Linear position with a rotary magnetic encoder

[@matthiasrandomstuff2221]

I figured out the encoding coming out of the device at the end. I was confusing because I was only looking at the bits, not the clock, and some of the clock pulses are a bit longer, so really confusing. At any rate, its synchronous serial, with the sensor on the rail the master (sensor provides clock). Sends out a 24 bit word, LSB first, last 3 bits always '1'. Resolution somewhere around 150 increments per millimeter (only a guess). On power up, it always starts at zero. Reason it doesn't go back to zero after turning it off is because turning it "off" only turns off the display. Removing the battery resets it to zero.

[@FriggOff361]

@@matthiasrandomstuff2221 i was just about to say that

[@HairyNumbNuts]

And the lesson is ... if you already have a dial indicator use that for positioning when drilling holes that need to be accurately located.

[@trevorholland5032]

Instead of interpreting the electrical signal, you could point a webcam at the display and use Optical Character Recognition (OCR) to read the numbers in to the computer.

[@johnheitke7617]

Your Rigol scope has serial decoding capabilities, see chapter 7 of the manual. Most likely I2C protocol

[@awesomefeldmanfamily]

Frigging genius

[@bazzatron9482]

I love that you literally have a machine lathe, you could have made a very accurate way to hold those magnets - but decided to shove some HST on a nail before using it to measure micron accuracy on a cheapo XY table. Never change Matthias 😅

[@alucard87]

that would be way less fun :D

[@AJMansfield1]

And that's the thing, too -- for him it that sort of thing usually works!

[@davidsadowski2886]

Try a fridge magnet! They're on the weaker side but polarized as a halbach array with very consistent spacing. Commercial magnetic linear encoders are built this way and work quite well for the price point

[@sparqqling]

THIS!

[@Love2Tinker2]

Sweet idea!

[@conorstewart2214]

Linear motors work similarly with a line of individual rectangular magnets arranged with opposite orientations.

[@skh7791]

FFT's in excel!!! Matthias makes me so happy.

[@dwalsh3469]

Been there for at least 20 years…. 😇

[@cashewABCD]

Matthais running Excel 2003 like a true nerd. :)

[@WangleLine]

Yesss!!

[@thephlophers]

Chasing microns with a wooden block, very nice. On those cheap DROs, if it's anything like what I have, it has a simple clock and a serial data line encoding the measurement in binary. Similar to UART in principle. You could absolutely have a raspberry pi decode it.

[@69dblcab]

Never boring stuff in this channel. Thank You Matthias. Looks like the other brianiacs (with affection) have begun chiming in. Always a fun read. I will need to check back later.

[@jcmo5900]

Agreed. I click on these videos even when I don't think I'm interested. I love the way you share your work, Matthias! Thank you, thank you, thank you so much for taking the time to film, edit and share. I know how much work that is, and we are many, who truly appreciate your efforts.

[@simonhopkins3867]

This is a great channel for the comments section as well as the content.

[@eliasleb]

I worked on the precision position control of a linear motor for industrial applications. The setup you present at the end of the video is actually very close to what we were using. We actually ran a "learning" sweep to determine field strength vs. position using an external sensor and compensated online with a look-up table.

[@rahveibo]

+1. Came here to say this, especially if the readings are somewhat repeatable over time, a lookup table is usually as good if not better than fitting polynomials and sinusoidals etc. As long as it is repeatable it is modellable. Cheaper than a dollar microcontrollers have enough memory to hold thousands of lut entries for tens of temperatures each. No reason to try to represent the interaction of bunch of complex parameters with a handful of variables only.

[@danielkruyt9475]

Excited to see how you do on decoding that cheap linear scale's outputs!

[@stevewurster]

LEGO bushings for standoffs! Love it!

[@illygah]

LEGO tolerances are insanely tight. This solution is divine.

[@clasdauskas]

It's a pity the rest wasn't as accurate :)

[@simonhopkins3867]

This just feeds my hoarding habit. 😉

[@hpda44]

I still love this stuff. I’m not sure why, a lot of it I don’t particularly understand, but it’s fun to watch. Great video

[@Don.Challenger]

I think more often than not even if not fully understanding the gist is pretty evident.

[@CrawfordAutomation]

I love this so much and I would watch an entire series of this. Please don't stop!!! What about making an optical encoder with a light and a photosensor with a wheel that has a hole???

[@jeffspaulding9834]

You know your audience, Matthias. FFT was the first thing I thought about when you said "sine waves." I appreciate the analysis - it's a bit lower level than what I normally deal with so it's nice to get a glimpse into that world.

[@ianrwin]

The calculations for periodicity of the error, and the coincidence with the gear teeth are excellent. This is so cool!

[@stefanopassiglia]

I can't even explain how much I love these videos

[@lewistaylor863]

For a piece of test kit I made at work a few years ago, I needed to be able to log the position of a carriage along some linear rails and take that data, along with some load cell data, into a computer. The main catch was that the rails were 1.5m long and I needed accuracy to be within about 0.5%. There were some expensive options, but in the end I made a cheap and accurate enough system using a 600 count rotary encoder and a gt2 belt. The encoder was fixed at one end with a gt2 pulley (I can't remeber the tooth count, but it was the smallest I could find), with an idler at the other end, and a belt loop spliced on the carriage. To get better accuracy, I hacked the stock quadrature encoder library to count based on each of the quad pulses (rather than taking the four pulses and outputting one count). If memory serves I could get within 0.02mm over the full length of the carriage (using precission reference rods as a calibration and verification tool). Belts, pulleys and encoders are dead cheap! I used an arduino and data streamer in Excel to manage the hardware and software requirements.

[@jaro6985]

Yeah you can get rotary encoders based off either optical or magnetic like this one, that are much better constructed, so you don't need to worry about radial runout errors etc.

[@JamesNewton]

My friend Ensari and I did this a few years ago and got better results; 1/4 thou instead of 4 thou. But we used a thin line (spiderwire) wrapped around a shaft in a bearing, with the magnet on the end of that shaft, aligned with the chip. Because the shaft was smaller than your gear, we got better resolution; more rotations per unit of linear movement. Also was very easy to make, no 3D printer required.

[@horatioswrld]

As someone who this is all well past for, I want to assure you that you Mr. Wandel that shouldn’t be a concern of yours. I’m locked in. The more you do it, the more I learn and suddenly I’m checking out the sine waves for my solar generators. Keep KZbin niche.

[@NotaRobot_gif]

Always so interesting, and I always learn something new. Thanks for sharing these silly projects!

[@wiseoldfool]

Another really interesting experiment!

[@LeesChannel]

What a fascinating video! I was hoping it'd keep going.

[@ronaldutrecht_nl8738]

This is why I love KZbin. Keep on going to make videos like this!

[@thomashverring9484]

This was so awesome! I didn't understand all of it, but still super interesting and entertaining.

[@chiparooo]

Very interesting investigative work! Thanks for sharing!

[@mrdrbernd]

Videos are always fun to watch, educational and inspiring. - great workmanship. Creative. Please do more videos.

[@LucasHartmann]

About the spinning compass. You don't have to approximate it, you can just have a full table lookup. Repeatability is what matters.

[@eric13hill]

Your talent and creativity are inspiring.

[@HexenzirkelZuluhed]

You still crazy infotainment! I hope the Scale readout works out, I'd be interested in the approach and the results!

[@lukahrastovec5022]

Some cheap Chinese calipers have internal UART outputs, and your linear DRO might have it too. SSI and BiSS/BiSS-C are also common encoder interfaces, but both include an additional clock signal, which wouldn’t match to only "asynchronous" signal on the scope.

[@tseckwr3783]

true. Even if there is any inaccuracy, it is consistent as far as I have observed. Easy to develop a calibration table for the caliper and the table.

[@a1k0n]

It looks like it's missing a start bit if it's a UART, though, otherwise signal traces would always be the same length since the scope would trigger on that; I would guess there is an additional clock signal here.

[@MRrwmac]

Very interesting. I’m glad it worked out for you!

[@artursmihelsons415]

Excellent video as always and great experiment!👍 By the way, same signal is on cheap plastic calipers as extra data output available on pcbs.. I looked few of them, and they have it.. 😉 I had an plan to modify them as cheap DRO with adding ATmega... How I read, that signal is exactly the same as from DRO sensors...

[@gedtoon6451]

I liked the section on Fourier Transforms where you realised your viewers may be as intelligent as you are!😊

[@AdityaMehendale]

3mm cubes in a Halbach-array (linear, anyway) --> way more "sinusoidal" transitions! with two Hall-effect sensors (TI makes some good ones) with a 1/4-pitch difference, you would have an excellent quadrature-encoder, far better than 0.1mm resolution. (realized by the small-ness of the magnets). The fact you are running it as a resolver means that you would be relatively immune to amplitude (distance) variations.

[@AdityaMehendale]

Clarification: "quadrature" --> sine and cosine pair "Resolver" --> Phi = atan2 (sin(phi)/cos(phi)) --> irrespective of amplitude/magnitude. (Although I am sure MW already knows this ;)

[@dosgos]

I thought CuriousMarc had a related NASA video. I'll see if I can dig that up.

[@69dblcab]

@@AdityaMehendale MW might but I didn't. Thanks

[@ericwazhung]

atan2(sin(phi)/cos(phi)), independent of amplitude, great, assuming amplitude is the same for cos and sin... yes? What if the amplitudes differ due to slight offset-differences, etc? Or, one sensor's being closer to a stronger magnet and the other to a weaker one?

[@AdityaMehendale]

@@ericwazhung I'd be far more worried about the magnetization-direction, placement-errors, stray magnetic fields and thermal elongation of the base material. Things can be "better" without being "perfect". Things that are "good enough" don't even need to be better. What I propose is an order-of-magnitude "better" without the need for complexity that far exceeds that what is already present.

[@TKC_]

I’ve had a similar need though for much longer distances. I used an allegro a1230 quadrature bipolar Hall effect sensor. Each change in magnetic pole is sensed and things like flat fridge magnets can have quite dense poles. You would need to use a couple offset to get enough resolution or use gears. Probably easiest to hack that cheap one you bought.

[@FiveEights-.625]

Great ideas, always stimulating, lovep watching. Sometime ago you mentioned there was some runout in your ER collets. I also had purchased an inexpensive ER collet set and noticed the many of the slits had metal chips in them. I didn’t measure their runout but really think removing the chips is gotta help.

How about a Fourier transform? :-)

[@Argosh]

Was about to ask that myself 😉

[@MCsCreations]

Pretty interesting stuff indeed, Matthias! 😊 Stay safe there with your family! 🖖😊

[@stuartgray5877]

Something we use at work (built the deployment and focus mechanism for the James Webb Space Telescope) is called a LVDT or " Linear Variable Differential Transformer". You can get extreme accuracy with these devices, but requires some EE skills to build one yourself.

[@matthiasrandomstuff2221]

On a project I worked at at the university as an undergrad, it used LVDTs and an A/D converter to sense position. Simple and reliable. I went looking where to buy one, but these things are quite expensive!

[@jeffreybernath6627]

I love the Lego Technic bushings, which first show up at 1:10! 😀

[@Don.Challenger]

Mr. William Ockham always suggests trying the simpler before the more complex and this is another good demonstration of that iterative improvement that relates science, mathematics engineering and construction in slowly perfecting our built world. Note that sometimes an improvement requires shelving one approach and proceeding by adopting another one originally thought more demanding. Matthias is a great proponent of that procedure (each time he circles back on one of his projects, improvements multiply - pantorouter, bandsaw and the rest / but only where the improvement betters the practical requirements, where good enough suits he tends to rein himself in.)

[@AngryArmadillo]

Well said. I think watching him put this mindset into practice over the years has made me a better engineer.

[@claudiomenesesc]

Entertaining and mindboggling (at least for me). Regards from sunny Peru.

[@pewpewlasergunz]

You could probably try salvaging a bunch of mini electro magnet coils instead of hoping all your permanent magnets were cast perfectly. Also quite surprised to see this upload as not only was I about to do the exact same project but also the exact same milling table, however I'll be using optical linear and rotary encoders I salvaged from a pair of old desktop printers.

[@tlum4081]

I wish someone would donate a Series 1 Bridgeport milling machine to Mathais. However, getting it into his basement would be a problem. I do admire his attempts to get precision with bearskins and duct tape which are entertaining.

[@m-vo]

Really enjoyed the video and the results! My first thought was using an ultrasonic sensor to directly measure the distance, though I don't know how good the cheap ones are in terms of resolution.

[@matthiasrandomstuff2221]

Played with them. Low precision, inaccurate and unreliable

[@viktorbublic]

My brain hurts but I still love your videos.

[@TimoNoko]

I recall James Bruton took those ultrasonic echo meters apart, so that it metered direct distance between transmitter and receiver. And result was very stable.

[@capermache]

After the apocalypse I want to be on Mathias’s team.

[@Proud2bmodest]

Use a non stretchable cord wrapped around an optical rotary encoder with quadrature output to give direction. The cord can be kept taut by using a weight hanging off of one end. This becomes an absolute encoder and can compensate for backlash.

[@madmodders]

The data format is usually either BIN6 or BCD. Six four-bit nybbles.

[@GilesBathgate]

AMS who make the rotary encoder chip also make linear encoder chips. Good luck finding a source for those though, and the linear magnet strips. The compass idea is interesting though.

[@meeponinthbit3466]

LOL... Our man knows his audience.... Nerds. 😂❤

[@lumotroph]

Fantastic. Your watchers are all such nerds, and we love this! 😂

[@alibronx2112]

Love this!

[@ChatNoirLe]

You could try some magnetic encoder tape, it is pretty cheap with fine pitches 2+2mm. Off center can be pretty annoying, I'm planning a project with a MT6835 and even with a 10mm dia magnet you still need to be preferably within 0.1mm centering and even with that it's got a whole bunch of DSP linearisation going on, so you got pretty good results for a bit of wobbly heatshrink.

[@semerhi]

A Short-time Fourier Transform might be able to capture localized frequencies, and thus factor in the differences between the magnets

[@gamerpaddy]

you can make yourself a sin/cos absolute encoder very easily: get two linear hall sensors (outputs a voltage around vcc/2 depending on the magnetic field and polarity) feed those two voltages into a microcontroller adc and put the reading straight into the atan2(a,b) function. it will output degrees. just use a diametrically magnetised magnet (or a normal magnet on its side) and place it above both hall sensors where one of the sensors is rotated by 90°. depending on the position and how close you can get the sensors together, you might need to compensate for some non linearity.

[@matthiasrandomstuff2221]

But even easier to find than a linear encoder is a rotary encoder. So why bother? Fact is, if I needed a linear hall sensor, I'd probably just use a rotary encoder and work it back from angle and magnitude, cause the linear ones are way more common.

[@gamerpaddy]

i needed a absolute angle encoder for a project and didnt wanna spend big bucks on one, incremental didnt work. and incremental can lose steps if the microcontroller cant keep up counting. some industrial absolute encoders work on a optical disc that has sections that encode their angle in binary. sometimes even with multiple gear stages and further discs to count thousands of turns and still keep their absolute position, no matter what the microcontroller is doing. combined with a spring loaded pulley whre you can pull out a string to measure distance it can be a proper linear position sensor.

[@FrankGraffagnino]

i'm looking forward to a video of hacking into those linear encoders. i think a github repo that could share how to do that and the link to the part number you used would be really cool and useful to lots of people. especially with videos to go with it!

[@StuartVonTRT]

search for digital caliper serial data or cheap digital caliper serial ... you have there one clock line and one data line, and I think 24 bits of data from which first (probably) 10 are some sort of status, rest is position (number which you divide by something to get position in mm or inch) . Or something similar like this. Most of china cheap calipers have same or probably very similar protocol

[@matthiasrandomstuff2221]

Alas, that's definitely not what I have here.

[@andrewianchen5700]

@@matthiasrandomstuff2221 Looking forward to the follow-up video about the encoding. I wonder if they used something different since those encoders look a bit fancier. In case anyone is looking for the encoding on those ubiquitous three button Chinese scales, the encodings are two 24-bit words, two's complement, little-endian. One for assumed absolute position and one for relative. With 20480 bits per inch.

[@osgeld]

the digital signals are fairly easy to get, if you think 1970's RC type stuff, each digit is within a time period then the pulses per time period give you an 0-X readout. start off at zero and increment very slowly you should be able to pick out the time period pretty quickly. its similar to asynchronous serial but in some special herbs and spices recipe

[@3ATIVE]

Could getting data out of Stepper Motor coils (with some gearing) work?

[@krisholt]

On my lathe, I have an rotary encoder attached to the main slide, and a toothed belt attached in both ends of the lathe bed so the belt rotates the encoder when it passes over it. The signal are transferred to a TouchDRO from Yuriy's Toys, and the precision is under 1/10 of a millimeter. on the cross slide I have a real glass scale with much higher precision.

[@musthavechannel5262]

leave alone the digital signal recognition, just do OCR on the screen, haha. This would be a portable solution in case you change the measuring slider thing in the future.

[@AK-vx4dy]

About magnet line, i observed that chip is used in diffrent postion (prependicular vs parallel) than in rotary, this may affect accuracy, also they are spaced more away, so stray fields have more input.

[@thenutdriver9685]

A temposonic would be your best bet imo. We use temposonics in our cylinders for position feedback. Basically a tube that goes through a magnetic field,,,going to an encoder.

[@MrQuickLine]

Martin just mentioned you in his video and then you released this almost immediately after 😁

[@BuddyCrotty]

You got a shoutout on the new Wintergarten video today!

[@dominpiano]

Give this man a 3D printer and he will do magic with it!

[@RnXeKrNe]

If that's an AS5600 encoder it might work better if you don't let the magnet touch any metal. That's probably the case for other hall effect encoders too. I just learned about this myself after reading up on encoders for BLDC motors, specifically on the SimpleFOC community forum and their docs.

[@DragonX5X]

Very interesting!

[@joruss]

Remember pre-optical mouse? It had two optical encoders on board and chip that dumps ready to consume data steam via serial or ps/2

[@kazzle101]

On the youtubes Curious Marc had a look at his broken DRO (late August 2024: Modern Fails: Newall C80 DRO repair), it uses a row of ball bearings instead of your linear track and some sort of magnetic magic and sensors simmilar to yours to find the position.

[@misterdudemanguy9771]

The signal could be SPI. Look for a companion clock signal. You should be able to interface with 3 wires, ground, clock, and data. An Arduino would be perfect, IIRC it has an I2C/SPI port. If not, bit banging it is trivial, and there's drivers already made for that. Use RS-232 or USB to the PC and send it the converted measurement. Doing this you could capture all 3 axes and send the PC data packets with X, Y, Z.

[@senorjp21]

You could use a camera to read the numeric display and use opencv to do optical character recognition. No mod required, but more software required

[@IlusysSystems]

That's probably better resolution, than I would expect from such system. There are a few videos where images were read from PC mouse, I wonder if mouse sensor would be viable as a position sensor and what the resolution would be. Difficulty would be printing the encoding on some medium, so instead, you would probably have to do what callipers (kinda) do - print bunch of triangles that you would have to interpolate to get position.

[@pjsparks4845]

That last bit reminds me of linear closed loop stepper motors - I wonder if there's anything worth exploring in that to get a position readout

[@longshot789]

Love seeing Lego show up

[@Farlig69]

8:07 We have faith you'll figure it out ;)

[@CrudelyMade]

you might also consider, instead of a magnet, use a slotted 'disc with a light meter. there are a variety of mice that use this as the internal mechanism that measures how the mouse moves. but.. now that I think about it.. you could also but a cheapish mouse and use its light sensor to watch the gear rotate from one side (instead of head on).. and this would translate as mouse motion. and with some mice you can change the resolution/sensitivity which can improve accuracy. and the mouse internals are already made to feed data into the computer via usb or bluetooth or rf. the test for this would be rather straight forward because you can position a mouse over the rotating gear (where you have your magnet).. and you could make that part slightly larger diameter to get an easier/more accurate read. but.. that might have been another 'diy' solution. ;-)

[@phrozenwun]

Instead of rack and pinion, what about a windlass style coupling? That should help remove gear lash and tooth variation issues... very cool all the same, thanks for sharing.

[@ScramblerUSA]

You could try using a set of cheap digital calipers. They have an output port on the board (some even have it exposed in a case). And the data format is already documented on the internet and is easy to use since it's a data/clock pair.

[@rasimbot]

You can make an oscillating loop with a laser diode, a mirror, a photodiode and a logic inverter. And then count oscillations with a binary counter per ms...

[@bpark10001]

C'mon, Mattias, you have good machining capability available! What you have now will be limited to telling you which end of the milling table you are at, with sloppy wood bearings, shrink tubing, & 3D printed plastic gears. Break out that lathe & mill & get stuff properly centered! All that calibration you so carefully measure will change as fast as the tide. (You don't need Fourier series, just a big lookup table of corrections with linear interpolations. The lowliest processors can handle them!) To implement GOOD rack & pinion, get 1' caliper & put encoder on the dial. The "open air" scheme will be perturbed by any ferrous stuff in the room closer than the nearest black hole. A better scheme would be to take coarse threaded steel rod, with steel gear with tooth pitch matching the thread pitch, mounted almost touching. Magnets on the gear cause it to be attracted to the threaded rod threads, forming "magnetic rack & pinion". That can be read out by magnetic encoder. You need magnetic shielding around the encoder! The worst problem with this is stiction of the gear. Mount it on precision ball bearings. Those linear readouts are available with readout capability (but readout may be too slow for CNC work).

[@ddutton0]

I did not understand most of this but got that Mathis is super smart 😅

[@NeitSotm]

Just try to use that last measurement (with the magnets in a row) as a calibration! You can simply interpolate the data points that you have. You only need to remember which cycle you're in.

[@dr.kraemer]

For some of us an FFT is much more familiar than the rest of the work you present here.

[@xl000]

Imagine people during the first industrial revolution slowly improving methods , materials, measuring device, but not as a hobby or a thought experiment. Like going from a wire that randomly fits through a hole, to a wire that fits consistently through a hole and create the exact same friction. Are there any books on this general subject ?

[@reddcube]

A flexible fridge magnet is usually a tiny Halbach arrays (alternating north/south poles in a stripe pattern) I wonder how accurate a position you could achieve with the sensor and a flexible magnet tape.

[@JernD]

Have you tried the gnumeric spreadsheet before? It is fairly similar to Excel 2003 but I like the graphing in gnumeric a bit more.

[@ColCurtis]

You could buy just the dro scales. They come with 9 pin connectors. I'm sure you could easily figure out how they work

[@Bbonno]

FFT is something I've never actually seen used. Explanations on what it does usually end with by simply saying that it's super important for 'lots of things '...

[@klave8511]

There’s always the good old mouse. Supposed to get 600dpi. The wheel has less resolution but uses a simple optical encoder.

[@TehMegaNoobZor]

Had no idea excel could do fft. Cool!

[@letsgocamping88]

Have you seen curious Marc's video fixing his dro and explaining how it works?

[@PatFarrellKTM]

Sure glad you ran a DFFT, otherwise, I would have had to ask about them.

[@Urban_Spaceman]

Just watched the "Square drill" video from 4 years ago (the Polish flat pack fixture drill) . Could you use that in this "mill" You were complimentary about it's build quality and accuracy.

[@GameBacardi]

I wouldn't believe that you can use magnet like that way until now I have seen it :)

[@lezbriddon]

The magnetic variation in absolute position reminds me of the bumps in 3d printer beds, so they do a a read of multiple points and create a look up table with error values to compensate. cant you take your readings and create an offset lookup so it can be driven to that value?

[@vasiliigulevich9202]

I wonder if instead of functional approximation, a set of tables associating angle and posotion in monotonic sections would work better.

[@handdancin]

you could probably make a pretty good dro with a couple draw wire encoders. dont know why you would, but it might be fun

[@matthiasrandomstuff2221]

True, if I want to make something more expensive and less accurate than a real DRO

What kind of precision do you want to achieve with it? I though 0.05mm was plenty good for woodworking.

[@mirkoragni3118]

Such a great video! By the way, Am I the only one seeing a DIY CNC coming in the future?