Kettler Stroker

[Double-A-92]

I'm trying to create a configuration for my "Kettler Stroker" rowing machine.

The rower originally comes only with reed switches on the handle drive wheel, which is not optimal.
Luckily the rower is easy to disassemble, so I added a threadmill reed switch (it already came with the nice cable) and 4 small magnets on the flywheel.

This is what it looks like on the inside if anyone is interested:

For the brains I used a cheap "ESP32-S3 Super Mini" from AliExpress. (I can create a pull request for the PlatformIO config I made for it, if it's welcome.)
My testing contraption with a salvaged vape battery:

Now I'm stuck with the settings...
I was bored and CAD modeled the actual flywheel, which gave me a moment of inertia of 0.0265 kgm²

Kettler Stroker Flywheel.zip

For the "sprocket radius" I measured the belt pulley directly attached to the flywheel axis. It had an average radius of 1.6875 cm

The stroke count seems to work perfectly without tinkering with any other settings (besides the number of impulses per revolution).
The Force curve also seems plausible (not sure about the absolute values though).

But I can't get the distance and drag factor to work consistently. I guess something is wrong with my Drag factor filter settings...
Is there anything I can log to discover the proper settings to get this to work?

[Abasz]

thanks for the interest in this project!

it's possible to simulate workouts on the computer that helps witht he calibration if you have the deltatime data of a recorded session.

I have a few question I need to understand before I can help:

• which profiles are you using are you using runtime settings enabled?
• please send me your current settings for the rower
• asuming you are using the web gui you can export the delta time data in json please post thatit here (assuming recording is enabled) .

There is an inherent ussue with reed switches namely that they produce bounce. so we need to see what is being recorded by the mcu to deal with the potential bounce.

[Abasz]

I guess on this rower only the first 3-4 levels are actually meant for rowing, and the rest is more like a rope machine for strength training.

Yes I totally agree. When I saw these readings I actually searched this rower and I found pictures where they do strength training on it :D::D

I will also create data for each damper level (1-8), to see which levels still produce a high GoF.

Yes that is a good idea. Also it would be very useful to see how the algorithm in general behaves in these cases. To be fair, alternative option is to increase the inertia (e.g. by adding more weight to it, though not sure how feasible this approach is :)) of the flywheel. this 0.025 is extremely week and small. Concept 2 for comparison has 0.1.

[Double-A-92]

I didn't manage to get e2e running. CMake is already giving me errors with it. Does it work on Windows or do I need to run try it on Linux?

I'll upload my current test files for now:
5-strokes-per-level-test.txt
kettlerStroker.rower-profile.h.txt

Not sure if the deltas are usable though... I used ENABLE_DEBOUNCE_FILTER and it seems to have discarded some of the smaller deltas (see the sudden step in each stroke).

[Abasz]

Can you post the cmake errors? I think the issue might be that make e2e-test expects the environment name (i.e. the name of the target in platformio.ini) but I would need to see the full thing.

So in the platfromio.ini you should have an [env:ketlerStroker-board-name] with correct defines. Then you can call ./build/build-e2e ketlerStroker-board-name

Indeed this is rather strange:

Not sure why this is happening as DEBOUNCE filter should not change the total amount of time and create a drop like this as it merely concatenates two. So when its not working you should get jumps instead . The good news is that there is a way to debug this:

in the GUI please set the log level to verbose and do a few strokes with serial monitor attached. It will output rawImpulseTime which is the actual time (not delta) of the impulse before the filter. With that we can compare what is going on. Should you want you can use an online serial monitor: https://googlechromelabs.github.io/serial-terminal/ which is very simple to use.

[Double-A-92]

I finally got the e2e-test to build and run!
I don't know if there is some easier way, but on Windows I had to install WSL and use that to run all the commands. And also make sure that GCC is installed and configured for CMake (by default it tries to use the Visual Studio C++ compiler...).

Now I'm not sure what to do with the output.txt. I can look at the force curves, but how did you get other things like the GoF?

[Abasz]

Yes, the e2e currently only works on linux out of the box. On Windows it can be made working with GCC under MinGW (meaning it can be compiled under MinGW but the support scripts are in bash). Any way WSL is a good choice, I develop under linux

So the output.txt cab ve anaylised. Generally you want to do few things:

• Plot the deltaTimes in an excel

• Plot the Force Curves in excel and see their shape

• change any settings rerun the compilation and see how things improve or get worse.

There is no need to worry about the goodness of fit in general but if you want you can put more logging statements (printf) in places you want to see things. Actually you can even run the vs code debugger with the e2e target if set up correctly (though this is probably an overkill).

The straight lines are where the algorithm detects the stroke, for instance if you increase the minimumDragTroque you will see that the lines move forward (recovery detected earlier as higher force is allowed). This may chop off the end of your force curve, but may provide more consistent stroke detection.

You can plot any metric you select, for instance power:

Its clear from this is power plot that on the 3rd setting something start misbehaving as the power is somewhat low.

Also I think you are better off disabling the bounce filter as currently it causes more issues than things it solves :(

[Abasz]

can create a pull request for the PlatformIO config I made for it, if it's welcome.

PR's are welcome thanks! I am less interested in the board profile and more in the rower profile:) you will need to include a few recorded session delta times with the proper naming (its in the test calibration folder) thrt includess the expected number if strokes so testing can happen when i change something.

I tried the super mini and i measured a deep sleep current consumption of almost 1ma so I decided its not worth it. But the onboard charging circuit and the very tiny size is cool.

I use now days a zero mini with AA battery: https://www.printables.com/model/1438697-waveshare-esp32-s3-zero-box it has half the feep sleep current.