3D printer input shaping with the energetic David Crocker at the Duet3D booth at formnext 

That moment when a bright mind visits another bright mind's chanel! Looking forward to the next episode of Hotends Olympics!

Glad it was helpful!

Indeed, he's such a great guy! We casually discussed the implementation at first, but then I knew we had to share his knowledge and found a camera to record everything from the top.

@@MihaiDesigns And I'm very glad you did! I've always used "control boards from the east" for all the reasons he graciously acknowledged in this video and have seen duet as a premium alternative but I've absolutely gained a whole new level of respect for them from this video.

Very interesting! Where can I find more technical details about this?

Having just implemented IS for Marlin, I concur. And I would go further: there are multiple things wrong with the explanation of IS in the video. The jagged path described, for example, is not the path the print head will follow. Inertia will smooth the path out but if the resonant frequencies are different for X and Y then the inertial effect will be different for X and Y and so the jagged path will be needed for the smoothed path to best fit the intended print path. Also the video implies that RRF does not apply IS to the direction change at each corner. If this is really true, a major source of ringing is not being addressed. The only way the explanation given could actually work is if at each corner RRF decelerates the print head to a complete stop and then pauses for a short period before beginning the next segment. Given that IS is supposed to increase print speed and quality, this would be counterproductive. I also baulked at the claim that EI3 applied identically to X and Y is better than, say, ZV applied independently to each axis. This is, indeed, true if RRF really does pause at each corner but for any firmware that maintains speed through the corner, it is misinformation: in general the least deformation comes from the simplest shaper. You only use more complex shapers if an axis has multiple dominant resonant frequencies and then it becomes a trade off between how much resonance is cancelled and how much distortion is introduced.

@@tombrazier6172 Thanks for chiming in. I agree with all your points. So, I think there are two possibilities: 1) the implementation of IS in RRF follows the classical approach and then the explanation in the video is wrong, and the suggested benefit isn't there; 2) the implementation more or less follows the explanation in the video, but then it's not how it is usually done and will have some other defects instead. As you described, due to different masses/stiffness, different forces on X and Y are required to better follow the expected trajectory. Then, RRF will also not be able to compensate for velocity jumps at corners (when junction deviation or jerk is non-zero) so there will be some ringing left, and it will not be able to input shape short moves when the input shaper does not fit the duration of the move (e.g. when filling some small infill areas or printing smooth curves composed of myriads of small moves). In fact, earlier I observed in the RRF code that if a move is too short (fast), first it tries to slow it down and then, after some threshold, disables input shaping for it completely. So, then perhaps it is the latter, but then this isn't an obviously better implementation. In fact, I'm really looking forward for someone not affiliated with either of the firmware to make a fair comparison of the input shaping implementations in them (RRF, Klipper and now Marlin).

@@MihaiDesigns You can check out 'Effects of input shaping on two-dimensional trajectory following' by W. Singhose, N. Singer, where more or less this problem is discussed in more details. They also mention that the way to avoid smoothing (rounding) is to introduce pauses between moves around corners, something which isn't practical for 3D printing as the material will be leaking from the nozzle during such stops, forming blobs at corners. Otherwise, there are many good articles on input shaping, for example by Singhose, or an overview article 'A review of command shaping techniques for elimination of residual vibrations in flexible-joint manipulators' by C. Conker et al.

@@dmitrybutyugin3857 I have not tested for this but I suspect that a lot of short zigzag movements at the resonant frequency (which infill could easily generate) is a potential source of layer shifts. IS is important here and from what you say RRF might switch it off for this case. I also would welcome a comparison between RRF, Klipper and Marlin - and also the GH Enterprises stepper driver which incorporates an MCU that does IS for you. I am also very interested to know how Klipper has implemented it but I don't really have the time to read the source code.

An dont forget Reddit LOL. Yes he is there.

Indeed, it looks like there are multiple approaches to IS and I'm excited to learn more about all of them. Perhaps I should do more interviews. A bit more difficult over the Internet, but still possible.

The asynchronous gcode streams is also a very powerful idea, let's see how it develops.

Great idea! I'll be using the boards over the next months anyways and might as well track what I need to learn and change and summarize in a video or two.

Glad you enjoyed it!

Yes! Working on PitStop for a while now. Got some partial builds. Waiting for the boards to arrive 🙂

If you don't have a compatible accelerometer, there are special prints you can make and measure to determine resonance frequencies, which you later use to configure the input shaper. You should be able to find tutorials on this.

Where can I learn more about this? Because in my mind (and this is more of a guess) if the accelerations match, then the cancelled frequencies must match as well. Or if we're targeting some middle frequency between the two, then aren't we only targeting one?

Is there any 3D printer firmware that implements this average target acceleration?

@@MihaiDesigns Yes, you would be targeting only one. But you could choose which one. Using a weighed average would target a frequency somewhere inbetween the x and y resonance targets. Or a simpler algorithm would choose just max(x, y) to target the more prominent one in each situation.

@@johnkim3858 I have no idea, but it could be a nice idea for a RU-vid video to test. Set 2 very, very different input shaping coefficients for x and y and then run with different firmwares and see how it behaves.

I couldn't find much out there. He's busy solving the technical challenges and leading the firmware development. It's why I was so excited to film this interview. Happened spontaneously.

Just noticed that now. Not sure which patent that is, but someone might know on the Duet3D forum.

No, which is why he drew the corner at an angle

No, they do industrial or oem stuff and not consumer.

From personal experience with years of Marlin frustration, tried klipper once it was too much programming for my liking I didn't like the RPi setting up, then tried RRF with Duet boards and will never look back at either klipper or marlin unless forced to. Duet has its audience being those who just want to configure the machine and run it. I can setup a Duet board for a 3d printer or pick and place in under an hour. That's why I love Duet they give you a platform where you can configure everything via gcode you never have to program C++ or python. Klipper is similar but you still have to install the firmware is has gotten super easy nowadays though. Also Duet's hardware just works seamlessly with each other and everything is backwards compatible for the most part. I have several "marlin" Chinese touch screens that turned out to be outdated and useless with newer boards and firmware versions are all over the place. With Duet most of the time using old hardware like the paneldue is just including a line of gcode in the config. David is just passionate about his baby would be too so would you, and he does give credit to klipper for innovation on the input shaping. Then proceeds to rip them a new one for independent axis shaping why? because the man studied the hell out of it to incorporate it into RRF. As an engineer myself I understand his passion and why he's so head strong because he's put in the work and understands the advantages and disadvantages of different implementations and he's very open to saying they went the wrong direction but they learned and improved. Lots of other companies aren't developing/innovating mainly just using whatever firmware is on the top sold machines. why marlin? It was on the prusas but now klipper is coming stock why? It's on the vorons and I think on the bambu maybe idk. Had Duet boards or just RRF been on those machines it would be the dominant Firmware. IMHO marlin is dying staying alive because of prusa, klipper is for programming enthusiasts and hobbyists, and RRF is for engineers making machines in industry. Also my favorite part of Duet boards and RRF you don't need wifi at all which is my personal preference give me my touchscreen HMI with GUI and SDCARD! Those are my two cents.

@@sculptaware4548 actually, with off-the-shelf hardware (I mean some widely used boards), you can install klipper with zero programming skills. The only compilation requires to jump once in a console and execute "make menuconfig && make flash". The rest could be (and should be) done from the web interface

@marsgizmo filmed it. Formnext is very exhausting. Soo .... On the other hand if you use the right OS and there the right player you can enable deshaking for youtube....