so buried in noise, it's hard to extract.

That also seemed to reinforce that something external to the print parameters is interfering. So I decided to take a break and look closer at the machine. I had indications from both a square and an electronic angle tilt meter that the vertical black post was likely tilted backwards off of square by a tenth of a degree.  I ended shimming the rear of the post by 0.002 inches. The tilt meter was indicating it was 90 Degrees to the base metal skin adjacent to the front of the post. The square still showed a small gap towards the front on the printer. I'm not sure how accurate either really is but I used the tilt meter to dial in a "home brewed CNC router that could hold to an accuracy range of less than 0.005 inches over a 4 x 5 ft table with a Z range of 8 inches. At this point it was better to try something. The fact that 2 of the M measurements had a thicker rear rail might be an indication that the shims caused a shift. Why wasn't it all three? Probably just wishful thinking on my part, but I'll reduce the shim to 0.001 to see what happens.

Another machine issue that still bugs me is that why do I have to torque the clamp screws so much (I stripped the original aluminum head and replaced it with a steel one with more threads to avoid stripping)? I Disassembled the head and removed the steel split cylinder clamp. I used a dremel to remove some of the steel creating "grooves" along the cylinder. It should make it easier for the screws to squeeze the cylinder around the ball post that holds the build plate. You can get an idea of what it looks like in the picture "sleeve after notching" below. Theoretically it should wrap around the ball with a bit less force. Frankly it's hard to tell.

Since I had some adhesion issues I decided to sand off the black powder coating. I didn't want to use power equipment so I put sand paper on my granite flat and went to work (it took hours). I had originally thought the build plate was flat because I couldn't slide a feeler gage under it around it's periphery. It had a high spot where the post is inserted and a couple of "troughs" running inside the plate. Nothing extreme as they were generally less than 0.0015 inches. You can see this on the image "100 Grit A". I followed up with 150 grit and then 220 Grit for which there are pictures below. The biggest impact I could see is that the rectangles adhere to the plate like they are attached with super glue. It takes a 5 in 1 and a hammer to separate the part. This level is probably not necessary for most but if you do this I'd maybe finish with 320 instead of 220. This may make a 40 second bottom exposure more reliable.

Lastly I've extracted the run parameters that gave best standard deviations. Beware of the delay times listed below (look for a separate thread on that.

Sorry for the long post, but maybe it might help somebody.

Greg

---

Here is the parameter file I could attach to the preceding post

[DDLLC]

I have a few more observations that may be useful for some folks.

I'm convinced that adding in a light delay can be beneficial to more accurate and precise thickness control. So far more delay is better. Details are in a separate post.

The lapping of the build plate to a 220 grit has a major impact on adhesion. In fact my parts are extremely difficult to remove now. I have to use a 5 in 1 tool and a mallet. Almost always I chip or crack the rectangles when I remove them. I've reduce the number of base layers (total base thickness is now 240 microns) in half but they still are tenaciously attached even with a base exposure of only 50 sec. I think I could easily reduce the base exposure to 40 seconds but have not made those experiments yet. I would recommend that most folks might be better off going to 320 grit; who wants to break their parts trying to remove it from the plate?

I reiterate that temperature is very important. The bulk of my recent efforts are all at 77 F, but the parts also are well adhered at 73F.

Since I've cut reliefs in the cylindrical split collar that grips the ball post on the build plate, I haven't had the plate tilt off parallel to the screen anywhere near what it had done before. The front bar of the rectangle is generally slightly thicker than the rear bar but that might be due to nonuniform light intensity on the screen. In the "Delay Anomaly" post you can see a repeatable thickness pattern around the plate perimeter, that might be due to either a slight alignment issue or light intensity. I may go back and revisit the alignment theory with more rigor as the ability to improve the light uniformity is constrained.

If your z trolley rolls down the rails without adding force to it's weight, it is too loose for repeatable, accurate prints. I've adjusted mine to just under a pound of force to get it rolling, but it might need more. Precision adjustments are not really possible with the design, but I think it's worth attempting in the future.

[cliffyk]

If your z trolley rolls down the rails without adding force to it's weight, it is too loose for repeatable, accurate prints. I've adjusted mine to just under a pound of force to get it rolling, but it might need more. Precision adjustments are not really possible with the design, but I think it's worth attempting in the future.

^^^This^^^ is vital.

Due to the poor design of the "3-pulley" Z-Axis rail there will be considerable inaccuracy if assemble with insufficient pre-load. Elegoo knows this--it is whay from the factory they are adjusted so tightly (mine and two others I readjusted for friends initially required over 5 lbf to move the trolley! Greg's ≈ 1 lbf is spot on.

There is a YouYube video out there misleadingly titled "Elegoo Mars Z Axis Rail Calibration" (It's not "calibration" as it's not being adjusted to some standard, it's just a plain ol' adjustment) that incorrectly recommends the trolley fall under it's own weight--this is BAD information. The 5+ lbf factory adjustment is excessive, however ≈ 1 lbf or maybe a tad more will be just about right.

This level of preload is needed due to the very sloppy fit-up of the 3-pulley's grooves and the cylindrical guide rods. Elegoo acknowledged this weakness in the Mars Pro and later models where they switched to a recirculating ball/grooved rail friction-less linear bearing.