There has been a spurt of papers recently pointing out that cheap BLDC motors rarely have their hall sensors precisely aligned and that this has deleterious affects on commutation efficiency. Cf:
https://www.iosrjen.org/Papers/vol3_issue1%20(part-1)/I03114955.pdf
http://yadda.icm.edu.pl/yadda/element/bwmeta1.element.baztech-3206a2c6-58ec-4be3-a136-4de722eea734/c/Kolano_Calculation_3_2015.pdf
This made me wonder if I could use a micro-benchmarking type technique to probe what's going on in my motor. Software, in other words, rather than hooking up oscopes and "ground speed" meters.
The motor controller I use -- Roboteq FBL2350 -- is programmable and lets you easily query speed and hall sensors state (0-5). The speed is of course calculated/inferred from a timer that that records how long it took to get from one transition to another. So I wrote code to record the speed at each transition to see what kind of cylicality I could find.
Speed did indeed vary in a periodic way: at constant voltage, speed would cycle up and down by about 2% (using trapezoidal communication), and by almost 5% using sinusoidal/FOC.
Though I ran it through a periodogram and PACF to confirm, a quick glance at the plot showed a fundamental period of 96. Since I was doing this on a MAC motor, that made perfect sense to me: 16 pole pairs * 6 transitions. (I need to do some noise reduction but it does look like there may be a tiny autocorrelation peak at 480, as well. The MAC is of course geared 5:1...)
Ok, not that thrilling (although -- personally -- I still call it victory when empirical results somewhat match my world view. )
What I'd like to understand now, is if any information about hall sensor misaligmnet -- or other motor imperfections -- might be inferred from futher filtering the data (ie subtracting the sinusoid). Any thoughts?
Thanks!
s.
https://www.iosrjen.org/Papers/vol3_issue1%20(part-1)/I03114955.pdf
http://yadda.icm.edu.pl/yadda/element/bwmeta1.element.baztech-3206a2c6-58ec-4be3-a136-4de722eea734/c/Kolano_Calculation_3_2015.pdf
This made me wonder if I could use a micro-benchmarking type technique to probe what's going on in my motor. Software, in other words, rather than hooking up oscopes and "ground speed" meters.
The motor controller I use -- Roboteq FBL2350 -- is programmable and lets you easily query speed and hall sensors state (0-5). The speed is of course calculated/inferred from a timer that that records how long it took to get from one transition to another. So I wrote code to record the speed at each transition to see what kind of cylicality I could find.
Speed did indeed vary in a periodic way: at constant voltage, speed would cycle up and down by about 2% (using trapezoidal communication), and by almost 5% using sinusoidal/FOC.
Though I ran it through a periodogram and PACF to confirm, a quick glance at the plot showed a fundamental period of 96. Since I was doing this on a MAC motor, that made perfect sense to me: 16 pole pairs * 6 transitions. (I need to do some noise reduction but it does look like there may be a tiny autocorrelation peak at 480, as well. The MAC is of course geared 5:1...)
Ok, not that thrilling (although -- personally -- I still call it victory when empirical results somewhat match my world view. )
What I'd like to understand now, is if any information about hall sensor misaligmnet -- or other motor imperfections -- might be inferred from futher filtering the data (ie subtracting the sinusoid). Any thoughts?
Thanks!
s.
