A DIY sensor to measure handle speed

[Nomath]

Recently someone commented that the C2 Performance Monitor provides the user all the information that he/she could desire. This topic is about collecting information that the PM does not provide. The PM shows us the force curve. But power = force x speed, so the speed curve during the drive is probably just as interesting. There is also interesting information in the recovery.



Let’s have a look the time axis of a complete stroke (figure above). It starts with the Catch, defined here as the point where the handle is closest to the chain gate. I assume that a stroke is not a single stroke, so the flywheel is still running at the catch. The axis that is rotated by the chain and the flywheel are connected by a clutch. This is a device similar to the freewheel of a bicycle. We can only add power to the flywheel when the speed of the handle matches the current speed of the flywheel. At the catch our arms are fully stretched, so in order to move the handle we have to move our body. To get the handle up to the speed the speed of the flywheel we have to bring our body up to speed. This takes time and a lot of power. This is phase-I in the figure. In phase-1 the PM cannot sense what we do because it only senses the rotation of the flywheel. In phase-II in which apply power to the flywheel, the PM can measure our force and the speed of the handle. A short time before the finish of the drive, the handle movement slows down and the PM is again disconnected from the handle movement (Phase-III). This is also the case for the recovery (phase-IV). Clearly, the PM is blind for movements during a big part of the stroke, viz. phases I,III and IV. What the PM or ErgData shows as Drive Length and Drive Time applies to a part of the drive, viz. phase-II.

Let's see how the handle speed is related to the flywheel speed. The chain of the C2 erg is 1/4 inch. That means that each link is 2.54/4 = 0.635 cm long. The chains runs over a 14-sprocket wheel, so one rotation of the axis on which the flywheel is mounted corresponds to a handle displacement of 14 x 0.635 = 8.89 cm. Typical drive speeds are 1.5-2 m/sec. This corresponds to about 17 - 22.5 rotations per second.

One of the options to measure handle speed is to measure the rotations of the sprocket axis. For this, we have to look to the side opposite of the flywheel. Behind a plastic cap (picture below) we find the axis capped with a hexagonal nut. I have read several academic studies where a rotary encoder was connected to the axis. This is an instrument that gives multiple pulses over each rotation. However, at a rotation rate of about 20/sec this would overwhelm my data collecting capabilities. So I looked for simpler means, preferably contactless,.



One option is optical. However, the available photoresistor was rather slow (typically 30 msec rise and 30 ms decay). Another sensor is magnetic. This is used in the flywheel (3 magnets with Hall effect sensor). I had a lot of experience with reed-switch sensors which are commonly used for measuring the speed of a bicycle. I thought to give it a try, although the rotations per second of a bicycle wheel are much slower (30 km/h = 10 m/sec ; typical wheel circumference = 2m, hence 5 rps).

I was long pondering about how to make a small wheel that was driven by the hexagonal nut, before I hit upon a simple solution : a DVD disc. It turned out that I only needed to carve 6 edges in the central hole of a DVD to match the geometry of the hexagonal nut. The next step was to screw 2 spoke magnets on the DVD and find a way of mounting the wheel sensor (reed switch) close to the circular path of the magnets. The use of 2 magnets means that each passage of a magnet corresponds to a handle displacement of 8.89/2 = 4.445 cm. This is the maximum resolution of this setup.







The sampling of the sensor was done with an available DATAQ DI-145 data acquisition kit that has a maximum rate of 240 samples/sec. The DI-145 is connected to an old laptop with a Windows-XP operating systems (this DATAQ has no drivers for later Windows systems). The raw data were then transferred to a desktop that has more advanced data processing software.
The first question was, of course, whether the reed-switch sensor could cope with the high rotation speed of the magnets. Fortunately, this turned out to be the case as shown in the next figure. The information is in the number of peaks and in the time-spacing of the peaks. The height of the peaks is not important. Each successive peak corresponds to a handle displacement of 4.445 cm. The drive has 34 peaks, which implies a drive distance of 151 cm. The recovery has 35 peaks. There are no peaks missing.



The information in the spacing of the peaks can be converted in a speed graph (speed=0.04445/(time gap)).



This graph shows several interesting features. The drive starts with a huge speed ramp, almost linear. This is the initial acceleration of the body (phase-I). The acceleration is much bigger than I expected : from 0.3 m/sec to 1.5 m/sec in about 0.15 sec amounts to an acceleration of 8 m/sec². Note that the acceleration for a vertical jump in the air is 10 m/sec². At an estimated body mass of 50 kg (lower legs don't move) this implies a power of some 350W! Note that there are 4 data points in the initial ramp, so a displacement of about 13 cm. All this happens before the flywheel sensor takes note. At 1.5 m/s I hit the speed of the flywheel and started the power input for the flywheel.





The reason I share these first results, which are provisional, is that I like to challenge other readers to come up with alternatives or better ways to measure handle speed. Maybe there are readers who have experience with Arduino or Raspberry-Pi computers, or have friends playing with such systems, which might do faster and better sampling. I am happy to share more detail. Send me a personal mail if you are interested.

(to be finished soon ; more text will be added)

[Tsnor]

Wonderful project. The shape of the recovery is a neat surprise. I was thinking constant velocity once arms away and back angle was set, you are seeing acceleration into the catch during recovery nearly until the handle stops. Can you capture the direction of spin to see the point the handle goes to zero speed ? Would you use the data you captured to speed up your drive and slow down your recovery to get further from a 1:1 ratio?

You can cross check the handle speed by using video of your stroke, then cross checking the number of frames/time spent / handle distance moved. There are companies that will do this for you from videos of a marked erg. For example: https://analytics.rowsandall.com/2017/0 ... -analysis/ At the 50K foot level, video is an interesting alternative way to instrument an erg vs speed sensors.

[Nomath]

Continuation...... (I didn't know that there is a limited time window to edit posts. My last edit and update was therefore rejected. Unfortunately I also cannot change a few errors, like 30 km/h = 10 m/sec ; should be 36 km/h).

Additional comment to the following graph



At 1.5 m/s I hit the speed of the flywheel and started the power input for the flywheel. The power input accelerates the flywheel from 17 rps to about 22 rps. At 31.1 sec the handle speed drops sharply. There are 1-2 points in the drop, amounting to a distance of some 10 cm before the finish. The drive time seen by the PM (phase-II) would be around 0.75 sec, but the graph shows that the full drive time (phase I+II+III) is about 0.95 sec. Also note that the total distance of the drive was 151 cm ; ErgData only 'sees' about 133 cm.
The recovery follows a more parabolic shape, with a maximum about halfway.

The following graphs are from the start of a regular 5K run. The power input is much lower (about 170W), so speeds are generally lower. The shape of the drive part is the identical. I didn't know that in the recovery there is a typical acceleration in the second half (possibly by leaning the upper body forward).



I have searched the scientific literature on indoor rowing for speed graphs. Although I found more than 10 papers that mention the use of some speed sensor, in most cases a rotary encoder, only one showed a handle speed graph. In this study several motion capture cameras were used and luminescent markers on the handle to measure speed. Camera: 60 frames/sec. It must have been a awkward task to inspect frame-by-frame! The rower was a French gold medalist (paper title : Forces Applied on a Rowing Ergometer, by Nicolas Decoufour and others, 2008). His handle speeds are higher. By comparison my simple system does perform rather well.



One of the first improvements of my system is to use 3 or 4 magnets. This will improve the resolution of the speed graphs and allow a better separation between drive and recovery.
It will also be interesting to compare speed graphs at different drag factors.

What are the most significant take-aways so-far :
1. The drive is more complex than the PM and ErgData show. Drive time and drive distance are significantly longer than shown by ErgData.
2. The start of the drive is a strong acceleration of the body. There is a lot of (leg) power involved that is not shown by the PM. I guess that the initial acceleration is a very good indicator for the strength of a rower.
3. The ratio drive time/total stroke time is larger than commonly assumed.

[Ombrax]

Interesting.

I'm not sure how easily this method would transfer to a Dynamic erg, but it would also be neat to try this for that and, of course, very easy to do it for an erg on slides. The body acceleration (or lack of it) would presumably show up in the data.

[Slothful1]

Very interesting read, thanks for doing the hard work and sharing!

A future enhancement could be to find a way to track the movement of the seat (using a series of equally-spaced magnets on the rail, with a sensor under the seat). You could then get an analysis of how synchronised the seat and handle speeds are during the drive, and also get a measurement of recovery time.

Dave

[Nomath]

Wonderful project. The shape of the recovery is a neat surprise. I was thinking constant velocity once arms away and back angle was set, you are seeing acceleration into the catch during recovery nearly until the handle stops. Can you capture the direction of spin to see the point the handle goes to zero speed ? Would you use the data you captured to speed up your drive and slow down your recovery to get further from a 1:1 ratio?

I am glad you liked it! With the present setup, I cannot capture the direction of the spin, except for the typical spacing of the peaks. I think the reversal point (zero speed) becomes less important as I increase the number of magnets on the disc to 3 or 4, which gives a higher resolution on distance.
I was also surprised by the shape of the recovery curve. I guess I lean forward and stretch arms in the second part of the forward slide. I haven't seen any reference shapes, so this is new territory.

[Nomath]

I'm not sure how easily this method would transfer to a Dynamic erg, but it would also be neat to try this for that and, of course, very easy to do it for an erg on slides. The body acceleration (or lack of it) would presumably show up in the data.

The method will be applicable to an erg on slides if the wire from the reed-switch sensor to the data logger is sufficiently long and flexible. This is a simple thin cable with two strands. Most wired bicycle computers come with about 1 meter cable between computer holder and wheel sensor.
Presumably on slides the body acceleration after the catch will be much lower. Quite interesting to see the differences with the static erg.

[Nomath]

Very interesting read, thanks for doing the hard work and sharing!

A future enhancement could be to find a way to track the movement of the seat (using a series of equally-spaced magnets on the rail, with a sensor under the seat). You could then get an analysis of how synchronised the seat and handle speeds are during the drive, and also get a measurement of recovery time.

Dave

Thanks! My present setup already gives an accurate measurement of the recovery time and also shows where you deliberately slow down or pause. I have no plans to extend the monitoring of the rowing action to seat movements.

I know that several rowing associations, e.g. the Deutscher Ruder Verband, use C2 ergs instrumented with an additional force sensor on the handle and a displacement sensor, to analyse the performance of their elite rowers . Strangely, you can find lots of force curves on the internet but no handle speed curves.

[Nomath]

Besides plotting handle speed as a function of time, it is also interesting to plot handle speed as a function of distance.
It is nice to see that the measurement is able to show a very high consistency in 50 consecutive strokes.
There is a clear inflection point in the drive curves at 17 cm from the catch. This is most likely the point where the handle speed starts to synchronize with the speed of the flywheel and power is delivered to the flywheel.
The single deviating drive is the first drive that starts with a standstill flywheel.



Profile of the recoveries (distance=0 is the point where the drive ends).

[Tsnor]

Strangely, you can find lots of force curves on the internet but no handle speed curves.

There is a handle speed curve here: https://analytics.rowsandall.com/2017/0 ... -analysis/ Search for ths text "Pages 5 to 7 show graphs for the five individual strokes. Here is an example, showing handle and leg speed:" then look at the "riem" chart to see handle speed over time. Text says "Dutch words: Riem = handle"

They are also seeing gradually increasing handle velocity until sharp deceleration right before the catch.

[Nomath]

There is a handle speed curve here: https://analytics.rowsandall.com/2017/0 ... -analysis/
...

Thanks for that link! Strange, I have visited that site several times but never seen these curves.
I did many internet searches but have never seen Row Analysis. Their presentation and method is very interesting. I think about contacting them (I am Dutch too!).

[mict450]

I think about contacting them (I am Dutch too!).

I had absolutely no idea that you were Dutch! But then, I shouldn't be surprised. Of all the people who are not native English speakers, the Dutch, by far, are the most fluent, followed by the Scandanavians. Also a lot better than most of us Yanks!

[Nomath]

I experimented with 3 and 4 magnets mounted on the DVD disc. With 4 magnets, there is a higher resolution in the low speed range, near the catch and the finish. However, at handle speeds above 1.5 m/sec the reed-switch sensor reacts too sluggish, which results in imprecise pulse timing and a high noise in the speed data. I therefore chose to continue with 3 magnets. This implies a spatial resolution of 8.99/3 = 2.96 cm.

Below is a run of 50 strokes at a pace near 2:00 per 500m. I plotted the handle speed relative to the time after the catch. The timing of the catch is done by software and the plot shows that a few drives have been mis-timed. Nevertheless, the broad band of overlapping drive profiles reveals a consistent picture. It's helful to discuss it with reference to the schematic figure in my opening post.





The single 'outlier' profile is the first drive. Because the flywheel is not rotating at the catch, the handle engages directly with the flywheel. The handle speed increases nearly linearly up to 1.5 m/sec at the end of the drive. Acceleration : 1.2 m/s². Handle speed and flywheel speed are directly proportional (1 m/sec = 157 chain links/sec = 11.2 rotations/sec). The first drive takes much longer than the next drives, about 1.5 sec.

The other drives start with a sharp, nearly linear increase in speed until the handle speed matches the speed of the flywheel. This is phase-I in the schematic figure. In these runs it takes 0.17 sec. The Performance Monitor is blind for this phase. There is no power going into the flywheel. All power is going into accelerating the body. The power developed in this phase can be estimated fairly well. At the end of phase-I the speed is 1.5 m/sec. My moving body mass will be about 50 kg (lower limbs don't move), so the kinetic energy at 0.17 sec is ½*m*v² = 56 J. This implies a power of 56/0.17 = 330 W, developed by the legs pushing on the foot stretchers. The acceleration in this phase is quite high : 7.6 m/s². Since there is no connection to the flywheel in phase-I, the acceleration does not depend on the drag factor. Only the duration of this phase depends on the drag factor.

In phase-II the power goes into the flywheel and to a much smaller extend in increasing the speed of the body. In the second part of phase-II the kinetic energy of the body adds to pull the handle and bring the body to a standstill. The handle acceleration in this phase is low, about 0.7 m/s².

Phase-III, characterized by a sharp drop in handle speed takes only about 0.1 sec. The PM is also blind for this phase, because the handle is again disconnected from the flywheel. The body has almost come to a halt, so there is very little power developed in this phase.
In phase-III the handle still moves about 10 cm. Added to the 17 cm movement in phase-I, there is about 27 cm distance in the drive that the PM does not recognize. According to the PM the drive distance was 133 cm ; the graphs show drive distances of 150-155 cm.



The recoveries have the same shape as shown earlier (distance=0 is at the finish of the drive). My earlier explanation for the higher speed in the second part is wrong: forward leaning of upper body and stretching the arms is done in the first part. The kinematics of these movements is intuitive. Possibly the lower speed in the first part is to enjoy a bit of rest.

[jamesg]

The 17cm catch slack on fixed ergs is much as expected, as no-load body mass acceleration is likely to be somewhat less than 1g. Using a dyno or slides it can be reduced; high drag is not the best way to do this since it wrecks the rhythm.
The catch lunge is often seen in races, afloat: sitting on the backstop lets the boat lose her peak speed before coming forward, which would reduce hull speed before necessary. I do that sometimes on erg, and afloat, it feels like a bounce as if kinetic energy could be stored in muscle-tendon and limits time spent waving the blades around in the air; dry the effect doesn't last more than a few strokes and in any case from an inertial point of view it's probably useless.

[JaapvanE]

Thanks for that link! Strange, I have visited that site several times but never seen these curves.

In stroke metrics just came out of beta a couple of days ago, and they are quite interesting.

With Open Rowing Monitor we started filling them in the beta-phase a couple of months ago, including handle speed based on the flywheel speed during the drive. We capture a lot of the drive, but as you convincingly show, not all of it.

The reason I share these first results, which are provisional, is that I like to challenge other readers to come up with alternatives or better ways to measure handle speed. Maybe there are readers who have experience with Arduino or Raspberry-Pi computers, or have friends playing with such systems, which might do faster and better sampling.

With our current set-up, we are capable of capturing data at a microsecond (us, not ms!) accuracy on a Raspberry Pi, as long they are binary (on/off) signals. We do this for the flywheel already, and adding this for another GPIO-pin should be doable, with the benefit that the data is reported in one consistent capture.

One issue is that this employs discrete measurements: for example the catch could be missed because you stop just before the magnet and go back again. These kind of inaccuracies are hard to tackle, especially when the rotary disk changes direction. I'm not certain if a phased grey encoder would help here.

I experimented with 3 and 4 magnets mounted on the DVD disc. With 4 magnets, there is a higher resolution in the low speed range, near the catch and the finish. However, at handle speeds above 1.5 m/sec the reed-switch sensor reacts too sluggish, which results in imprecise pulse timing and a high noise in the speed data.

An alternative might be an optical sensor and drill small holes in the CD. In collaboration with OpenErgo machines (self built Model A's ), we typically see people choose the optical sensors as they are simpler and faster.