The Homebuilt Dynamometer (Dyno) Thread!!!
 

Several RCTech users are building, or have built, dynamometers ("dyno" for short), but there is no generic thread where we can all come together and share our experiences and show off our projects. So here it is!

To start the ball rolling, I'm re-posting this from another thread:

Yes, I do know of some expensive ones. They would be the ones made by the guy I know who owns an NC lathe!

Yes, larger diameter does give more MOI for the same mass. But it also gives more leverage on the motor shaft for secondary imbalances, more aerodynamic drag, and reduces the maximum operating speed. Spokes may look nice, but for drag and structural integrity, I would prefer to stick with a solid disk.

Since I'm going to be using a gear drive, I can easily select the MOI that the motor "sees" by varying the ratio, which makes it easier to get good results with a large variation in flywheel sizes. That's important when being forced to use what's available, rather than what is optimum.

Yeah, a disc is better, I just meant changing the shape so that the maximum amount of mass is at the outer diameter, to make the most of what mass you have. Not really concerned with "looking good"! ;)

And it should be possible to compensate for the effects of gearing by looking at the deceleration curve when spinning down, right?

I'm now thinking it might not matter anyway, if we don't care about converting to real units? As in, if one setting goes from 5,000 to 5,075 RPM between two samples (at fixed time intervals), and another setting goes from 5,000 to 5,100 in the same interval, the second is better, and it doesn't matter if it could have gone to 5,150 without the gearing, right?

Maybe one of those "rolling roads" dyno from McPappy would be fine (with the right data capture and analysis), after all? :)

That's a big question at the moment. Determination of things like aerodynamic drag and bearing friction should be quite accurate while coasting down, since they are dependent on speed. Determination of gear losses might not be as accurate, if they are dependent on transmitted torque. The torque from the motor is very high during acceleration, and very low during deceleration, so it's apples and oranges.

I do like the idea to measure data on the track:
http://www.rctech.net/forum/nitro-ro...mes-alive.html

It is possible to compare data to see if there are improvements.

Back in the day (1987-1993) I built my own dyno's. Basically I used a Computer power supply for a 5v constant, slave motor, with 3 resistors in the circuit and controlled by 3 switches. Whole thing was about 1" x 3” x 4". Had meter lead receptacles to attach a multi-meter for voltage or amps. I bought the stuff from Radio Shack or another electronics store. Cost was about $15.
RPM was equal to output voltage from the slave motor. Swap the lead and read the amp draw which equaled torque.

I would run it in free mode then flip the switches one at a time and apply the resistive load to see how the motor responded to loads from a start and from a running motor. Worked very well to determine RPM motors and torque motors.
Plus it was great when you could change brushes, springs, timing, etc. Lot more tuning options than with today’s motors.

Thought of building one again, but with so limited tuning options, the G-Force motor analyzer reads timing and RPM plus amps, though does not check loads is sufficient for now.

The only thing with a slave motor and resistive load is that it can't go very low, and you have only very granular resistance, which will slow it down to a certain level (so you only get the torque at the RPM that it'll go down to).

Or does it?

Would a slave motor with a set resistance could be used in place of a flywheel to do a spin up, to get most of the torque curve? It wouldn't go all the way to zero (it'd stop short of the free running RPM), but would it generally work for the rest of the RPM range?

I'd be a bit worried that the torque reading wouldn't be linear through the RPM range, but if you optimize for each given RPM (use the timing that gives you the best torque at each RPM), it could make sense, right?

For blinky, there's not much to tune, but for boosted, there's a lot of information that you need to optimize the motor, and that's information a dyno can give you (more or less) easily. :)

The G-Force motor analyzer sounds good enough for blinky, IMHO. Getting the true timing measurement from each sensor is good for getting (more) consistent results from one motor to another, and the amp reading can let you see how much timing is too much. ;)

It was only a $15 (poor man's) dyno :lol: It was a good indicator of motor performance (top RPM and AMP draw). :tire: To design what you are asking will cost $$$$$.

Ha! Yeah, that's not wrong! :lol:

For brushed motors, you could get a lot of mileage out of that kind of dyno setup, when tweaking brushes and such...

The RC Benchmark dyno that recently came out is just shy of US$450! It looks like the current version is aimed at other applications than RC cars (like quadcopters, robots, etc), but they seem interested in developing something that would be good for car racing... :)

That's awesome work-- definitely what we have come to expect from you!

It's not difficult nor terribly expensive to use an electronic current sink (controlled by a microprocessor, of course!) on the slave motor to vary the load on the motor under test. That's how the fancy dynos for brushed motors worked. But torque output needs to be measured with a separate sensor like a strain gauge to get good accuracy. Just measuring the current loading the slave motor isn't good enough.

The slave motor could simply replace the propeller as the load on the RCBenchmark dyno, since it already has a strain gauge to measure torque.

The flywheel dyno still gives more complete data in a shorter time, but if one needs to run a motor under a continuous load for an extended period of time (perhaps for stress testing), then the absorption dyno the proper device.

What I meant was to try to have the slave motor give a (high enough) constant load, and rather than run the tested motor continuously, as you normally would with a slave motor setup, instead do a full-throttle acceleration, like you would with a flywheel? You'd then analyze the data the same way as you would with a flywheel.

Like I said, calculating "real" torque won't be easy (or maybe even possible), and it won't spin up all the way to the free-spinning RPM, but for the range covered, you should be able to get data to compare different timing settings at the same RPM, and come up with some sort of reasonable timing advance curve.

Maybe the data at the top of the RPM range could be gathered with a lower load setting on the slave motor, and yet more runs?

Back in the day I had a Fantom Dyno. Thing was great! Could do alot with it based upon the data you were getting. There was a slight learning curve when you made changes. The total opposite happened that what you would think.. I figured alot of it out over time.

I see HobbyKing has a meter for props now.

http://www.hobbyking.com/hobbyking/s...arehouse_.html

I was thinking about buying a Novak Sentry Data Logger, but they're a bit old, maybe newer hardware would be better (high sampling rates, for example)? I wonder if some Arduino would able to do that? I'm not that good with hardware, I'm more of a software guy! :)

I see you're working with nitro cars, but could this be modified to gather some other information, putting probes on the motor power and sensor leads, and measure how much timing advance the ESC is applying?

A bit off-topic, but that whole telemetry/data recording business also seems very interesting, albeit in a palliative way. I could do drag runs, with various timing settings, and at least get my boost settings in the ballpark... :)

Eagletree used to have components. I havent looked to see if they are even in business anymore.

This is an old data aquisition box with 8 analog, 8 digital and 1 timing input. You can connect all kind of sensors as long you make the range of 0-2.5v or digital max 5v on/of switching signal. The software is modulair so you can set it up with even re-calculations on the range.

Off-track testing and tuning can be usefull but still you have to translate it to the setup of a car. Looking at a lot of toplevel drivers I do see they are not working with test benches but getting as much tracktime as possible to setup the whole car and also important: to get the track in the fingers.

And then a real time (recorded) telemetry will show directly if there are improvements or not. But finally the laptime is the most important and there are enough situations where the maximum performance of the motor does not count.

These guys?

http://www.eagletreesystems.com/

What kind of sampling rate can it do? Quick enough to directly sample the Hall effect sensors (would need at least 1 kHz)?

I generally agree, other than with the one exception of boosted racing. I've never felt the need for a dyno in brushless blinky racing, very little to test on the bench, my testing is done with lap timing equipment, at the track! :)

But in boosted (dynamic timing) racing, getting the most power out of a motor is a completely straightforward process, if you have the torque/timing curve data (which should be fairly simple with a flywheel dyno). It's not even a matter of taste or set-up tuning, once it's set up right, you'll never need to tweak it (unlike timing in blinky racing), but without the data, it's some very difficult guesswork (as you're also trying to work out the right gearing, so a lack of punch or top speed could be either, with almost nothing to tell you).

For blinky racing, it's a lot simpler, and I'd say that other than some simple "how much timing is too much" (that you could do on the track without too much difficulty), there isn't much that a dyno can tell you that you couldn't find out at the track, for a given motor (a big budget racer could use a dyno to pick out the best motor out of a bunch, but I'm not interested by that).

This system does 2 whole string (17 inputs) of samples per second. One of the reasons that is is that slow is a reliable transmitting data on larger distances with low power.

Data aquisition wit a 1khz sampling rate is possible, but does require more advanced systems.

A 2000kv motor will spin at about 230Hz with no load. 1k might be a bit low, but it might just do it. The wire that goes from the ESC to the motor usually has 6 wires. What kind of encoding is used?

I guess that timing is implemented using interrupts in the ESC microcontroller. Instead of fast analog sampling, the microcontroller monitors a digital pin. There is a comparator in the microcontroller, and as soon as the comparator is activated (either by a up, down, or up or down signal), a piece of code is executed.

I built this thing with my collegue about 10 months ago: i.imgur.com/3vYNipR.png

It is too bad we did not take a picture...We were attempting to build a direct drive micro helicopter. The whole helicopter was mounted on bearings. There is a load cell at the back to measure torque. The motor was loaded with the prop. We could change dynamically the motor loading by moving the swash plate, and we were measuring the thrust.

The whole thing could run completely automated tests, and it had a nice matlab gui. It was using arduinos for data acquisition, op-amps for load cells, load cells savaged from kitchen scales, and a current sensor.

Here is a picture of the micro helicopter prototype:

i.imgur.com/acuyEaf.jpg

1 kHz sampling would capture all of the transitions on a sensor output from a 17.5T car motor (which runs at about 300 Hz), but the resolution would be insufficient; 30 kHz would give 1% resolution in the measured period.

Here's a diagram of the sensor outputs and motor drive:

http://i1191.photobucket.com/albums/...pshxxbayhs.jpg

A comparator isn't needed, since the Hall sensor outputs have digital levels.

I have tapped into one of the sensor signals with a 100K-ish series resistor to reduce loading and attenuate the signal going into the microphone input of my laptop, which is what the SimpleDyno freeware program uses for speed sensing. The mic input can be sampled at 44.1 kHz, so the resolution is pretty good.

For the pin out, you saw the section 8.5.1.3 in the ROAR rule book?

I was thinking "at least double it, for Nyquist". I know just enough to be dangerous... ;-)

Actually, you are correct if the RPM is determined by counting the frequency over a gate time. One second gating with a 300 Hz input gives 300 cycles, which is 1/3% resolution. But that requires that the motor speed remain stable over the gate time. That isn't the case for a flywheel dynamometer, but could be for an absorption (brake) dynamometer. So my answer was biased by my preference for a flywheel dyno.

So it would work, but be very noisy during acceleration, gotcha.

This standardization is really great for consumers! It prevents a manufacturer from developing a proprietary system. It is interesting that they label phase A, B, and C, as I understand it, it is sensor A, B and C.

The comparator I was talking about is just an implementation detail. It is how the ESCs are doing it, and it is very accurate. Most microcontroller will have pins that can execute code when triggered. The comparator is actually inside the microcontroller. Those pins are monitored internally in the MHz range. The trick is that you tell the microcontroller, each time this pin is triggered, save the time in a variable. The time is very accurate, so you can subtract the old time to the new time, and obtain the phase length in microseconds. The phase length is inversely proportional to the RPM.

Actually, it was a proprietary system created by Novak. When it came time for ROAR to create the rules, they just used Novak's arrangement (with the full blessing of Novak, as I understand it).

By the way, the thermistor is unfortunately not required for a motor to be approved by ROAR.

Not all ESCs monitor and process the sensor inputs with an interrupt structure. I believe some are hardware based, since the dither in the delay from sensor change to motor drive change is virtually non-existent. (ROAR rules specify a maximum delay of 10us, but the delay may vary within this range during operation.) But I don't know the details of the hardware used.

I just took some rough data from the flywheel shown in one of my posts in the RC Benchmark thread. It weighs 458g, and has an OD of about 2.25" (the weight is not evenly distributed across the disk). I ran it using a 21.5 motor on 1s LiPo with 23T/44T 32 pitch gears (a 1.91:1 reduction of speed from motor to flywheel).

My initial target was to have sufficient effective MOI to get about a 1% change in velocity between readings (one reading per revolution) at the motor's power peak (which occurs at about 50% of the free-running speed). Modeling showed that requires about 40 total motor revolutions with the required MOI.

The data showed that the motor reached 50% speed after about 4 revolutions, or a factor of 10 less than desired. If I change the gear ratio to 1:1, that gains me a factor of about 3.66 (1.91^2). That means the flywheel MOI must increase by a factor of 2.73 to get me where I want to be. Time to try a larger flywheel!

http://i1191.photobucket.com/albums/...psafbmw9ih.jpg

Interesting. Did you have any vibration problem? Using your measurements, and assuming a cylinder, your wheel has an inertia of about 7.5*10^-4 kg.m^2.


I was wondering what is the equivalent inertia of a car. I started with a B44. I obtained 3.4*10^-2 kg m^2, or about 45 times larger than you wheel. You can see my calculation here. It is freely editable, so people can add their own cars.

I know that in theory, the result of the characterization should be the same with a small or a large wheel, but there might be small differences due to delays in the ESC. I mean, you could even use the inertia of the rotor for the characterization, but if the acceleration is too fast, the ESC might not perfectly keep up. The question is, at what point is the inertia wheel large enough so that the results of the characterization don't change anymore?

howardcano, if I understand well, you reached 50% of the speed in 4 revolution of the inertia wheel? What speed is that? If I assume 4000RPM, I obtain a torque of about 0.2Nm assuming a constant acceleration (it is not), which sound pretty reasonable.

What is the maximum axial load tolerable by the motors bearings? Lets assume that during maximum acceleration, the torque goes up to 0.5 Nm. A pinion might have a diameter of 9mm (I measured that, anybody knows the gear size used? I could get this number exactly, as I know the number of teeth). With those assumptions, the force on the shaft is 10kg during acceleration, and a bit more if we take into account the radial force due to the pressure angle. Consequently, I think it is safe to use heavier flywheel, but I would mount the flywheel so its axis of rotation is parallel to the ground.

I corrected the flywheel for the imbalance caused by its set screw, but not for any other direction. There was very little vibration. The motor's free-running speed was very close to 8000 RPM, so about 4180 RPM on the flywheel.

I monitored the sensor signal on the motor. It took about 4 motor revolutions to reach 50% speed. It could easily have been 3.5 or 4.5 revolutions, since there is some ambiguity over how fast the motor is already going by the time the monitored signal changes to trigger the scope.

Many car ESCs have an adjustable ramp-up time, called "punch control" or something similar. I set this to minimum. However, the acceleration time was so short with my test rig that I'd guess that it is affected by the ESC's punch control. Actually, it wouldn't surprise me to find that most car ESCs have a limited ramp-up time to avoid very high current peaks during acceleration. I remember using (and designing) some old airplane ESCs that exhibited the same response, though current was not necessarily the reason behind it.

I suspect that once I get the flywheel as big as needed to reach my goal of 1% change in time period between two successive revolutions at the power peak, the ESC ramp-up limitation will no longer contribute significant error to the test.

The front motor bearing is 3/16"IDx1/2"ODx0.196", but I don't know its load rating. (In any case, it is not intended for significant axial loads.) I'm less concerned about the bearings than the 1/8" motor shaft with a big flywheel. A small imbalance could cause the shaft to flex enough that it leads to dynamic divergence, resulting in the flywheel beginning a new journey sans motor!

The smallest pinion I have used on my cars has a pitch diameter of 1/4".

Here's a somewhat different idea I just had, to try to approximate the information I'm looking for, possibly on the cheap...:)

We know the torque curve for a brushless motor with static timing is approximately flat, starting high at the stall torque, and going down to zero at the free-running RPM, right?

Well, figuring out the free-running RPM isn't difficult, with the max RPM feature in most ESC: just remove the pinion, rev it up briefly, and you can find out.

If we assume the torque curve is flat, all we need is another point, preferably at the low end of the RPM range, and we can get the (very) approximate torque curve!

Now, the question becomes this: what's the simplest way to find out the stall torque, or the torque at some low RPM?

Since our other torque data point is zero, we don't even need a real unit, only something that we can compare other motors with...

Any ideas?

If we accept that these would be only approximate results, then you are correct. You can find the stall torque by just connecting a lever arm to the motor shaft, rest the other end of the lever arm on a scale so that the force from the arm is normal to the scale, feed a known current into the motor, read the force on the scale, and calculate the torque. Repeat this for several angles of rotation of the motor within each commutation position (since the torque varies with rotor position), and again for the other commutation postitions (6 total), then average all of the results. This of course assumes that the torque vs current transfer function is linear.

Wow, that still sounds like a lot of work just to get approximate results!

Yeah, I thought about something like a lever arm, but that didn't seem so good. :)

I also thought about using a massive propeller, but that's also inconvenient, not very accurate, and requiring something like the RC Benchmark setup. ;)

A spool with a heavy weight, and winch it up? :lol:

I tried a larger flywheel. It took about 9 motor revolutions to reach 50% speed, so once I change the gear ratio to 1:1 I'll be getting within spittin' distance of my target. To that end, I ordered a Custom Works Spur to Pinion Adapter (#CW8225) so I can use the same size spur for drive and driven gears.

I also ordered a 1/2" diameter shaft and bearings to support the larger flywheel, and some pillow blocks to make assembly of the test rig easier, plus another identical flywheel to double the inertia.

The flywheel I used is a weldable steel hub for V-pulleys. It has two set screw holes at 90 degrees from each other, which makes it easy to balance by simply attaching weight as necessary to the threaded holes. Of course, that also means it is necessary to balance it! The best part is it only cost $12 on the evil auction site. Here's a picture:

http://i1191.photobucket.com/albums/...pstzjrpnxf.jpg

In theory, only Kv is sufficient...I spent a lot of time on that last week.

In SI units, KV = 1/KT (or KV = KT, depending on how you define your constants).

On the following graph, the bottom right corner of the triangle is at point (KV*Vbat, 0). The top left corner of the triangle would be at (0, KT*Current_max). Current max is calculated with the internal resistance of the motor and the battery voltage. The triangle cut at the top, as the motor is limited thermally to 20A. In an ideal motor, the torque on the vertical axis is proportional to the current.


http://i.imgur.com/tLzRMa5.png

Unfortunately, due to the ESC and other effects, brushless motors do not follow theory that well. To obtain the motor constants, including the equivalent resistance of the motor, the ESC and the wires, I had to do a fit of a lot of data points, and even then, the model predicted power and efficiency within about 10% precision.

Consequently, I think that characterizing a motor with only 1 or 2 points will give a quite inaccurate model. Characterizing it with a lot of points will give a better model that could allow narrowing down the selection of motor to 2-3 possibilities. If you want to know which motor is best for your application with more than 10% accuracy, or if you are using your motor/ESC for anything unusual, I think experimental data will be required.

No, not for determining torque or power. If I take a motor wound bifilar with two wires of the same gauge (like a ROAR-approved 17.5T with 20 gauge wire), then cut one of the wires, the Kv remains essentially constant, but the torque and power both drop by a factor of two.

If you mean "Kv and resistance are sufficient" then I think you are correct.