Suspension Data Acquisition

https://www.vitalmx.com/features/vital-mx-pit-bits-2026-anaheim-1-supercross

More picture here. Looks like the have a mount off the case.

Surprised that you don’t see a rotary pot and a link arm being used, far cheaper and more robust.

Been using string pots a lot recently as I can’t trust the guys using them to not bash them off. I’d like to back to back them with linear pots to see the response difference.

I kind of wish I got sensors for my moto.... maybe another year. The moto feels so good that its hard for me to know if I am that far off but I would still be interested in the data.

I bought a couple of these sensors years ago - they are ride height sensors used by GM, super cheap and pretty much bomb proof. A bit bulky and they run on 5v so I haven't been able to use them on a bike yet, but I still think are a good option. V8 supercars do/did use them on racecars, plus Motec sells the exact same thing for about 5x the price. You can buy/make different length arms and brackets to mount on different bikes

https://www.ebay.com/itm/177392721144?_skw=Delphi+Right+ER10031+Suspension+Ride+Height+Sensor+for+08-11+Cadillac+CTS&itmmeta=01KF994E16FQD5Q7BV6WYHYF0B&hash=item294d6e1cf8:g:W9UAAeSwYyhovDQS

FWIW you could probably make something identical with a rotary potentiometer and some 3D printing. Granted, it won't be weather sealed like this one is... 

Yeah rotary pots can work, I've seen people make angle sensors and even a type of string pot from a basic Rotary potentiometer, something like the Motec steering sensor here. You lose a little bit of resolution if its only working in a small portion of its range (a 1 turn Pot with 260ish degrees of working range and a 12 bit ADC/4096 values measuring a link with 45 degrees of rotation and a 65mm stroke shock is roughly 0.1mm) vs a 75mm linear pot that gets about 0.02mm on the same system. It's probably the easiest one to integrate with any system though, I've looked at throttle position sensors that are non-contact magnetic sensors which look very tidy but again seem to all use a 5V output

Funny to read here about string potentiometers and link driven potentiometers too 😅 A few days ago I noticed a pretty big flaw in my system. The rear sensor is mounted on a pivot of the rear linkage system, similar to Motion Instruments. zero position and max position are calibrated, the rest between is handled completely linear between the two points. But pretty much no pivot on a bike moves linear through the travel (I don't know how I didn't notice that till now)

Because of that and because of the annoying mounting differences between each bike I'm currently redesigning the rear sensor and currently exploring exactly those two options (Still using an AS5600 rotary encoder though).

The new version will probably use a rear wheel travel > sensor travel calibration to get the best and most precise result possible.

Will let you know once I have more to share :-)

FWIW Suss My Bike also used the rolled up string in their concept. They were an early shockwiz competitor. 

You could use a level converter to use it with a 3.3V microcontroller. But this adds another layer of complexity. Maybe they even have limitations in regard to updatespeeds (I have no experience with them, just as an idea)

Yeah I experimented with a voltage divider and it seemed to work OK, but felt like it would need too much extra work to integrate with the system. As in adding an extra 5v power supply then either making a small board or maybe an inline connector/adapter with the extra circuitry....so kept look for other sensors. 

There could be a way of making a small module that "breaks out" the 5v sensors and houses a small battery too? Maybe it doesn't need to be that complicated but I've been keen to try adapting the Sram AXS batteries to power a logger since they are pretty small and commonly available now....

Hi all, 

I would like to try the BYB on different Bikes.

- Supernought Forbidden

- V10.8 Santa Cruz

The Bikes are not yet here any mounting experiences with these bikes?

Best regards 🙏

I've been doing some more testing and slowly have inched up in air pressure in the fork. I did try full open compression and full open rebound for a few laps as that is what the auto tune was telling me to do. I can say it was very active but felt uncontrolled and I found my self wanting more support or getting offline easier than I wanted.  I slowly crept up in pressure and compression but didn't want to loose to much front end feel. I tried to then get the rear to have similar compression speeds but could never get it as slow as the front. I found a spring that I had laying around and thought it would be interesting to see what would happen if I dropped about 100# and get something closer to 30% sag. I figured I would have to run more compression and faster rebound to not use travel to easily. I was expecting to have faster compression speeds and slower rebound speeds which I did but not as much as I expected. I'd like to try a spring between my 343 and 325  and a shim stack with more HSC and lighter rebound tune. Both were completely ride able although the softer rear setup did get the faster time for the day. I also felt like I wasn't being pushed forward/riding as high up with more rear bias sag which helped but I was also feeling that I was using travel easier.  The duration of the run is about 2 mins and I was able to put a start/stop in to compare runs and times which is a nice feature to trim everything down. 

First prototype of the new rear sensor is on the bike! :-) The sensorarm is waaay too long but I did that on purpose to test different positons for the ballhead.

Update from the Thinker Garage ;-)

The new sensor and calibration are finally implemented. While calibrating the system, I now create a map between rear sensor data (the link arm sensor) and the rear wheel vertical travel (the fishing rod looking sensor). This map is then used while logging to know which sensor value means how much rear wheel travel.

Main advantage: No leverage curve needed anymore unless you want to reverse calculate the shock movement. But honestly I haven't seen a use case for that yet in terms of bike setup. Do you guys use raw shock movement data?

First testride, without the "fishing rod" of course, is yet to be done.

The Honda program was legit 

@enricorodella22 Is it possible to train an AI model using this data and have it provide insights or recommendations? That would be a really nice feature. I’m not a data scientist, but maybe someone with more expertise in this area could weigh in. The sensors already collect a lot of information but I find it difficult to interpret and use it to make adjustments.

AI isn't amazing at providing insights or recommendations for "novel" things. It's great at historical pattern recognition and replaying patterns into the future. The problem with trying to train an AI model with the tooling we have now is kind of twofold. 

First, wholistic system data. To recognize patterns that would be meaningful, the model would have to know more than what the position sensors can provide. It would have to take into consideration things like terrain (which you could in theory get from gforce data) but also rider body position and dynamics. It would have to know (from historical context) how you would respond to various terrain features (like if you tend to land nose heavy or crumple in g-outs) for it to start providing any real insight in a meaningful way. The more variables a system has the more quality data it needs on each of those variables to make a decision. 

Second, model targeting and outcomes. Models are trained by coming up with solutions deemed "good." We've done this for Large Language Models by giving it a book and saying "this is a good book" or an image saying "this is a good image." This is kind of in conjunction with the wholistic system data point above, as not only do you need a LOT of data, but that data also needs to be flagged as "good" in some capacity. What is good suspension? What is a good setup? Not to sound too philosophical on that but the fact that so many different setups can produce similar outcomes (like lap times) makes this a very hard problem to solve.

Also, I'm not convinced (at this point) that pumping all this data through an LLM would give you any different/better insights than the algorithmic based approach BYB and Vorsprung's Tuning Hub provide. Those work by having target values and either comparing that against positional measurements (BY or by doing some math (expected force at shock based on rider weight, rear suspension kinematics and "rider aggression special sauce"). So I don't think you'll see any improvements to what those things provide by giving the raw data to the LLMs.

However, I am in the AI space for my job and working on various processes/tooling to help deal with physical AI and robotics stuff and I desperately want to get a BYB system that gives me the raw access to the data to start doing some cool stuff around plugging data into models and producing outcomes. But just asking to chuck AI at it is similar to how "everything needs an app" felt when smartphones came out, AI is a hammer but not every problem is a nail. 

With all that being said, I think someone like BYB could use AI to help explain conceptually what's going on and why its advocating for the recommendations that it is. But honestly you could do that yourself as is by just taking some screen shots and having a chat with Claude.ai. 

Also all of ChatGPT are trash and I cannot wait till the world is free from that pile of garbage. What shit models. 

Which software are you using to upload your photo to? I searched around a bit and didnt really find what I was looking for. 

It's probably the linkage X3 software, I don't know about many others that can stuff like this (The syn bike software maybe too)

https://www.bikechecker.com/

Linkage

Hey folks, thought I'd share my DIY data acquisition project. It's been about 6 months in the making (so far) and is up and running. My intention, once it's cleaned up and a few more of the wrinkles are worked out, is to open-source this.

Hardware-wise, it's based on an ESP32 dev board on a custom PCB that breaks out the connectors and does some simple analog signal processing. It supports 4 analog channels and digital sensors on I2C. It has a small OLED screen and keypad, and optional handlebar-mounted switch for starting and stopping logs and marking events of interest.

Apart from logging data (obviously) the logger runs some simple quality-of-life functions like sensor calibration, turning sensors on and off, and setting the sample rate (I've tested with 2 sensors up to 500Hz and have nice clean signals, at least to my eyes). The logger also supports wifi access for more detailed setup and downloading files. Optionally, you can apply transformations at the source (e.g. shock travel to wheel travel) either as a lookup table or polynomial. The firmware is designed to be extendable to other types of sensors but for now it's limited to analog inputs.

Sampling and recording sensor data is simply (ha!) an engineering problem; deciding what it means is of course the hard part. With this project I'm aiming for somewhere between a black-box “just trust the result” tool and dumping the raw data on the user: the goal isn’t to provide instant answers for everyone but rather to build an extendable set of tools that are useful straight away and provide a platform for "power users" to take further.

To that end, I've built a data pre-processing pipeline and some general-purpose widgets for browsing and analyzing the outputs. These are Python-based and run on Jupyter Lab, a free and open source interactive computing environment.

One aspect of the analysis pipeline I'm particularly pleased with is 'event detection from schema'. Users can specify in a configuration file particular things they are looking for in the data - deep compressions, say - and what they want to know about them from the available data series. This is a work in progress but is already quite flexible and powerful.

I've been running the system on my Stumpjumper Evo with a couple of AliExpress linear potentiometers and 3d printed mountings:

I'm conscious I'm jumping into the middle of a thread that is 12 pages deep, so if you think this belongs somewhere else, feel free to let me know. Otherwise, any thoughts, feedback or questions would be appreciated. I'm also on the hunt for people who would like to have a go at building one. It's not a beginner project as such, but none of it is actually rocket science either.

If you've made it this far, thanks for reading :-)

Sunn and Voullioz were earlier than that. Images below are 1998 And I imagine someone/team tried even before that? Nico and Bruni's mechanic Jack would be the guys to ask. Cool we are now at a point in DH MTB that we can chat about the "history of Data acq". It's certainly come a long way. I'm also old enough to remember Volvo-Cannondale team using data acq at some stage in the late 90s, but can't remember for sure (part of old age). 

https://www.instagram.com/p/C2b1kqJxNAO/

Seems to be a DAQ box of some kind on this Cannondale? But no sensors visible?

I think this is the correct place for this. What's the sampling rate you can go to? Does it stop at 500 Hz?

Who made that mount for the SJ link?