Answers to the questions I had going into the Ultimaker 3 launch event (plus a few surprises):
The nozzle lift does not require an extra motor. The left nozzle is fixed, and the right nozzle drops below it or lifts above it using a switch on the side of the effector, and there’s a piece of plastic on the side that the it runs the switch into to toggle it (judging by the force needed to switch it by hand, it’s probably close to the stall torque on the Y axis).
The OLED display does appear to have been replaced with an LCD, for some reason.
The construction is almost identical to the UM2 aside from the effector. It looked like it might have been built out of different materials because of the lack of black dibond edges, but it’s the same material with a white core. They did, however, remove the indentations from the top that allow the machines to be stacked.
The Matterhackers guys said that they’re still using the same old ultimaker controller boards (still no 32-bit), but have added a single-board computer (Raspberry Pi equivalent) running linux to handle the networking, flash drive, and camera. Camera images are retrieved by Cura using an HTTP request, so it will be easy to set up something else to monitor them.
Nozzle height is calibrated with a capacitive system (most likely measuring the movement of the nozzle/heater block when the printcore spring compresses).
Everything but the print cores was a bit underwhelming, and I would rather have seen those as part of the UM2+ upgrade (probably the only reason this didn’t happen was because of the added wires needed to connect to the new components).
Between this and the proliferation of multi-input-single-output bowden systems, I think 2017 will be the year that desktop multi-extrusion really catches on.
Not much information on that NFC system for the spools. With 2 spools and these distances it must have 2 antennas.
The new shaft doesn’t seem to accommodate the large 2.2Kg spools anymore. With these tiny spools and no detection when it’s empty and needs to pause mid-print… that is going to be a problem.
Didn’t find out anything about the NFC, but there are larger spools. They didn’t look like a full kg, but more than twice the size of the 350g spools that seem to be “sample plastic” that comes with the machine.
The interesting question is if multi-input-single-output will reliably work with PVA in combination with other material. PVA extrusion still has its own set of problems.
True. The advantage of MISO is that the PVA isn’t in the hot end when it’s not being used, which avoids the dreaded cooked PVA clog. I suspect that E3D’s scaffold will work, even if other PVA-based filaments have trouble.
To me the most impressive prints on the UM3 were Nylon with PVA support. This is a 50°C gap in extrusion temp. Also a challenge to meet with MISO. Commonly seen cool down and heat up speeds are at some 1°C/sec and 2.5°C/sec respectively.
The UM3 heatcores have some of the smallest heater blocks I’ve seen. Small block means less thermal mass, so heat up and cool down can be faster. It also means your control loop needs to do more work, ideally with some feed-forward.
That’s one of the things I like in Cura 2.3. It already uses heat up and cool down speed settings to cool down and reheat the idling extruder on time. Works great not only for the UM3 but also for other printers like the UM2 tool changer.
The lifting has more then enough motor power to spare.
The display is OLED, but white. It’s just a white version of the blue one in the UM2. (yes, ver underwhelming, but not LCD)
The board is indeed a new iteration of the AVR board that we used before. Still running Marlin. It does need to do a whole lot less in the UM3 then the UM2. Because…
The other board is an Olimex LIME2 A20 board. Running anything from remote APIs, to the display.
The measuring system for the bed and difference in nozzle height is a pretty clever system that never requires calibration. (could make a whole post about it)
The temperature control loop has a feed forward. If I knew how much difference this made, I would have done it for the UM2 as well. I will publish the code soon, as it’s in Marlin.
The NFC is currently only used to identify material type. We also wanted to write back the amount of material left, but this wasn’t finished in time. Nothing depends on using this, and it’s completely optional. The machine even keeps working if you remove this hardware.
NFC uses a single antenna, inside the spool holder.
@Markus_Seidt I would personally avoid PVA in any kind of mixing or MISO extruder, since you can’t guarantee there’s no residue during material changes, and PVA pyrolyzes into vile tar at the normal printing temp of every other material. Multi-nozzle is the way to go if you want to use PVA.
@Daid_Braam Can you describe the feed-forward mechanism in simple terms? Heat loss model? Extrusion melt flow compensation? Several other people have been working on this sort of thing, including RepRapFirmware and Redeem. Curious what you’ve done with Marlin.
@Ryan_Carlyle in control mechanisms, a feed forward (Kff) is actually very simple. It’s “target temperature” * “factor”, which is added in the normal PID loop. Which, for Marlin, means the I also should allow to become negative.
So, straight copy from the code:
pid_output = constrain(Kff[e] * (target_temperature[e] - FEED_FORWARD_MINIMAL_TEMPERATURE) + pTerm[e] + iTerm[e] - dTerm[e], 0, PID_MAX);
We’ve had someone with an actual background in this stuff tune the parameters, and it lowered our heatup times and it increased our overall temperature stability. For example, our PID settings are now: {“Kff”: 0.57, “Kp”: 9.0, “Ki”: 0.2, “Kd”: 40.0, “Ki_max”: 40}
The FEED_FORWARE_MINIMAL_TEMPERATURE is 35, as that’s that’s where I got with linear interpolation from our normal temperature range vs power required to 0 power.
@Daid_Braam Ok, yeah, that’s a heat loss model, thanks. (You probably know this, but for the other people in the thread…) For normal printing temps, convection and conduction dominate loss of heat energy from the hot block (radiation can be neglected) and heat loss to the environment is linear with the difference between the hot block and ambient temp. The Kff factor you’re using is simply the slope of the heat-loss-vs-temp line, but in units of PWM output per degree instead of watts per degree.
RepRapFirmware is doing the same thing. Except it also has a special autotune routine to match since it supports lots of different hot ends. Feed-forward / partial model control is SUPER easy when you have a known hot block configuration, and frankly I’m confused why it hasn’t been implemented by more printer OEMs.
I get the point that the PID loop can be greatly improved. But be it 1.5°C/sec or 2.5°C/sec heat up speed which I see in the UM3 atm, to me the important thing is that the pausing extruder will start reheating in advance exactly at the right time.
It’s not an LCD? Must just look different because of being refreshed by a linux machine instead of Marlin. It flickers horribly on video. I’d guess that you’re blanking and rewriting it several times a second. You might consider slowing that down so that the pixels have more on–time.
The decisions they have made here are interesting and sort of split across two different markets. Dual head printing and price makes this machine squarely targeted at businesses that have more skilled users on hand. But the USB stick, camera, wireless printing, z calibration and filament detection are nice ease of use additions that anyone with little experience and no desire to learn more technical details would enjoy. Of course, ease of use has benefits in any market and trickle down features could benefit any printer targeting lower price points and less experienced users. The fact that dual printing is the flagship feature screams “for professionals!”
Interestingly, with many companies and the press, including Makerbot and ultimaker, backpedalling a bit on the perceived hype and demand of consumer 3D printers, landing on both professional and educational markets… I see these two markets as very different with two important things in common: bigger budgets and eager buyers. You could also argue the expectations of an understandably higher learning curve and benefits from adoption are significant. This makes those two markets the obvious choice for a safe bet on where to focus time and energy. The higher price point may reduce overall sales, but the higher margins make sense. Making a very high resolution printer that is very reliable is the natural progression for some companies- but it does come with the trade off of complexity and price.
It will be interesting to see people contrast Printrbot’s new Simple (soon to be released) and the decisions we have made. We do not believe that consumer 3D printing is dead, nor a bad market to target. It is a bigger gamble since this point remains to be proven. The difference in approach across the “major” brands is widening. China is focussing on cheap printers. MB and Ultimaker are focussing on education and professionals. Printrbot is choosing to focus on newbies- which are not served by MB or UM due to price and ease of use (that last point is arguable, I know) nor served by Chinese made bots due to a variety of trade offs: quality, openness, ease of use, support, speed and reliability (again, these may be arguable).
I am impressed with the mature decisions that ultimaker, and yes, Makerbot, has made to target their desired audiences. But I am making different choices.
My prediction, with no ill will intended towards MB, is that Ultimaker will win the hearts of professionals and higher education.
My hope is that Printrbot will win the hearts of education-with-limited-budgets and consumers.
@Whosa_whatsis I think that’s just a side effect of the OLED display, the driver isn’t clearing the display at all. And I’ve seen the same effect on some UM2 videos.