[Swede]

I've been experimenting with two motors, one brushed, one brushless DC, both of similar torque characteristics. The brushless is definitely quieter and smoother. But I was very surprised when I did a fairly "scientific" test using software by Logosol.

The servo amps are Logosol, and accept serial commands, outputting digital step/direction. A utility called DCN allows manipulation of the PID parameters, and further will graph position error during a sample move.

I set up the software for a 4000 tick, high speed, high accel. move. For both, I manually manipulated PID to minimize error, yet keep the servos from vibrating themselves to death.

The conclusion: The brushed motors perform significantly better than the brushless, at least in terms of position error. The graph of the brushed motor has the position error spiking upwards as the mass accelerates, then quickly correcting close to 0 for most of the move. As the mass decelerated, the error spiked in the other direction before settling down. The brushed graph was tighter and of lower magnitutde than the brushless.

This was very surprising to me. I thought the lower inertia of the brushless would tip the test in its favor. Any suggestions or comments? I cannot see switching to the brushless unless I can get comparable performance.

[HuFlungDung]

I don't know, Swede, I've never tuned a brushless, but they are used successfully at higher speeds than brushed, so logic dictates that there must be a way.
The graph of position error I try to keep within modest limits: it should look like a smooth exponential curve. If there is a spike, then the settings are too aggressive and hard on the machinery.

[kevincnc]

Did you have a load on the motors? If not, you may get different results with a load.

[Al_The_Man]

One characteristic of DC brushless is the cogging at very low speeds (resembling a stepper) because of the construction and nature of switching, you would have to go to true AC servo's to get the comparable or better low speed performance. If you require very low speed operation with a BLDC it is best to use some form of reduction. Because BLDC and true AC servos have low armature mass and no brushes, higher speed and high acceleration are the recognized plusses.
Al

[Swede]

The moves I programmed were pretty intense for this small machine, and under load only in the sense that the work table is the load, no cutting. In my research of PID and as much hands on play-time as I can do, I feel pretty comfortable tuning a servo with the software that I have. My understanding is to stress the machine for tests like this... high accel, decel, and high top speed. This will exaggerate errors as the mass is accelerated. Then, the PID parameters are slowly modified to minimize the error over the move. Minimizing position errors when the mill is stressed this way will result in even better performance during what might be called "normal" operation.

I'll try to get some screen shots up to show what I refer to. I think that I am going to go brushless anyhow. Ultimately, they'll probably give better performance running actual g-code than the brushed.

[HuFlungDung]

Sure, you should use high accel/decel and as high a speed as is practical.

Doesn't a "spike" in a position error graph indicate overshoot? That may not be what you are referring to (I'll wait for your pics), but that is not something you want to allow.

Do both motors have the same encoder resolution? I'm not sure of this, but I'm supposing that very high count encoders are used to give better control of the motor, not merely because we want to split hairs on positioning.

[Al_The_Man]

Swede, This is a copy of a Tuning manual I use on the Acroloop equipment. It has a built in 'scope in the software so you can see imediatly what the effect of tuning is.
Although the parameters used here may not be exactly the same as is used in your case, it does give a pictorial view of what the effect is on changing tuning parameters and curing instability etc.
They shoot for a following error of less than +/- 5 encoder counts for the entire range of RPM.
The feedback encoder must yield at least 4000 counts/motor rev. Also the torque mode of operation is recommended.
The PDF file is too large so here is the link ftp://www.compumotor.com/bbs/acr/torque.PDF
Also see a very usefull ac servo Handbook ( you have to register with the site) by Sanyo Denki www.motiononline.com
Al

[Swede]

Here's a screen shot of the PID utility w/ the brushed motor. I set distance (in encoder counts, 1000 line = 4000 counts = 1 turn), velocity, and accel/decel.

Note the initial position error as the servo begins to move the table. It spikes to 25 counts as the table mass "lags" the commanded position. The system corrects and drives the error towards zero. As the table decelerates, the error occurs in the other direction, as the servo brakes the mass to a halt.

The brushless curve is similar except it doesn't drive towards 0 as aggressively as this one does. Note the constant velocity error, which is +/- 5 counts, or +/- 0.0002".

The KP, KD, KI, and IL are the primary parameters I have to mess with. It's fascinating stuff, I'm enjoying it thoroughly!

[HuFlungDung]

Wow, Swede, you've really got KI cranked there. Is that a typical setting? I've never had anyone to bounce these adjustments off of, but I've always kept KI pretty low, like less than 10 or so. Otherwise, I'd get the mad oscillation effect. I'm open for comments and advice, though

[Swede]

This software has an "auto-tune" feature which will slowly adjust the parameters and zero-in on a minimum error state. The values it generates vary with the motion parameters I enter. A set of PID for slow motion will be different than one involving faster speeds. I've probably run 4 or 5 different servos through the auto-tune function, and one servo generated values near 8,000. The auto-tune works, but the servos are always "edgy" for lack of a better term. During rapids, they vibrate slightly, resonate, and simply aren't very smooth. I often back off on select values after auto-tuning to smooth it out. As I learn what the parameters do, I am becoming more comfortable with adjusting them manually.

I'm wiring up yet another motor to see how it does. This one is a brushed NEMA34 which is overkill, but I'm curious to see if a more powerful motor will drop the overall error. Ultimately, when I get the mounting figured out, I'm going to test a pair of brushless vs. the brushed jobs that I have been using with a complex g-code.

[ESjaavik]

Could you give us a block schematic of your setups?
I'm a bit puzzled that you can drive a BLDC or a brushed DC using the same setup except the motor? It's a very interesting test. Please let us know more as you go.

And kevincnc's comment may be a good point. Did you check that the ratio of inertia of the motor and the load is within the manufacturer's recommendations?

Also make sure your motor mounts, couplings etc. is very stiff. If not, it will throw in a lot of unknown's. Especially when doing a step function that is very demanding also on the mechanics. You are not tuning the motor(s), but a complete system of interacting masses, compliance and other factors.

[unterhaus]

I assume he's using the same brand, but different drives for the brushed and brushless. Of course, a brushless drive can handle a brushed motor given the proper programming -- there just would be an extra set of output transistors. AMC, Compumotor, MTS, and Electrocraft drives I have in my basement will do that.

Integral gain will add oscillation and the derivitive gain would be all good except it adds noise.

[Swede]

In "researching" for this reply, I have learned something interesting... I think I have been messing with AC, not BLDC surplus servos.

The drives are Logosol LS-182-2010, on a serial bus.

http://www.logosolinc.com/

They apparently sense brushed DC vs brushless AC, looking for commutation signals. I haven't had a problem swapping the two. I've only had one motor fail, and sadly that was a set of six NIB Parker motors of the finest quality. In retrospect, these must be BLDC, and mounting one consistently blows the fuse of my system, in the power-supply region.

My knowledge of motors, motion control, and hardware is weak... I'm trying to learn as I go. The brushed DC servos obviously connect, power wise, with their two power lines; the encoders are wired, quadrature, A, A~, B, B~, I, Vcc, Gnd. Off they go with no problems.

The surplus (AC? BLDC?) motors I've tried have, strangely enough, ALL been either unlabeled, or OEM labeled and every bit of searching I've been able to do hasn't revealed their origins. They use hall commutation. Are they AC? Probably. The Logosol LS-182 docs specifically state Brushed DC or AC, so all along I have been stupid with this sytem, thinking that BLDC was what I was messing with rather than AC. How embarassing!

This explains why my sweet Parker motors haven't worked, being BLDC.

Here is a picture of the two competitive motors, an MCG ID23005, and a mysterious OEM MCG motor of high quality. Labeled MCG 2281-MEB3551. But nothing on the MCG site about them. The long shaft length at the rear was for a brake, which I have removed.

[Al_The_Man]

I
They apparently sense brushed DC vs brushless AC, looking for commutation signals.
They use hall commutation. Are they AC? Probably. The Logosol LS-182 docs specifically state Brushed DC or AC, so all along I have been stupid with this sytem, thinking that BLDC was what I was messing with rather than AC. How embarassing!

.

The presence of three hall effect (or equiv) outputs on the motor generally indicate BLDC, Many manufacturers make an amp that can be used with either DC brushed or BLDC, The commutation feedback for true AC comes in a few different forms, resolver being one. The logosol site indicates that they are the type of amps I mentioned above.
Just looked at at the MCG site and they do not mention AC only BLDC. Those amps they show there I swear are A-M-C.
Al

[Swede]

Thanks, Al, the mystery continues and is not yet solved. I too thought that the presence of hall effect commutation indicated BLDC. I'm going to dig through that Sanyo Denki doc and try to get smarter before I release the smoke from either a motor or the Logosol units.

I went through every page of the Parker SM232 docs, and oddly enough, not once is the acronym DC present, other than the Peak Current in the specs page. More often than not, it's simply referred to as a "brushless" motor. I wish I could get better docs for all the surplus motors I've gathered up from eBay.

[Al_The_Man]

Have you tried contacting the Parker technical dept, I have found them always very helpfull, By the way it is now common practice for the BLDC commutation outputs to be on the same disk as the encoder track, instead of using discrete hall effect devices.
Al

[CNCAddict]

There is really no difference between BLDC and AC synchrnous motors. The difference has more to do with the driver. AC motors are commutated with a sine wave. While BLDC motors use a simple trapezoidal wave. The sine wave provides smoother movement and better torque. AC motors use the encoder to provide precise feedback for commutation while BLDC mostly employ hall sensors. Some AC motors still use hall sensors to provide orientation before AC commutation begins. So like I said in the beginning, saying a motor is AC has more to do with marketing than anything else.

Cheers,
David Bloomfield

[Al_The_Man]

And then there is this............

A "sinusoidal" brushless motor is wound so it produces a sinusoidally varying amount of torque per unit of current in a phase as a function of rotor angle. That is, if you plotted the torque per unit current for a phase vs. angle, you would get a sinusoidal curve. With a "trapezoidal" brushless motor, you would get a roughly trapezoidal curve.

The difference is in the details of the copper winding pattern, often on a common iron-core structure. For this reason, they are often called "sinusoidally wound" and "trapezoidally wound", which is clearer. Of course, neither will yield a perfect curve, but a sinusoidally wound motor comes much closer to a true sinusoid than a trapezoidally wound motor comes to a true trapezoid (as commented above).

The first commutation algorithms were simple "six-step" switching algorithms using hall-effect sensors that matched the trapezoidal waveforms better. They were "on" during the "flat" part of the wave form, and "off" for the "slope". Now "sinusoidal" commutation algorithms using the feedback encoder or resolver are cheap and common, and these match the sinusoidal waveforms better.

The lowest torque ripple comes from a sinusoidal motor with sinusoidal commutation. Next lowest comes from a trapezoidal motor with six-step commutation (worse because of the sudden current switching from phase to phase -- you can often see the marks this leaves in cutting applications -- and the imperfect trapezoidal waveform). Higher ripple comes from a "mismatch" between the algorithm and the winding.

most servo motor manufacturers these days have moved exclusively to producing sinusoidally wound motors. You have to look in the very low-performance market for trapezoidally wound motors.

Curt Wilson
Delta Tau Data Systems
Al

[Swede]

Is there any way, with a use of a DVM or other crude methods, to determine if a motor is BLDC or AC? Not that it apparently matters.

Model R/C planes using brushless motors obviously use battery DC, and the brushless controller then apparently phases the output over three lines to drive the motor phase legs. Technically, then, are these motors DC or AC? They are referred to as sensorless, meaning the motor has no encoder or hall, nothing except 3 phase legs into the motor. I suppose given that the current travels only in one direction, then they must be technically DC.

Maybe I should use whatever motor works and be happy with it!

[unterhaus]

And then there is this............

A "sinusoidal" brushless motor is wound so it produces a sinusoidally varying amount of torque per unit of current in a phase as a function of rotor angle. That is, if you plotted the torque per unit current for a phase vs. angle, you would get a sinusoidal curve. With a "trapezoidal" brushless motor, you would get a roughly trapezoidal curve.

I suspect this is very overstated. Winding a motor is a pretty crude process. We have some three phase motors made by a major American motor manufacturer that have one winding that only pulls 2/3 as much current as the other windings. This is no doubt a quality problem, but the motors work fine. This same company makes perfectly adequate servomotors.
There is just very little difference between the trapazoid and an equivalent sine wave of the same period, expecially since it may be sampled at 3 or 4 points in space.
Eric