Driver for very small brushless DC motors?
in [Design]
|
From: John Larkin on 23 Nov 2009 10:04 On Mon, 23 Nov 2009 01:16:35 -0500, Phil Hobbs <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >John Larkin wrote: >> On Sun, 22 Nov 2009 22:33:18 -0500, Phil Hobbs >> <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >> >>> John Larkin wrote: >>>> On Sun, 22 Nov 2009 17:23:32 -0800 (PST), dagmargoodboat(a)yahoo.com >>>> wrote: >>>> >>>>> On Nov 22, 5:43 pm, Phil Hobbs >>>>> <pcdhSpamMeSensel...(a)electrooptical.net> wrote: >>>>>> John Larkin wrote: >>>>>> >>>>>> > What a coincidence... I've been thinking about the same problem. >>>>>> > >>>>>> > How about a small, cheap stepper. One could run it in microstep mode >>>>>> > and tweak its drive waveform to get very smooth rotation; I know that >>>>>> > works. Then couple it to the load platform through something >>>>>> > torsionally compliant, like a spring or a rubber tube or a piece of >>>>>> > piano wire or something. Maximize the mass of the load platform to >>>>>> > make a mechanical lowpass filter. >>>>>> > >>>>>> > Over the top, but I suppose one could make a multipole rotational >>>>>> > lowpass filter by adding mass to the motor and/or insert an >>>>>> > intermediate mass and use two compliant couplings. I've seen >>>>>> > Collins-type mechanical filters like this, and it resembles a >>>>>> > microstrip lowpass filter in concept. >>>>>> > >>>>>> > The stepper gives exact, controllable rotational speed open-loop, >>>>>> > which is nice. And small steppers are cheap and easy to drive. >>>>>> > >>>>>> > We could program one of our multichannel arbs to test some motors and >>>>>> > find a nice pre-distorted waveform that gives smooth rotation. I think >>>>>> > adding some third harmonic is classic here, but whatever works. How >>>>>> > would one instrument the resulting angular rotation? Optically, I >>>>>> > guess, or maybe drive a variable capacitor? >>>>>> >>>>>> I'm mostly interested in very smooth motion at small scales, which is >>>>>> why I want an ironless BLDC. The gizmo's operation will require a lot >>>>>> of curve fitting to pull out the amplitude and phase of a >>>>>> small-amplitude tone burst of about 10k cycles over about 5 degrees of >>>>>> shaft rotation, once per rev. Any cogging or other bad behaviour of the >>>>>> motor will cause nasty spurious peaks in the spectrum, among other problems. >>>>>> >>>>>> Steppers are never sufficiently well made to avoid periodic errors--I'm >>>>>> at the level where I have to worry about whether the ball bearings are >>>>>> smooth enough, or whether I need to use jewels, which would be fragile >>>>>> and expensive enough to dim my enthusiasm quite a bit. (A galvo is >>>>>> another possibility, but those cost the Earth.) My hope is that because >>>>>> the balls' motion doesn't have the same period as the shaft rotation, I >>>>>> can sort out the bearing junk from the desired signal. >>>>>> >>>>>> In the real system, I'm expecting to have optical clues as to what the >>>>>> actual motor phase is, but I'm not too worried about that at this point. >>>>>> >>>>>> I'm currently gearing up to do a sanity test with a nice Maxon brush >>>>>> motor from my junk box, a He-Ne, and an HP 35665A dynamic signal >>>>>> analyzer to do the data acq and so on. (I just got a Prologix >>>>>> GPIB-Ethernet gizmo, so I don't have to use the floppy drive to get data >>>>>> in and out.) >>>>>> >>>>>> Cheers >>>>>> >>>>>> Phil Hobbs >>>>> >>>>> Even microstepped, steppers shake, rattle,& roll. And they sing >>>>> (resonate). I never imagined how much until I tried a few. >>>> >>>> But they can be silky-smooth if you drive them right, in the speed >>>> range they like. >>>> >>>>> >>>>> As far as COTS, CD, DVD& hard disk spindle motor drivers? They use 3- >>>>> phase BLDC motors& integrated controllers. >>>>> >>>>> Here's an old BLDC datasheet off ye old hard drive: >>>>> http://www.freescale.com/webapp/sps/site/prod_summary.jsp?code=MC34929 >>>>> >>>>> But won't you be wanting ultra-fine control over commutation, PWM, >>>>> position-interpolation and such? You'll probably have to do that >>>>> yourself. >>>>> >>>>> Atmel, Microchip, and Freescale all have good application notes on >>>>> BLDC-driving with uCs. >>>>> >>>>> e.g. Atmel AVR444: Sensorless control of 3-phase brushless DC motors. >>>> >>>> I think of a BLDC as a 3-pole stepper that hard commutates based on >>>> crappy Hall sensors. And I think of a stepper as a 100-pole BLDC that >>>> soft commutates using precisely the waveform that produces the >>>> smoothest rotation. >>>> >>>> So there. >>>> >>>> John >>>> >>> >>> But iron-rotor steppers and BLDCs both cog like absolute mad on the >>> scale I care about--the signal I'm looking for is the equivalent of >>> ~10*6 cycles per rev, and I need to resolve 1/8 cycle or better. I can >>> average out random stuff, or things like out-of-round ball bearings, but >>> cogging is the same on every single revolution, so it survives averaging >>> and looks just like signal. >>> >>> Cheers >>> >>> Phil Hobbs >> >> Then mimimize the amount of cogging and maximize its frequency, and >> mechanically lowpass filter the rotation. The imperfect microstep >> wiggles are essentially harmonics of the step rate, so are relatively >> easy to lowpass filter. A clever drive waveform will minimize the >> lower harmonics and make the filtering more effective. >> >> If you really want to run in spindown mode (sounds mathematically >> messy to me!) use any junky motor and disconnect it during spindown. >> Some Bendix drive variant maybe, or a centrifugal clutch, or an air >> gap viscous coupling. >> >> Air motors are very cool, except that they need compressed air. We're >> working with some guys who use a gas motor to spin the prism in a drum >> camera, at 20 KHz; 1.2 MRPM. >> >> John >> >> >Disconnecting it is probably the ticket, just as you say, which is why >I'm coming around to the eddy current drive. Linkages don't get much >floppier than that, it's dead simple, and there's nothing to wear out or >get broken. Got any links to such motors? Induction motors should have zero free-spin torque too, except for windage. > >The pattern I'm looking for is symmetrical with rotation, so it isn't >too hard to get the deceleration rate to high accuracy. It's all about >curve fitting, and I have 50k data points and only need 7 parameters. >So if I can minimize the systematic errors (especially those harmonics >of the rotation rate you mention), I should be in pretty good shape. > >At 1.2 MRPM, the glass is probably becoming birefringent due to the >stress! Or it breaks up into little pieces, bad for the optics too. > Must be bomb photographers. > Yup. John
From: John Larkin on 23 Nov 2009 10:07 On Mon, 23 Nov 2009 01:24:46 -0500, Phil Hobbs <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >John Larkin wrote: >> On Mon, 23 Nov 2009 00:42:36 -0500, Phil Hobbs >> <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >> >>> dagmargoodboat(a)yahoo.com wrote: >>>> On Nov 22, 9:23 pm, John Larkin >>>> <jjlar...(a)highNOTlandTHIStechnologyPART.com> wrote: >>>>> On Sun, 22 Nov 2009 17:23:32 -0800 (PST), dagmargoodb...(a)yahoo.com >>>>> wrote: >>>>> >>>>> >>>>> >>>>>> On Nov 22, 5:43 pm, Phil Hobbs >>>>>> <pcdhSpamMeSensel...(a)electrooptical.net> wrote: >>>>>>> John Larkin wrote: >>>>> >>>>>>> > What a coincidence... I've been thinking about the same problem. >>>>>>> > >>>>>>> > How about a small, cheap stepper. One could run it in microstep mode >>>>>>> > and tweak its drive waveform to get very smooth rotation; I know that >>>>>>> > works. Then couple it to the load platform through something >>>>>>> > torsionally compliant, like a spring or a rubber tube or a piece of >>>>>>> > piano wire or something. Maximize the mass of the load platform to >>>>>>> > make a mechanical lowpass filter. >>>>>>> > >>>>>>> > Over the top, but I suppose one could make a multipole rotational >>>>>>> > lowpass filter by adding mass to the motor and/or insert an >>>>>>> > intermediate mass and use two compliant couplings. I've seen >>>>>>> > Collins-type mechanical filters like this, and it resembles a >>>>>>> > microstrip lowpass filter in concept. >>>>>>> > >>>>>>> > The stepper gives exact, controllable rotational speed open-loop, >>>>>>> > which is nice. And small steppers are cheap and easy to drive. >>>>>>> > >>>>>>> > We could program one of our multichannel arbs to test some motors and >>>>>>> > find a nice pre-distorted waveform that gives smooth rotation. I think >>>>>>> > adding some third harmonic is classic here, but whatever works. How >>>>>>> > would one instrument the resulting angular rotation? Optically, I >>>>>>> > guess, or maybe drive a variable capacitor? >>>>> >>>>>>> I'm mostly interested in very smooth motion at small scales, which is >>>>>>> why I want an ironless BLDC. The gizmo's operation will require a lot >>>>>>> of curve fitting to pull out the amplitude and phase of a >>>>>>> small-amplitude tone burst of about 10k cycles over about 5 degrees of >>>>>>> shaft rotation, once per rev. Any cogging or other bad behaviour of the >>>>>>> motor will cause nasty spurious peaks in the spectrum, among other problems. >>>>> >>>>>>> Steppers are never sufficiently well made to avoid periodic errors--I'm >>>>>>> at the level where I have to worry about whether the ball bearings are >>>>>>> smooth enough, or whether I need to use jewels, which would be fragile >>>>>>> and expensive enough to dim my enthusiasm quite a bit. (A galvo is >>>>>>> another possibility, but those cost the Earth.) My hope is that because >>>>>>> the balls' motion doesn't have the same period as the shaft rotation, I >>>>>>> can sort out the bearing junk from the desired signal. >>>>> >>>>>>> In the real system, I'm expecting to have optical clues as to what the >>>>>>> actual motor phase is, but I'm not too worried about that at this point. >>>>> >>>>>>> I'm currently gearing up to do a sanity test with a nice Maxon brush >>>>>>> motor from my junk box, a He-Ne, and an HP 35665A dynamic signal >>>>>>> analyzer to do the data acq and so on. (I just got a Prologix >>>>>>> GPIB-Ethernet gizmo, so I don't have to use the floppy drive to get data >>>>>>> in and out.) >>>>> >>>>>>> Cheers >>>>> >>>>>>> Phil Hobbs >>>>> >>>>>> Even microstepped, steppers shake, rattle,& roll. And they sing >>>>>> (resonate). I never imagined how much until I tried a few. >>>>> >>>>> But they can be silky-smooth if you drive them right, in the speed >>>>> range they like. >>>>> >>>>> >>>>> >>>>>> As far as COTS, CD, DVD& hard disk spindle motor drivers? They use 3- >>>>>> phase BLDC motors& integrated controllers. >>>>> >>>>>> Here's an old BLDC datasheet off ye old hard drive: >>>>>> http://www.freescale.com/webapp/sps/site/prod_summary.jsp?code=MC34929 >>>>> >>>>>> But won't you be wanting ultra-fine control over commutation, PWM, >>>>>> position-interpolation and such? You'll probably have to do that >>>>>> yourself. >>>>> >>>>>> Atmel, Microchip, and Freescale all have good application notes on >>>>>> BLDC-driving with uCs. >>>>> >>>>>> e.g. Atmel AVR444: Sensorless control of 3-phase brushless DC motors. >>>>> >>>>> I think of a BLDC as a 3-pole stepper that hard commutates based on >>>>> crappy Hall sensors. And I think of a stepper as a 100-pole BLDC that >>>>> soft commutates using precisely the waveform that produces the >>>>> smoothest rotation. >>>>> >>>>> So there. >>>>> >>>>> John >>>> >>>> Depends on how you drive 'em, of course, and how fast. >>>> >>>> I think of BLDCs and kin as linear motors--almost like a voice-coil >>>> motor--wrapped around a spindle: drive them with sinusoids at low >>>> speeds, and interpolate smoothly between positions. >>>> >>>> Or you can drive them all--steppers too--at high speeds with >>>> rectangular or crapezoidal waveforms for higher torque,& the >>>> mechanical low-pass of the rotor's inertia still yields smooth >>>> rotation. >>>> >>>> Stepper resonances aren't a problem at all if you crawl, or if you >>>> fly, but they sure are a pain at mid-band. >>>> >>>> But for super-fine angular resolution stepper poles just aren't >>>> mechanically or magnetically accurate enough. >>>> >>>> I'd think ironless rotors would still have several once-per-rev >>>> periodic errors, but at least they don't have a magnetized cog with 50 >>>> hungry poles, lusting for iron fingertips across a small gap. >>>> >>>> So, that's my boneheaded appreciation of it. >>>> >>>> Phil's app sounds like it needs a 1,000,000 line optical encoder (or a >>>> 100,000 line analog encoder and a 14-bit a/d)! >>>> >>>> -- >>>> Cheers, >>>> James Arthur >>> >>> Nah, just Newton's laws and good timing accuracy, hopefully. I used to >>> pal around with a guy named Ed Yarmchuk, who invented self-servowriting >>> for hard disks--he replaced insane laser interferometer spin-stands for >>> writing the servo tracks, with a bit of drive firmware, good timing, and >>> Mr. Newton. You couldn't make terabyte hard disks without it. He >>> retired a year or so ago (very young). Smart guy. >>> >>> Cheers >>> >>> Phil Hobbs >> >> Right. If you had some sort of reference signal, equivalent to a >> million-step encoder, rotational noise wouldn't matter... you could >> timebase correct it out. And chance of mixing a reference signal with >> the real thing, or having one on the side? >> >> That might get to be compute-intensive, but so is doing Fouriers >> linked to a spindown system. >> >> Some truly constant-speed spinner sure would be nice. >> >> John >> >> >> > My processing algorithm will probably be something like this: >(1) Count the fringes from the time they become visible till the time >they go away; >(2) Estimate the deceleration rate by comparing the transition rates in >patches at the two ends of the pattern, and maybe a few pairs of patches >in between, which gives the deceleration curve; >(3) Knowing the wavelength and path length vs time (because we do by >now), estimate the other things we care about. > >No big huge FFTs, I don't think. That could be grim on a PIC! (On the >other hand, it doesn't matter if it takes 5 seconds or so to do the >measurement.) > >Cheers > >Phil Hobbs Pass the raw data to a PC and do it there? We're starting to use an NXP uP, ARM architecture, 280 MHz or some such, that has a floating-point vector co-processor. About $7. John
From: Phil Hobbs on 23 Nov 2009 12:16 John Larkin wrote: > On Mon, 23 Nov 2009 01:16:35 -0500, Phil Hobbs > <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >>>> But iron-rotor steppers and BLDCs both cog like absolute mad on the >>>> scale I care about--the signal I'm looking for is the equivalent of >>>> ~10*6 cycles per rev, and I need to resolve 1/8 cycle or better. I can >>>> average out random stuff, or things like out-of-round ball bearings, but >>>> cogging is the same on every single revolution, so it survives averaging >>>> and looks just like signal. >>>> >>>> Cheers >>>> >>>> Phil Hobbs >>> >>> Then mimimize the amount of cogging and maximize its frequency, and >>> mechanically lowpass filter the rotation. The imperfect microstep >>> wiggles are essentially harmonics of the step rate, so are relatively >>> easy to lowpass filter. A clever drive waveform will minimize the >>> lower harmonics and make the filtering more effective. >>> >>> If you really want to run in spindown mode (sounds mathematically >>> messy to me!) use any junky motor and disconnect it during spindown. >>> Some Bendix drive variant maybe, or a centrifugal clutch, or an air >>> gap viscous coupling. >>> >>> Air motors are very cool, except that they need compressed air. We're >>> working with some guys who use a gas motor to spin the prism in a drum >>> camera, at 20 KHz; 1.2 MRPM. >>> >>> John >>> >>> >> Disconnecting it is probably the ticket, just as you say, which is why >> I'm coming around to the eddy current drive. Linkages don't get much >> floppier than that, it's dead simple, and there's nothing to wear out or >> get broken. > > Got any links to such motors? Sure--made by Maxon and MicroMo (Faulhaber), e.g. http://www.maxonmotor.com/product_overview_details_ec_motor.html > > Induction motors should have zero free-spin torque too, except for > windage. > >> >> The pattern I'm looking for is symmetrical with rotation, so it isn't >> too hard to get the deceleration rate to high accuracy. It's all about >> curve fitting, and I have 50k data points and only need 7 parameters. >> So if I can minimize the systematic errors (especially those harmonics >> of the rotation rate you mention), I should be in pretty good shape. >> >> At 1.2 MRPM, the glass is probably becoming birefringent due to the >> stress! > > Or it breaks up into little pieces, bad for the optics too. > >> Must be bomb photographers. >> > > Yup. > > John > -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
From: Phil Hobbs on 23 Nov 2009 13:00 John Larkin wrote: > On Mon, 23 Nov 2009 01:24:46 -0500, Phil Hobbs > <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >> My processing algorithm will probably be something like this: >> (1) Count the fringes from the time they become visible till the time >> they go away; >> (2) Estimate the deceleration rate by comparing the transition rates in >> patches at the two ends of the pattern, and maybe a few pairs of patches >> in between, which gives the deceleration curve; >> (3) Knowing the wavelength and path length vs time (because we do by >> now), estimate the other things we care about. >> >> No big huge FFTs, I don't think. That could be grim on a PIC! (On the >> other hand, it doesn't matter if it takes 5 seconds or so to do the >> measurement.) >> >> Cheers >> >> Phil Hobbs > > Pass the raw data to a PC and do it there? That would be quite possible, and the prototype will certainly be done that way, via the HP 35665A and Prologix. I'd love the actual instrument to be lighter weight than that, though--more like a Fluke 87, which you can shove in a drawer for a year, then pull it out and it just works. > > We're starting to use an NXP uP, ARM architecture, 280 MHz or some > such, that has a floating-point vector co-processor. About $7. > > John My hope would be that this could be done with a 16-bit PIC and some SPI-attached SRAM, or something like that--battery-powered, anyway. Do those NXPs have power control features that are reasonably easy to use? Being able to crank it up to full speed (which I assume means a watt or so) 1% of the time would be great. A nice development system, a good library, and a decent display would really help. Does it run COMMAND.COM? ;) (I'd happily settle for bash over telnet.) Ideally I want to measure: 1. Wavelength, 2. Collimation, 3. Power, 4. Beam profile, 5. Wave aberrations through fourth order (i.e. spherical, astigmatism, and coma), 6. Temporal coherence, 7. Spatial coherence, 8. AM and FM Noise, and 9. Pointing. I have some idea how to do all these things with a simple optomechanical gizmo and some not-too-ferocious software, though they won't all be in the first version, that's for sure. (Some of them need 2D motion, for one thing.) It also won't do an equally good job on all of them, nor will it work well on all beams. But the things it does do, have to be done really really right--as the teacher might say, "The essence of Fluke nature, Grasshopper, is no ambiguity: give trustworthy results or confess failure." It sure won't happen by accident. To get much further with it, I need some data on beams with accurately known problems, which is my spare-time project for the next month or so. I'll certainly try the VCR and hard disk ideas--I've used salvaged 2.5 inch hard disks to spin hologon scanners, which survive 3600 rpm with no problems, but I'm not optimistic about how they'll do at spinning some comparatively gigantic interferometer thing, much more slowly. Maybe an unpowered HDD spun by hand, to start with. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
From: whit3rd on 23 Nov 2009 13:18
On Nov 22, 9:07 pm, Phil Hobbs <pcdhSpamMeSensel...(a)electrooptical.net> wrote: > >> I'm actually just going to spin it up and do the measurement as it spins > >> down unpowered. That way I should have zero cogging and no jitter due > >> to commutation. > There are no iron teeth (or any other iron) in the rotor, so when it's > unpowered, the only things left to cause angular acceleration are... This reminds me of a toy I used to love. It was a gyroscope, with a wrap-string-on-shaft motor. Do you have to have electric drive? |