Finding optimal components for servo drive

[Xerxes]

Hi,

I thought it would be wise to ask for component suggestions for an AC servo drive that is most likely going to be produced industrially. I discussed about it on these threads:

http://www.cnczone.com/forums/showthread.php?t=14477
http://www.cnczone.com/forums/showthread.php?t=22028

The goal is to make the best price/performance drive ever and that's why also component selection plays an important role. Because component prices directly reflect to drive price it is favourable not to use the most expensive parts.

------

Here are some components where I would like to see if there are better alternatives. I listed some features that are the most important criteria for choosing part.

MOSFET/IGBTs (current choice: IRF640N)
- 200-250 Vdc breakdown voltage
- higher than 15 amps rated current
- low total gate charge (Qg) (IRF640N, 67nC)
- fast internal diode, short reverse recovery time (Trr) (IRF640N 167-251ns, not very good)


Current sensing op-amp:
-Rail-to-rail inputs and outputs
-Operates on 5V (or even 3.3V) single supply
-Low input offset temperature drift (less than 5µV/C)
-Low input offset voltage (5 mV MAX)
-Low input noise voltage (5-15 nV/squareHz is very good)
-Bandwidth at least 1MHz, 2-10MHz is very suitable
AD8604 looks quite good. Now only if that has less noise... (has 33 nV/sqHz)


Step/dir optoisolator:
-Fast, determines how short pulses it accepts
-High current transfer ratio (CTR) or sensitivity, determines how much current requires to switch on. CTR of about 50% would allow control directly from parallel port without buffer circuit.
-Possibly logic gate output, or just transistor output

HCPL2630 is quite nice but probably needs buffering after parallel port.


Suggestions?

[pastera]

You're asking your conflicting specifications on the FET. A large die FET is going to take some time to recover. Add a fast recovery diode to compensate
MURB2020CT - $0.83 each - 25nS Trr - 25nC Qrr

As far as the op-amp, it's hard to recommend one without the current sense schematic. Are you using the circuit Mariss suggested?

Aaron

[Xerxes]

That's a very fast diode but costs as much as mosfet itself. I think power is not too high to be handled with internal diodes but I'll see if there would be enough board space for optional external diodes. I have read from somewhere that some recent STMicros stripfets would have good internal diodes but they seem to be harder to find.


The current sensing method is quite typical low side current sensing. There will be 0.01-0.03 ohm resistor between the ground and lower mosfet source pin. The voltage over this resistor will be amplified to 0-3.3V range. When current is 0 Amps, the outputs (+ and -) will be at 1.65V potential. See attached schematics (not final version).

[NC Cams]

Find a place on the PCB for the most robust reverse recovery diode you can find. You can always leave it out. Even avalanche rated fets sometimes need external diodes because of the inductive kick they see.

It is better to be looking AT the diode that to be looking FOR it.

Finally, just because it is RATED for something, that don't mean you RUN it at rated values. You have to run at a fractional percentage of the rating if you expect it to live.

[pastera]

The 15ETH03 is only $0.66.
I am quoting IRF parts because they have great site other vendors do have great devices at good prices.

You should look long and hard before committing to low side current sense. What happens when you get a motor fault? You will not be able to sense it. High side sense catches this issue and is not all that hard to impliment.

Aaron

[Xerxes]

MURB2020CT seems to have a more favourable pinout. I try to add some space for them.

Aaron, the low side current sensing is only for phase current measurement. A short circuit detection needs a different circuit but I'm unsure whether I'm putting it in the drive.

NC Cams, of course component "rating" is very misleading. But mosfet may survive if I use a well cooled 18A rated part at 8 to 10 amps at ~50% duty cycle. If they don't then I choose bigger mosfet.

[NC Cams]

If you're going PWM, you can't just look at the current being pulled - you need to know the instantaneous maximum.

Let's say the load is capable of drawing 25 amps at locked rotor. So you run a 50% duty cycle PWM. The average current is only 12.5 but your fets are still seeing 25 amp spikes. A fet rated at 18 amps would live at 12 but would not be expected to live at 25 PWM'd at 50%.

The next thing you need to consider is Vgs. We learned while switching fets on R/C car speed controls, the you needed to run the highest Vgs the fet could repeatedly tolerate to get it to turn on and turn on fast.

For example: an IRFZ40 driven at about 12v Vgs would run warm and needed to be heat sunk under the current demands we asked of it - namely, max rated, sometimes more if you OOPS something.

Yet, an IRLZ44 driven at 10 Vgs could be run with no heat sink whatsoever under the same loading conditions.

For whatever circuit you run, the IR application note AN969 shows an easy to implement voltage pump that will easily provide the needed gate voltage to turn on a fet fast and hard - even if the source Vgs drops to as low as 6 or so.

The only change I"d make would be to change D5 in the circuit from a 15v to an 18v zener for a conventional fet. If you use a logic level fet, a 9.6v zener should work nicely.

Intersting note as an asside: when we drove the Vgs's as hard as I'm suggesting, we didn't see as many popped fets when the occasional overload occured. Almost none as I recall. We literally had none with the IRLZ44's.

Should you drive the fets at/near their max Vgs rating. definitely use a totem pole drive (if you use discretes) or a true fet driver of your choice.

The honest to goodness drivers work MUCH better than merely using a logic level pin out of a logic IC - the drive of a logic level IC is not intended to drive the capacitive load of a fet gate.

We drove with 2N4401/4403 totem poles but that was in the stone ages of fets and their incorporation into R/C cars. Life is much simpler now. I'd find the board space to do it right.

[Xerxes]

Originally Posted by NC Cams If you're going PWM, you can't just look at the current being pulled - you need to know the instantaneous maximum.

True. 10 amps at 50% duty cycle over resistance equals about 7 amps RMS.

I'm using IR2104 gate drivers with regulated 12V supply and with 47 ohm gate resistor. No problems yet. I'm avoiding logic level mosfets since they break down easier even when used correctly.

I couldn't find the AN969. Typo in number?

[NC Cams]

According to my dated, yet reliable 4th edition albeit PRINTED IR parts catalog:

AN-969 = "Economic, high performance, high efficiency electronic ignition with Avalanche-Rated HEXFETS" by Brian Taylor.

It should be on the International Rectifier site someplace. I saw it there the last time I looked - its in the automotive section, not the stepper motor section for what should be obvious reasons.

Moreover, I've posted the reference a number of times and this is the first time I got a reply that the AN could not be found.

We found that logic and conventional fets had faster turn on times with about 10-15 ohm gate resistors. It is the slow turn ons that really boost the heat/increase the on resistance of the fets. It don't take much to shut them down.

If you rely solely on the epitaxial diode instead of an external Schottky to bypass the inducted voltage, yes, you'll over tax the logic level fets. However, with 10 volt gate and external Schottky's, they should work fine...

[Xerxes]

AN-969 seems to be in archived app notes. Good to read.

Using low gate resistor value may not give much boost since IR2104 has quite limited output current. But I'll try different values on next prototype and find the optimal. Very fast switching requires external diode due to dv/dt limitation of internal diode.

[NC Cams]

We drove IRLZ44's (pair of them) with a SINGLE STAGE of an LM324 Op Amp. Gate resistor was 10 ohms and Vgs was about 10vdc. Ran fine at about 100 hz. Surely an IR2104 could easily out drive an LM324 op amp.

We used op amp drives up to 3.2khz in PWM motor speed contols with discrete totem pole drivers and or op amps. Never tried 324's at 20khz.

I'd think a legit driver would have no problem driving them real fast with 10 ohm gate resistor at speeds up to and perhaps above 22khz.

Re: the dv/dt thing = that's why I've advised use of external Schottky since my earliest post.

When you switch off the fets, look real close via a current shunt for a very short duration, very high amplitude inducted current spike. We saw spikes of nearly 50 amps or more in some instances - only a few millisec duration but HUGE amplitude. These are brutal

[Xerxes]

LM324 is a slow op-amp and slew rate is the limiting factor (0.5 V/µs). It shouldn't have made much difference whether you used 0, 10 or 100 ohm resistor since op-amp would take about 20 µs to reach 10V.

I measured my first proto Vgs (IRF540N and 68 ohm gate resistor) and it seems to take about 500 ns to rise from 0 to 10V. IR2104 drives about 200 mA which can be easily exceeded with low value resistors (e.g. I = U/R = 10V/47ohm = 212mA).