October Update - New Feedback System

October Update - New Feedback System

Hey Everyone!

Its time for our October update, we’ve got another beefy one for you. We’re going to take some time to run you through the calibration procedures we’ve been working through, introduce you to a couple issues, talk about some sensors we’re considering putting into The Palette, and talk through some of the software pieces we’ve been digging into. We’ve been discovering some really interesting things, some good, some not so good. Buckle in, its time to get you another behind the scenes look.


3D printing isn’t a perfect science. It’s an amazing technology that lets us build some revolutionary things, but its not perfect. We’ve always worked to take into account these imperfections, and build our system in a way that lets us work with them. This has worked quite well up to this point, but a few extra measures are needed.

We recognize the fact that there’s a difference between us using The Palette in our office, and having hundreds of them being used all over the world. We need to achieve a higher level of consistency and reliability to give you the best product experience possible.

So what goes into calibrating The Palette and your printer, and how are we going about getting you the best system we can?

There’s lots of variables in play – when you dig down into it we know you could argue that filament manufacturer, temperature, humidity, and filament diameter variability all have an effect on the calibration. Luckily, none of those have a measureable effect. But, The Palette’s drive system, The Scroll Wheel’s Accuracy, and the accuracy/repeatability of the printers drive system all have the ability to cause issues with quality and calibration.

We knew from day one that the printer’s drive systems’ aren’t perfect. These imperfections are why we developed The Scroll Wheel. The Scroll Wheel lets us measure these imperfections, and account for them in future filament lengths.

But this is assuming The Palette always creates the right amount of filament, all the time.

We realized this wasn’t the case, and noticed variability between different drive systems. To account for this we have developed a series of calibration tests to run at the end of the assembly process. These tests measure the constant for each drive module, and record it in the firmware.

These adjustments have (literally) gotten us to the point where we’re 99.9% of the way there. If you do the math, that means we’re off by 1cm on a 10 meter long length of filament.

One of our more recent test prints – lost calibration after around 32 meters.

Here's the only thing,  that’s not good enough. This is where the engineering starts to get really interesting.

Printer drive systems are somewhere between 95% and 98% accurate. This accuracy is defined as the amount of filament they actually extrude divided by the amount of filament they’re supposed to extrude (as per the .gcode calculations).

So why are we worried about our drive systems (at a 99.9% accuracy), instead of the printer’s drive system (at an accuracy of 95%-98%)?

It comes down to real-time measurement, corrections, and constants.

We’re starting by experimenting with the calibration constant of The Scroll Wheel. We know that variability in the feedback measurement system can have some really negative effects on the calibration of the print. We have seen success with this approach – it got us an extra .03% of the way there, but this still throws the print off after a filament length of approximately 18 meters. Our goal is to make this number between 100 and 120 meters.

So how are we going to do this?


The Palette knows how much filament your printer is pulling in, as measured by The Scroll Wheel (covered in our last update, here). This lets The Palette change the lengths of filament depending on how much the output (printer) is using.

We know the actual amount of filament being used by the printer, but we only know the amount of filament being used by The Palette to a 99.9% accuracy level.

We’re looking into the feasibility of including another sensor inside of The Palette to account for the actual amount of filament being used. This would be another scroll wheel type sensor that would give us real time feedback to let us know how much filament The Palette has created.

We’re also investigating the feasibility of using a measurement for the buffer tube.

To refresh you – the buffer is where the two overlapping pieces of Teflon meet. It is what lets the printer continuously use filament, while The Palette is creating more.

We’ve been taking time lapses of the buffer to see if it decreases over time (that lost cm has to show up somewhere), and have noticed that it can be measured in the buffer.

The buffer is an absolute distance, not a comparative length. By measuring the buffer, instead of measuring the filament (via a scroll wheel), we’re able to be sure there is no additive error over time (maybe through some inaccuracy on behalf of the scroll wheel). We can measure the buffer, and know that it is 7cms where it really should be 9cms. The benefit of the buffer loop measurement system is in its absoluteness.

And the third option we’re messing with….. new drive systems. 

Fundamentally, the way that printer extruder drives work is wrong. They have little contact points, lots of slippage, and general inconsistency.

By choosing to re-design the drive systems we would be able to reach new levels of accuracy, with 0 slippage. It would ensure that the electronics and firmware stays simple, something that would let us keep extra variables out of the system.

We’re trying to avoid the situation where the entire system gets incredibly complex, which would lead us to a host of new problems. It’s a balance of fixing the problems of today without running into new problems in the future.

We were under the impression that the variability was from The Scroll Wheel, but we discovered during our tests (the ones mentioned in our last update) that there were more pieces to the puzzle.

We're going to be putting out an update near the middle on November on some of the progress we're making, keep an eye out for it in the next 3-4 weeks.

If you have any questions, feel free to chat with us on Slack here, we're available to help however we can!