ACTIVE TOUCH Rowing machine (Aldi) [WIP]

[MKO1640]

Coming from #29

this is my first attempt at implementing this. Therefore, I'm grateful for any tips.

It's a magnetic rowing machine that I acquired used and broken 2-3 years ago.
Unfortunately, the existing sensors are on the handle and can't be used.

I started up my CAD program and started designing a magnetic mount for the flywheel. Six 6x5mm neodymium magnets fit in the mount.

By adding or removing magnets, either 1, 2, 3, or 6 magnets can be used.

I split the mount into two parts to avoid completely disassembling the rowing machine. I hope the accuracy is still sufficient. The screws will certainly need to be tightened evenly.

https://www.thingiverse.com/thing:7294422

I'm starting the initial tests with an AZDelivery ESP32 DevKit C, and I also had a 49E linear Hall effect sensor breakout board with a digital output and adjustable sensitivity.

The 49e's mounting on the crossbar is somewhat improvised, but it can be adjusted well and precisely.

Because of the imperfect linear sensor and the fact that it is still uncalibrated, the data is currently quite inaccurate.
(6 magnets)
2026-02-13 16-03-41 - session.csv
2026-02-13 16-03-41 - deltaTimes.csv

However, an ESP32 S3 and a KY-003 digital Hall effect sensor, which are on their way to me, should solve the problem.

[Abasz]

So this is pretty bad data:

The issue is not the "wolf tooth" which is systematic (and relates to magnet placement and algorithm can clean it up massively) but rather the jumps (missed impulses). I mean we could try to use it and there is a high chance that the algorithm would be able to handle it in a large extent but its less than ideal.

Can you explain how you are using the linear sensor or which one you have? There should be a comparator on those when they produce the digital signal and my experience is that they work fairly ok and provide consistent switching.

What you also need to be aware is that these magnets are rather beafy which is fin on its own but it is important that they do not interfere with each other (meaning that they produce clean signal) so if they are too close they could prevent proper switching. Data shows that there is systematic error but there are random noise which is surprising for hall sensors (though again linear halls work a bit differently):

Which suggests inconsistent switching from the sensor. I think there is no point on trying to use this data and better waiting for the proper sensor board. Note the esp you have chosen is ok, S3 will not help in this regards (but it is a better board).

There is already a magnetic rower (kettler stroker) and please review the https://github.com/Abasz/ESPRowingMonitor/blob/master/docs/faq.md as there are important statements for Magnetic rowers. This is the discussion that we had over this machine: #20

I am happy to help with calibration I just need the delta Times once you have a cleaner signal.

Also just as a side comment, recently someone used an IR sensor with coloured discs: JaapvanEkris/openrowingmonitor#182 the good part in this is that there is no magnetic interference (since yours is a magnetic rower). If the new hall does not work, its possible to switch to this IR as a fallback. Everything works the same though it will use more battery (and potentially you need to manage powering the IR board off in deepsleep if you want full battery operation)

[MKO1640]

Unfortunately, I can't say for sure what kind of A/D converter is behind the 49e. Unfortunately, I couldn't find anything about it.

For now, I've installed this HC-89 interrupt sensor.
The sensor disc is 3D printed.

The readings look much better. Unfortunately, probably due to the 3D printed disc, I'm getting quite strong oscillations and a kind of sine wave.

2026-02-14 19-06-15 - session.csv
2026-02-14 19-06-15 - deltaTimes.csv
The undefined noise seems to be gone.

The KY-003s have just arrived, maybe I'll try installing them on the other side of the flywheel as a test.

[JaapvanEkris]

The readings look much better. Unfortunately, probably due to the 3D printed disc, I'm getting quite strong oscillations and a kind of sine wave.

If it is a repeating pattern that is structural, it can be filtered (latest version has a filter for systematic noise). And the regression filters are capable of handling quite a lot of noise as well.

[Abasz]

2026-02-14 19-06-15 - deltaTimes.csv
The undefined noise seems to be gone.

You can use the e2e simulation to replay a deltaTimes csv and tweak the settings: https://github.com/Abasz/ESPRowingMonitor/blob/master/docs/faq.md#9-how-do-i-calibrate-esp-rowing-monitor

This is kind of noise is fine

[MKO1640]

I wasn't sure how to interpret this zigzag pattern. It seems to happen every other time, and the slight wave then appears every 12th time.
It looked to me like the interrupt on the rising and falling edges was creating a timestamp. My disk is drawn so that it should be evenly bright and dark (every 30°).

That's why I printed another disk where it's only bright for 5°, and the zigzag pattern disappeared.

The slight "sine" wave is still there, though.

2026-02-15 12-08-55 - deltaTimes.csv
2026-02-15 12-08-55 - session.csv

I'm not really getting anywhere with the settings now.
The rudder strokes are being counted and displayed, but the drag factor and distance are both at 0.

[Abasz]

It looked to me like the interrupt on the rising and falling edges was creating a timestamp.

Not sure what you mean by rising and falling. If you have not changed anything in the code delta time is only registered on the falling edge. Rising is ignored. (

ESPRowingMonitor/src/globals.cpp

Line 42 in cbd6c9c

attachInterrupt(digitalPinToInterrupt(Configurations::sensorPinNumber), rotationInterrupt, FALLING);

)

I am thinking that this board already has a pull up resistor built in. You could try disabling the internal pull-up:

ESPRowingMonitor/src/rower/flywheel.service.cpp

Line 18 in cbd6c9c

pinMode(Configurations::sensorPinNumber, INPUT_PULLUP);

by changing INPUT_PULLUP to INPUT.

In terms of settings have you checked the faq: https://github.com/Abasz/ESPRowingMonitor/blob/master/docs/faq.md#15-strokes-are-getting-lost-or-stroke-detection-needs-tuning

Magnetic rowers need changing the max drag threshold and a bunch of other things.

Can you post the setting you used other wise we are not able to help? Are you using the kettler stroker as basis? I recommend starting with that. There is a full discussion where a lot of these issues (drag factor being zero, no distance etc.) have been addressed: #20

2026-02-15 12-08-55 - deltaTimes.csv

The wave is fairly ok. However, on slows speeds there are a lot of bounce that is bad (not a massive issue just not good):

[MKO1640]

Not sure what you mean by rising and falling. If you have not changed anything in the code delta time is only registered on the falling edge. Rising is ignored.

That's exactly what I mean. Since an interrupt is only triggered on a falling edge of the signal, I'm guessing that my signal bounces slightly on a rising edge due to reflection or a shallow edge, thus triggering another interrupt. Since the signal is now much shorter, this second signal will probably be ignored or filtered out. The simplest way to test is certainly to reduce the open areas. Currently, it's 5° across the 6 slots over 360°. A print with 2° is running right now.

The wave is fairly ok. However, on slows speeds there are a lot of bounce that is bad (not a massive issue just not good):

which might confirm my suspicion. Even at slower rotation, something is noticeable that cannot be ignored.

In terms of settings have you checked the faq:

Unfortunately, I overlooked this approach until now, but I'll work through it.

I'm currently using the web GUI and its settings options. I'll try the ??? Kettler 07982-500 ??? setting there and get back to you if I run into any more problems.

So far, I've tried navigating the settings.md file. Unfortunately, the settings options differ between the code and the web GUI, making it difficult for me to assign them correctly.

I also haven't yet figured out how to use the end-to-end simulation in Windows to generate the TXT files for calibration_helper_gui from the CSV files. I suspect this is only easy under WSL or Linux.

Thank you so much for your tireless efforts.

[Abasz]

That's exactly what I mean. Since an interrupt is only triggered on a falling edge of the signal, I'm guessing that my signal bounces slightly on a rising edge due to reflection or a shallow edge, thus triggering another interrupt. Since the signal is now much shorter, this second signal will probably be ignored or filtered out. The simplest way to test is certainly to reduce the open areas. Currently, it's 5° across the 6 slots over 360°. A print with 2° is running right now.

there is a debounce threshold setting which tells the system to ignore any impulse for x amount of milliseconds

So far, I've tried navigating the settings.md file. Unfortunately, the settings options differ between the code and the web GUI, making it difficult for me to assign them correctly.

What is different? if there is difference its a bug on the GUI side so if you tell me or create an issue there which setting does not reflect the code in ESPRM I will update.

Just to add something here, actually I parse the rower profiles automatically on every release : https://github.com/Abasz/ESPRowingMonitor-WebGUI/blob/master/tools/fetch-profiles.ts So I would be very surprised if something is missing. the above link shows the mapping.

I also haven't yet figured out how to use the end-to-end simulation in Windows to generate the TXT files for calibration_helper_gui from the CSV files. I suspect this is only easy under WSL or Linux.

Windows is a bit tricky but cmake will handle it for your. You will need MinGW set up and run cmake -S . -B build -G "MinGW Makefiles"

I recommend consulting with some LLM (chat gpt pherhpase) it will help set this up. But basically you will need mingw and the gcc build tools and cmake you can set up vscode https://code.visualstudio.com/docs/cpp/config-mingw and then cmake will generate the bulid files and build the e2e target.

Alternatively you can use WLS which is linux and there everything should work.

[Abasz]

Finally the deepwiki has updated to the latest: https://deepwiki.com/Abasz/ESPRowingMonitor

This also means that you can ask the AI assisted wiki for explanation of settings. It is very good in assisting but it is new :D So if you get stuck I am here to help.

EDIT:

once you have something that you are ok with I can help you refine finetune. I would need a rowing session of 20-50 strokes and if possible an interval session (i.e. where you vary the intensity)