affordable brushless motors

[drawbar]

Originally Posted by Mariss Freimanis Sorry but I can't be as sanguine about 6-step BLDC drives. I designed a PID trapezoidal BLDC drive 8 years ago to the point of a finished G320 size drive. I couldn't in good conscience release it as a product owing to the 4 to 5 position error discontinuity every 60 degrees of rotation. A small flaw certainly but large enough to where I couldn't overlook it. I resurrected that drive when I started on this full-tilt FOC adventure I'm involved with now. I wanted it for a reference of where I left-off 8 years ago. I dusted it off and it runs as nicely now as it did when I put it away, warts and all. I'll post pictures or a video of it running if anyone is interested. Just for fun, name just one true DC motor. I can't and I'll bet you can't either.:-) Mariss

I guess I always thought a homopolar motor was a true DC motor. It has no commutation whatsoever.

Brian

[Al_The_Man]

That odd, I have been using BLDC motors in conjunction with AMC drives for many years successfully in CNC applications, I particularly like the Tamagawa BLDC motors as they pack alot of torque into a small frame, used with something like the Galil motion card with analogue control, I have not really seen that significant of a difference between them and the 3 phase sinusoidal type motor.
The 3 phase AC motor is often harder to mix and match motor and drive due to most AC using resolver feedback for both commutation and encoder extrapolation.
I have successfully converted some AC motors to BLDC, although purportedly the rotor magnets have a different geometric shape to the magnet top. I have not yet confirmed this on the motors I have used.
The latest successful venture was converting Fanuc 8 pole AC PM motors to BLDC for both spindle and servo application.
I have talked to both motor and drive manufacturers over the years, and all have differentiated between BLDC and AC as DC Brushless as so named due to only two winding energized at one time, likened to a DC motor turned inside out, the AC has a sinusoidal supply to all three windings.
Their words, not mine.
Al.

[jmelson]

Originally Posted by drawbar I guess I always thought a homopolar motor was a true DC motor. It has no commutation whatsoever. Brian

I think he's got you there, Mariss! But, drawbar, just try to buy one from
a motor manufacturer. If they don't say "huh? Never heard of that!" then
they will say "I heard of a company that made a few many years ago, but
we've never, ever, seen one."

Jon

[Mariss Freimanis]

Does it reverse direction? If it does then it's technically an AC motor. Same goes for moving-coil motors.

Mariss

[jmelson]

Originally Posted by Al_The_Man That odd, I have been using BLDC motors in conjunction with AMC drives for many years successfully in CNC applications, Al.

Yes, all of the early brushless servo designs were 6-step, or some related form, and not sinusoidal. Obviously a drive with 25 SSI TTL chips couldn't support the math to even do sinusoidal drive from a look-up table. Fanuc started doing this in the mid-1980's, so you can be sure there were no DSP processors in each servo amp. The CPU for the entire CNC control was something like an 80286.

With careful design, these work quite well. You CAN detect the commutation, it does cause a slight disturbance, but it is going to be QUITE small compared to all the other disturbances in a real system. Since my engineering budget is a LOT smaller than yours, I really can't invest the time in distilling space-vector into a tiny CPLD. I will be very interested to see how you do this (or at least how much you reveal about how you did it). Obviously, you are going to have to simplify the math quite a lot.

I've been thinking about whether I could make a torque-mode amplifier with sinusiodal drive, but so far it looks too complex to fit in a very simple drive like I make.

Jon

[Mariss Freimanis]

Jon,

The math is way too complex to fit inside a small CPLD. At minimum eight "+/-x sin theta" calculations have to be done and each requires 4-quadrant multiplication to an 8-bit signed precision. Oddly enough a mixed analog technique using digital potentiometers in an unusual way does the math simply and cheaply. The result is a torque mode BLDC amplifier.

Furthermore, the exact same technique applies to step motors except it's a little simpler; the input and output signals are already in quadrature. This allows me to dispense with the Clarke transform and its inverse.

Because of the similarities, it makes sense to develop the BLDC servo and the step motor servo simultaneously.

Mariss

[freak_brain]

WOW!!! you guys have been busy today! It's official.... I'm never asking Marriss to make motor comparisons again!! It keeps you guys away from work all day.

The reason I asked is because I run into them from time to time. I never buy them because I know less about them than pm motors. And trust me, I don't know much about the pm motors. So, maybe I will mess with them one day soon.

Allen

[jmelson]

Originally Posted by Mariss Freimanis Jon, The math is way too complex to fit inside a small CPLD. At minimum eight "+/-x sin theta" calculations have to be done and each requires 4-quadrant multiplication to an 8-bit signed precision. Oddly enough a mixed analog technique using digital potentiometers in an unusual way does the math simply and cheaply. The result is a torque mode BLDC amplifier.

WOW, leave it to Mariss to come up with an insanely devious way to do in pseudo-analog what anybody else would have done with very expensive FPGAs. That is quite tricky!

Furthermore, the exact same technique applies to step motors except it's a little simpler; the input and output signals are already in quadrature. This allows me to dispense with the Clarke transform and its inverse. Because of the similarities, it makes sense to develop the BLDC servo and the step motor servo simultaneously. Mariss

Damn, now he's going to put me out of business! I think my only hope is to stay out of the step/dir business completely.

Jon

[Mariss Freimanis]

Jon,

I hope you are kidding; that certainly isn't my intent.

About digital potentiometers: One way of looking at them is, well.., they are digitally adjustable potentiometers so you use them where you would use a potentiometer. I saw analog x,y multipliers where digital 'x' (0 < x < 1) multiplies analog value 'y'. I see things in a funny way.:-)

Analog can add, subtract, solve integrals and derivatives but sure as heck cannot multiply without a great deal of trouble and then poorly. I'm familiar with most analog methods; differentially cascoded dual differential amplifiers, antilog of summed linear to log, variable amplitude PWM and others. The G203 and the G250 use the variable amplitude PWM method for current set. None are without limitations. 4Q multiplication of two variables is the Achilles Heel for analog.

Multiplication is not so hot in MPUs either. Usually only add and subtract are native instructions. Mutiplication requires the "shift and conditional sum" algorithm which consumes at best about 800 clock cycles for a 16x16 multiply versus 4 for a 16+16 add. The MPUs that have a native MUL instruction still requires at least 12 cycles. The reason is ADD or SUB gate complexity is a linear function of bit resolution while MUL gate complexity goes up with the square of the bit resolution. That is why it's called a "multiplier array unit".

FPGAs have enough gate complexity (>50K gates) that they can afford multiple MUL parallel gate arrays. They are also expensive when support circuitry is taken into account (XTAL oscillator, Flash ROM, voltage regulators, etc.).

CPLDs are way too simple. They are just small gate-arrays; even ADD and SUB aren't native instructions (you got to teach them how). ADD and SUB are expensive, MUL is completely out of the question with CPLDs.

Then there is the problem of having to convert from analog to digital, do the math processing in digital (FPGA) and then convert the results back into analog. ADCs and DACs require support are expensive by themselves. Digital Signal Processors (DSPs) include these converters and do the math superbly but are more expensive than FPGAs.

My idea is to dispense with all this rigmarole and keep all the math processing in the analog domain. No ADCs, DACs, FPGAs or DSPs, no XTAL oscillators or Flash-ROMS, no serious-current low voltage regulators requiring switching-type pre-regulators. What a mess! It was an unworkable idea until a suitable 4Q multiplier element was found. Then it all came together.

Just $5 of low-level circuitry: A small 64-macrocell CPLD, 4 quad op-amps, 8 digital pots and a few dozen RC passives. That and 2mA of 3.3VDC "juice" to run it all. Very sweet and it works!

Funny story: I'm using a very mild-mannered 2A / phase Sanyo Denki 103H step motor at 24VDC for development work. The simplest implementation was to have the amplifier section act as a transconductance (torque-mode) driver. I'd give it a max torque command and the thing got real ratty at high speed. It got nasty actually; nasty enough I thought I had something seriously wrong with the design or the circuit.

I sat down and interpreted the numbers. The motor was supposed to go to 10,000+ RPM at full-torque command and the encoder was set to 1,000 lines. The encoder was rated to only 7,500 RPM at that resolution. Duh,..

Mariss

[jmelson]

Originally Posted by Mariss Freimanis Jon, I hope you are kidding; that certainly isn't my intent.

I was kidding, but we are competitors at some level. I was thinking I had a leg up since you were not making a brushless drive, and now you are going to come out with one that may be better than mine.

About digital potentiometers: One way of looking at them is, well.., they are digitally adjustable potentiometers so you use them where you would use a potentiometer. I saw analog x,y multipliers where digital 'x' (0 < x < 1) multiplies analog value 'y'. I see things in a funny way.:-)

Right, I immediately thought of them as digitally-controlled multipliers and saw how you'd use them. I did use them in a product some years ago, for a more traditional gain-control purpose.

Multiplication is not so hot in MPUs either. Usually only add and subtract are native instructions. Mutiplication requires the "shift and conditional sum" algorithm which consumes at best about 800 clock cycles for a 16x16 multiply versus 4 for a 16+16 add. The MPUs that have a native MUL instruction still requires at least 12 cycles. The reason is ADD or SUB gate complexity is a linear function of bit resolution while MUL gate complexity goes up with the square of the bit resolution. That is why it's called a "multiplier array unit".

well, 12 cycles doesn't seem so bad, you shouldn't need all that many multiplies to complete your algorithm.

FPGAs have enough gate complexity (>50K gates) that they can afford multiple MUL parallel gate arrays. They are also expensive when support circuitry is taken into account (XTAL oscillator, Flash ROM, voltage regulators, etc.). CPLDs are way too simple. They are just small gate-arrays; even ADD and SUB aren't native instructions (you got to teach them how). ADD and SUB are expensive, MUL is completely out of the question with CPLDs.

Sure, no argument there. I was kind of wondering how you were going to produce a drive in a similar price range to your current offerings when it needed FPGAs, ADCs and so on. Typically, you found a BETTER way!

My idea is to dispense with all this rigmarole and keep all the math processing in the analog domain. No ADCs, DACs, FPGAs or DSPs, no XTAL oscillators or Flash-ROMS, no serious-current low voltage regulators requiring switching-type pre-regulators. What a mess! It was an unworkable idea until a suitable 4Q multiplier element was found. Then it all came together. Just $5 of low-level circuitry: A small 64-macrocell CPLD, 4 quad op-amps, 8 digital pots and a few dozen RC passives. That and 2mA of 3.3VDC "juice" to run it all. Very sweet and it works!

Wow! Quite cool, I will be interested in seeing it when it is a product. I might even have to get one just to see how bad mine is!

Funny story: I'm using a very mild-mannered 2A / phase Sanyo Denki 103H step motor at 24VDC for development work. The simplest implementation was to have the amplifier section act as a transconductance (torque-mode) driver. I'd give it a max torque command and the thing got real ratty at high speed. It got nasty actually; nasty enough I thought I had something seriously wrong with the design or the circuit. I sat down and interpreted the numbers. The motor was supposed to go to 10,000+ RPM at full-torque command and the encoder was set to 1,000 lines. The encoder was rated to only 7,500 RPM at that resolution. Duh,.. Mariss

Yes, I have overrun an encoder once, and it also took me a moment to realize what was happening. The EMC2 driver for my control boards has a raw encoder counts velocity output that can be "scoped" by EMC's HAL subsystem, so I was quite easily able to see the encoder velocity showing severe artifacts. I pretty quickly recognized the symptom when looking at a velocity vs. time trace.

Don't reveal too many of your secrets, or I may start thinking up my own solutions!

Jon

[KTP]

Tamagawa TBL-S brushless motors perform HORRIBLE on trapezoidal type drives...you must be talking about a different Tamagawa motor. The ones I have sound like a machine gun at low speeds.

They do perform perfectly on Emerson EN-204 or EN-208 industrial vector drives though

I wonder about Mariss's brushless drive. Can he enter a market where voltages exceed 200V? I always felt there was a liability reason or a cost reason why he keeps the voltage at < 80V on all the step and servo drives.

The new Atmel AVR32 32 bit microcontrollers might fit the bill for a digital design. They fully support ethernet and high speed USB and are about $4 each in quantity 100.