But no matter what we do, we continue to see some 4mm periodic waves in the prints. The ball screw has 4mm lead, and I can feel the circulating balls when I turn the screw.
So we replaced that with 2mm lead trapezoidal screw.
Now we have 2mm periodic waves in the prints; it’s uglier.
Next is to replace the linear rails with higher quality (I hope) rails that I just bought; the same brand that I used in my corexy.
While the printer is apart, I’ll look into a bearing replacement in the X stepper motor.
Well, it changed with the lead screw, so that’s definitely a contributor.
The extruder can still move a little, and I think it’s play in the junk linear bearings. Within the extruder is solid, and it’s the same type as I’m using on my corexy.
Tore it down to replace linear rail, extruder, and hot end.
The X stepper motor bearings were bad. I tried replacing them, but with the new bearings on place the rotor is catching and I don’t know why.
The Z axis linear rail was stiff and crunchy. It was not great when I installed it in the first place, but now it is worse. Oh well, that’s why I bought new rail. The new rail is smooth and runs free without perceptible play.
But when I put the X rail back on, it became clear that the torque of the cantilevered arm is just too high, and the rail doesn’t move smoothly.
The original gantry design with the central screw would have worked better.
I’m going to declare this design as it stands now a loss. Now I need to figure out whether to keep the base and make a new gantry, redesign to use two stacked carriages on one Z rail and buy a set with two carriages, or just buy a second working printer and retire these parts for now.
The Y axis is really good. It’s 240x270mm bed with mains heat and an SSR. The X isn’t really that bad: I have a pretty good extruder and hot end setup I think. But I really learned a lot about what not to do on Z.
This morning I realized that the Z stiffness was binding against the post opposite the lift stage, and relieved it. Then the Z axis moved smoothly.
Changed out the extruder, changed out the old 24V fan for a 12V Noctua. Added a buck converter, and attached the old 24V fan under the electronics hooked up to 12V to still move some air through the electronics box for cooling, but be quieter. Changed to a pancake stepper for the extruder, and replaced the X stepper motor cable. (The 17HS4023 pancake stepper is rated at 1A but gets hot running it at 800mA and I think I might reduce that current.)
The buck converter is taped underneath the electronics box, which doesn’t look beautiful, but there isn’t enough room for it in the box, and it at least works for now.
Finally put the ball screw back on and got it very nicely aligned.
So I tried another test print.
I still have a 4mm pattern of more or less squished layers, as before. So none of the improvements I made help. And I printed with a cold bed to make sure it isn’t periodic thermal cycling. Sure seems like I should try the lead screws I bought
I guess I need to get a good KFL08 bearing for the lead screw and make an adapter for my lift plate and see whether that helps next. But wow the KFL08 bearings on Amazon have a lot of terrible reviews, even discounting the people who don’t understand the intentional self-aligning feature.
That usually indicates slightly bent or over-constrained Z screws, and/or layer heights not matching screw-pitch/full steps. At least that’s what I’ve found on some I’ve built.
So 0.32mm or 0.25mm layers would probably be better. Clearly the next step.
In addition, I could set higher current to the Z motor for more torque for better holding in microsteps. Right now it’s 800mA. It is capable of at least 1A, maybe 1.5A; I have to look it up because I’ve forgotten.
I’ve been so focused on a mechanical defect I was wearing blinders about level height and aliasing errors.
Also, depending on what firmware you are using, and I could be wrong, but I believe Marlin has a feature where the Z axis motors can be de-energized if not used for (n) seconds. You would want that disabled. Otherwise if they de-energize between microsteps, they will fall to the next full step.
EDIT: found it.
DEFAULT_STEPPER_DEACTIVE_TIME
In configuration_adv.h
Although, probably not an issue unless you have some huge models where the layers take a while.
I never set that option. I don’t think it makes much sense in an open-loop system, and especially with microstepping. And I don’t have encoders on this printer
I’m using Marlin now and thinking about klipper. I expect that .25mm layers printed with klipper might be faster than .3mm layers printed with Marlin, based on what I’ve read so far.
I’m just so used to Marlin after all these years; they really have done a good job IMHO. But resonance compensation is compelling…
I’ve been doing tests with a cold bed to make sure that there’s no periodic impact from thermal cycling. There shouldn’t be (it’s PWM not bang-bang), but I’m just trying to remove variables.
Merely changing to 0.32mm layers (evenly divisible) didn’t make things better. (Maybe worse? Hard to tell.) I also confirmed that my corexy also has 800 steps per mm in Z, and doesn’t have this artifact.
I disconnected the arm from the “stabilizing” tower on the other end, going back to pure cantilever, and it made it noticeably better. I think the tower over-constrained it. I added it in the first place because it looked like it was “wobbling” in Y, and I was trying to constrain against Y movement in the X system, but actually that was primarily a loose mount for the heat break into the extruder, and secondarily a loose X linear rail. So poor diagnosis led to a cure worse than the disease. It’s now definitely better than the printer that I tore apart for spare parts to build this one.
Now I’m testing increasing Z current from 800mA to 1A for additional holding torque at microsteps.
1A didn’t help. I have a 4mm pitch ripple with an amplitude of 0.1mm. Although the ball screw is running true visually, it must be imperfect, and I’m not even sure how I want to try to measure it. I think that the printer isn’t stiff enough for me to be confident that I’m measuring it accurately, either with calipers or by stacking gauge blocks.
The last time I tried a lead screw, I just zip-tied it to the adapter plate. It seemed solid, but it probably wasn’t. Also, I let it rest on the coupler instead of using a KFL08 bearing. Therefore, before I try a lead screw again, I think I need to make an adapter plate for the lead screw nut, as well as borrow the KFL08 from the ball screw. Shouldn’t take too long, but I’m behind on my todo list around the house this weekend so likely will not be immediate.
Turns out the bearing I have is a KFL10, so I had to order KFL08. I’ll want to work the bearing free before using it so that stiffness in the self-centering feature doesn’t add a bad constraint.
I don’t think I need an anti-backlash nut because gravity is a spring that doesn’t fail. But I ordered an anti-backlash nut anyway. Probably won’t hurt? I’ll have both the standard nuts that came with the threaded rod and anti-backlash nuts to compare.
I’m partway through making a mating flange to connect the trapezoidal lead screw nut to the lift plate. 16mm square hole pattern in the typical 22mm OD lead screw nut, but to orient them on 45° the way I wanted requires the holes at (±5.655, ±5.655) centered around the 10mm center hole. But I forgot to account for the radius of the edge finder and made the hole pattern off center, so I have to start over again. Then I have to de-burr carefully to get the nut to seat flat.
The KFL08 reasonably has a different hole offset than the KFL10, so I had to drill new holes in the mounting bracket. The KFL08 I got had holes 37 mm on center instead of 36 that I think is standard — glad I measured before I drilled. I also made sure to work the bearing loose in the flange so the self-centering feature would work. I finally kept my brain turned on while making the flange adapter, and it works. Using the front panel knob to adjust from 800 steps per mm to 1600 by 0.1 increments was 8000 detents. I remembered to save the new setting!
I am trying a test print now and I can already see some waves, which are now 2mm in pitch. I won’t be able to measure them until the print finishes, of course. I think the print quality is better than the donor printer but I’m still not satisfied. It should be better! Given that the pitch changed to follow the pitch of the lead screw it’s not a hidden thermal cycling problem. I have some sort of mechanical defect here.
The difference measured across the cube between peaks and troughs is about .15mm, no matter which side I measure on. I’ve validated that I truly have thicker and thinner layers; it’s not Z-periodic shifting in X and Y.
I’ve tried replacing the stepper motor and increasing stepper current to 1.2A. I’m using an even number of whole steps per mm, so no partial step aliasing. I made sure the motor is aligned with the Z lead screw. The Z lead screw doesn’t wobble while it runs. I measured extrusion to calibrate, but reducing flow to 95% didn’t make a measurable difference.
I still have no mental model for what could be causing this defect.
There have been a number of successful cantilever designs over the years but I think they’ve all used linear bearings. If you are using the Delrin wheels on Z or X axis(like ones used on Ender3s) there is likely too much compression of the delrin to be without variations in accuracy.
All using linear rail. The delrin on the far side was an attempt to stabilize it, and I have since removed it. This isn’t misumi or hiwin high quality rail, and the tiny bit of play is magnified over a fairly long arm. A lot of the successful cantilever designs have a shorter reach. And/or maybe higher quality rail.
I’m sure this could be modified to work, but I just don’t want to design new components right now. For example, I could put two linear rails on Z, but then I’d have to redesign the whole stage and I just don’t feel like it…
Cantilever based printers could use linear rails or linear bearing. If you go with linear bearings, you have 2 rods.
I tuned in a bit late, but I noticed the last two posts where dougl was saying linear bearings and mcdanlj was saying linear bearings and it sounded like a disagreement. I just figured I would stick my nose in this and say both options are valid. I agree that delrin rollers would not be very reliable. I suppose if you really had no other option, it might work somewhat for a little while to get you another printer printed, but it would be prone to issues the moment they wear or get misaligned.
In the long run, I think an ABS printed rail with a no bearings piece wrapped fully around it might have more lifetime than a delrin setup. Not that it would be the best. It would be interesting to see someone try that too.
Good point. Yes, if you run two linear bearings further apart that is more stable than the two bearing races about 9mm apart in the linear rail I am using. However, I don’t think that difference is responsible for the regular pattern I’m seeing in the output, since it keeps following the pitch of the Z screw I’m using. I guess I could have tried using a geared stepper and a Z belt without throwing away the rest of the design, which would work only if I used high-quality pulley for the belt; any wobble in the pulley would result in an even worse version of the same problem.
For all my complaints about this printer, it is working way better than the i3-style printer running on linear bearings that it replaced. It’s just not as good as I wanted. But for pieces where the very slightly inconsistent layer height doesn’t matter, I’m using it to print parts for the SK-Tank I ordered.
I can’t see myself ever using printed rail for anything, but since you were curious about seeing someone try it, 3D Printing World did with several different filaments. Start here:
I don’t think it makes much sense in an open-loop system, and especially with microstepping. And I don’t have encoders on this printer 
