I’m not the only one who’s experienced this phenomenon. @Pedrodemio has also experienced this on his ebike build. I’ve used motors sized from 5065 to 63100 with KV ranging from 130 to 270 and experienced this. I wish I had the data to prove this, but I don’t have a dynamo setup. I will try asking someone I know who does have a dynamo if they can try testing this.
Just a by the way, my TRAMPA VESC 6 mk3 says my maytech 6374 motor has a phase resistance of 16mOhm and inductance of 9.98uH.
I know for a fact by using a power supply and RCL meter that the phase resistance is actually 19.25 mOhm and phase inductance (at 30kHz) is 35uH.
I suspect this discrepancy in the measurement is what contributes to the less efficient torque generation at lower speeds, because the different resistance measurement causes a ratio of the phase current to show up in the estimated BEMF.
That is also one of the goals I want a dyno, teste between sensorless, hall sensors, encoder and now HFI
Before I say anything, this is just an edge case, VESC and VESC derivatives is still the best we have, and is really few scenarios that this is an issue
That the VESC has lower torque at really low speeds is a fact that anyone can test. Find the steepest hill that you can climb at speed, it has to be on the edge of what your board is capable, or lower the motor current until it’s just enough for you to climb it at speed
Now try to climb at walking pace or slower, it won’t be capable, while in theory it should since torque is directly proportional to current, and the current your are applying is the same
Or it’s due the hall sensors due to it’s nature not being capable of reporting exact position and so the maximum torque isn’t being produced, if that is the case a encoder will solve, or is something deeper
@Deodand and @Gamer43 am I right in saying that the new HFI offer better angular estimation than hall sensors? And if that’s the case should low speed performance be better with it?
Genuinely asking, I haven’t had time to update anything to 4.0
Amazing work guys, hope to see this trend continue on the future
Also since you are here, @Deodand what do you have to say about the temperature measurement using motor resistance? Is it feasible now the the whole project is more mature? It would be in line with your thinking of less wires = more reliable
Not to go OT, but just updated to firmware version 4 and now my vesc tool 2.01 (android) and metr do not work. I can only connect to the VESC (flipsky 6.6 dual 200a) with a usb connection and the vesc tool from my cellphone.
Any troubleshooting steps that I should be taking right now, or should I attempt to roll back to 3.64 (last working firmware)
I don’t know if I’ve noticed this or not, but I’ll try and focus on it during my next ride. I always kind of thought it was a perception thing as accelerating at higher speed always feels more intense to me than lower speed. I’ve never had a problem accelerating up hills on any of my boards in the absence of some other issue like thermal throttling.
If you had a very even hill and set amps quite low then maxed throttle and logged RPMs you could back out a noisey torque estimate… or better yet I think the vesc 6 logs inclination angle so you could crunch the data of that as well.
Either way on hall sensors with 6 updates per rev you’re looking at an error of ± 30 degrees worst case which would cost you ~ 13% torque from standstill, as you get going a phase locked loop tracks speed and interpolates so you do a bit better even at very low speed.
It definitely offers a higher resolution/update rate as compared to hall sensors (~ 500hz is the rate at which the DFT is computed) but as to the accuracy it is hard to say, testing on the bench it appears very accurate when you overlay plots from encoder but that might get skewed at higher currents.
I’ve been working on this. It’s a simple SISO system and I think a basic Kalman filter with a first order thermal model and scaling measurement update should do the trick. I think the secret will be to scale the measurement update weighting with the motor amps and speed, at low speed and high amps you can get a very accurate and clean estimate of the winding resistance. You can get the high frequency thermal behavior from system dynamics and then the slow dynamics can be taken care of by a heavily averaged resistance estimate.
Hall sensors with even just a first order speed estimator will offer better performance. The whole HFI thing is for when you either don’t have any sensors or your sensor connection is unreliable. I am just highly skeptical of Trampa’s claims. While it is impressive it can spin a motor at low and zero speed, I question whether this is either efficient or worth the additional software overhead with high inertia loads.
Just want to reiterate what I found here. My Haggy (maytech) 6374 has a phase resistance of 19.25mOhm.
The VESC 6 mk3 I purchased from TRAMPA reports a phase resistance of 15.88mOhm.
The Flipsky 6.6 ESC reports a phase resistance of 21.7mOhm.
My own power stage reports something between 18-19.8mOhm depending on which samples I average.
This estimate is definitely within margin from what I feel, will to try to do some more tests to pinpoint where the transition occurs, finding a hill, setting the current to be just enough to climb it and progressively lowering the speed
Hall sensors will give trapezoidal control performance. A first order estimator at low speeds will bring it pretty close to FOC, that’s pretty hard to beat with any sensorless algorithm, regardless of which it is. Hall sensors have hysteresis lag that make them impractical for use at higher speeds, but have little effect at low and very low speeds.
I have shown that the inductance measurement taken by the VESC is not accurate (RCL meter, Fluke brand, at school gives more than double the estimate, my power stage gives more than triple the estimate) so I have a difficult time believing it will beat sensors.
This is of course assuming a perfect hall sensor connection, if the connection is unreliable, then hall sensor performance will be questionable.
I feel like an unwilling beta tester here. The vesc bluetooth tool basically forced me to upgrade. My profiles stopped functioning and it was telling me I had a limited connection, so I ran the firmware upgrade. Kicking myself now. Is there an easy way to roll back to 3.64 firmware?
If you took the time to actually examine the changes you’d note that this new algorithm is also now used for the inductance measurements because it’s more consistent than the previous measurement. However the accuracy isn’t even relevant to how the algorithm works because its based on the phase of the signal not the amplitude.
I’m still not sure why you are talking like you understand exactly how this is working when you haven’t taken the time to get informed.
I’d also say that claiming one inductance measurement is right and another is wrong is naive. The inductance is a linear parameter estimated from a non-linear system. Measured correctly at different amplitudes and frequencies will yield quite different values, all of which might be accurate locally.
I just can’t understand the blind bias. I’m not even claiming what we did is amazing but at least take a look at it? It’s actually a really cool method.
