Prusa Nextruder to Voron 2.4 Mount

prusaprinters

The goal of this project was to put together a test rig to prove out load cell probing in klipper. I've been working on this for the past 2 years and in that time we haven't seen much hardware developed. Facing the classic 🐔 or the 🥚problem I decided to write the software first use whatever hardware became available.Should You put a Nextruder on your Voron?Well I really hope you people do, I need feedback on the load cell probing code. Here is why you would want to do this:You get load cell probing, which gives you the best first layer performance I have ever experienced on a DIY 3D printer.You want to experiment with the Nextruder which has those big gears. Maybe with a klippers Pressure Advance and Resonance Compensation you can improve your print quality over what Prusa has shown on the MK4 and XL?You want that bleeding edge rig that no one else has. 😎You want to help me bring load cell probing to klipper.Talk is cheap, Show me what you got!How hard is this going to be?Hard. You have the following things standing in your way:Prusa, currently, wont just sell you a Nextruder. Most of the components are available as spare parts but the Nextruder gears and motor cant be bought that way. The cheapest thing you can do is buy an MK3.9 Upgrade kit, which will set you back $499. (Perhaps we will get Nextruder parts as spares soon, bringing the cost down)You'll need a wiring harness for the Love Board. It uses these really nice Molex CLIK-Mate connectors, for which no economical crimping tool exists. Parts required: connector, pins, less than ideal crimping tool, and some magnifiers. There is a post here outlining the pin outs for the LoveBoard. Alternately you could splice into the wiring harness in the upgrade kit if you get that.The HX717 chip on the Love Board communicates back to the main board using SPI. Normally this wouldn't work because the wire run is too long and the signals would get garbled. Prusa used an ST490ABDR to transmit the signals back to the main board. You'll need one of these chips at the other end to decode the signals back into standard SPI. You will also need a board to mount the chip on.The Fan ports on the love board pulse the positive pin for PWM. Basically all klipper compatible MCU boards pulse the negative pin for PWM. This means you'll have to directly wire the fans.You'll have to run a fork of mainline klipper with the load cell probing code in it. I'm working on merging it in but I expect it will take many months to complete this work.But is it worth it?You'll have to decide if this sounds like a project you want to take on.The first layers are the thing my Voron has been missing. My z-offset is 0. I have thermal compensation for the nozzle expansion but that's it. It's truly fire and forget now.The Nextruder seems to be worth the effort. It requires half the pressure advance of the E3D Revo + LGX.Just how good is the code?I have brought 3 ideas from Prusa's code base into Klipper:Filtering on the MCU: I've implemented configurable IIR filtering on the MCU using fixed point math so its fast on ARM Cortex M0+ chips. This is how Prusa rejects drift caused by the filament and bowden tube when trying to detect the probe contacting the build plate.The “Pullback Move”: Prusa does a small, relatively slow move away from the build plate before retracting. The point where the probe breaks contact with the build plate is the “true” z coordinate. This allows you to probe at 5mm/s but still enjoy the high accuracy results of a slower speed even with the 320Hz HX717 sensor.Tap Decomposition: The probe and pullback move are sent through an analysis engine in Python that decomposes it into 5 line segments. I've used the kneedle algorithm to do fast elbow detection in Python.The code is now based on Kevin's Bulk Sensor work and uses linear regression to get precise timing of the load cell samples. I'm committed to keeping the branch up to date and merging support in to klipper.There are some caveats:The software is “in development”, its not done yet and it might break your printer. I have tried very hard to protect your printer as best I can, but I cant make you a promise.I have not ported any of Prusa's proprietary Machine Learning based code for detecting “bad taps”. This code is machine generated and based on data Prusa gathered only from their printers. As such, we cant re-use this directly.You will have to probe “warm” with a clean nozzle. This is the same strategy as Voron Tap. Keep the nozzle under 150C for homing and probing. I have made an enhancement to z_thermal_adjust to make it easy to compensate for the growth of the nozzle when it heats up to printing temperature.You'll need a nozzle brush or a manual nozzle cleaning before each print.Probe AccuracyOn a Voron2.4 with the Nextruder toolhead and HX717 sensor chip I'm getting:6 microns of range of across 50 probes1 micron range across 10 probesless than 1 micron range probe-to-probeA micron is 0.001mm. So there are 200 microns in a 0.2mm layer. The next best competing probe is probably the Euclid switch probe at ~12 microns of range over 50 probes, but it has nearly that same range probe-to-probe. So on the standard PROBE_ACCURACY test with 10 probes, Load Cell is about 10x more repeatable than Euclid.Most of the range in the load cell probe is hysteresis of the toolhead: meaning it slowly deforms across many probes. Your results in a bed mesh are probably better than these tests indicate.How do I get the code?There is a branch here on GitHub: https://github.com/garethky/klipper/tree/adc-endstopPlease should read the documentation.The config for the reference printer is on GitHub: https://github.com/garethky/klipper-voron2.4-config/tree/nextruder-hx717There is an active thread on the klipper discourse: https://klipper.discourse.group/t/strain-gauge-load-cell-based-endstops/2134/The OnShape model for the toolhead is here.