3 phase inverter short circuit protection

Why does the logic take so long? 10usec is an eternity.

Maybe it has to first file a shut-down request, in triplicate, with some new-fangled operating system ...

I am not too familiar with IGBTs but why don't you hang a logic gate or FF up there that doesn't only limit but that trips it shut instantly, signals the controller, and only the controller can release it out of a shut-down? Ok, 600V is no small feat but opto-couplers are cheap. If the last penny counts you could probably even get that transferred to 3.3V or whatever level with resistors and BJT since now it (probably) doesn't have to be fast anymore. It's only two lines, error signal and release line. Could also be combined.

Why not use a fast scr? It would shut off the igbt for that drive cycle, not merely current-limit it.

It would need a bit of lowpass filtering into the gate, so spikes wouldn't fire it. The transistor would, too.

Incidentally, the classic current limiter that you drew tends to blow out the npn, even when it seems impossible. So add a base resistor, at least.

John

why not make the controller a LOT faster ? what are you using ?

I wouldnt suggest using analogue current limit as this doesnt limit the power disipation very well.

thyristors are too slow, IME they dont recover from fualt conditions quickly enough, they might conduct from one pwm pulse to the next, and as they stop the fualt the logic keeps on banging out the pulses.

my 3ph pwm controller mcu has about 25ns delay on the fualt inputs or though i dont actually use them :o

Colin =^.^=

Won't that just force the IGBTs into the linear region and result in huge dissipation?

Firstly speed up your control circuitry. 10uS is a bit sluggish. You should be able to get down to 100nS without any real effort and much quicker if you really try. Then put an inductor in each output line which will limit the rate of current rise to below your circuitry protection time delay. That's the way I've always done it and it works well.

The case should be earth not HV return. A phase to phase/earth short is very hard to protect from and as its such a rare event allmost no commercial design implements it at these small current levels. It's not worth doing to protect a 1$ igbt, a different story with a 100Amp drive however.

Not enough detail. Small IGBTs like these are usually controlled with a purpose chip. IR makes a whole line of them look on their website.

Only 1, it wont work. With full gate drive voltage you IGBT will die quicker than 10uS from a phase to phase/earth short.What you need to do is reduce the gate drive voltage during the short, this cuts the short current considerably allowing the IGBT to survive for much longer. If the short is still present after say 20uS you kill the drive voltage completly, if its gone (ie the freewheel diode in the other half of the bridge has recovered) then you allow the full IGBT gate voltage to keep conduction losses low.

In this case there is no "earth" as the machine (motor and power electrioncs) in moving vehicles. Vehicle chassis is tied to HV return.

Yes, not shown in the ascii art is the gate drive transformer with a small

This is needed because the armored shielded cable between the motor and the electronics will take a beating and will likely get smashed,squished or cut in its lifetime. If, when it shorts, I don't want the drive electronics to be destroyed.

This is exactly what I'm trying to accomplish. In addition to the linear clamp shown above, there is some (currently) slow current limit shutdown that kicks in at 10uS after a short is detected. The limiter above will limit the peek current until the shutdown dignal makes the loop back to the gate of the IGBT.

Yes but it 600V @ 5A in the SOA box for the device

Yup I need to speed up that turnaround time. The logic currently in place uses some SLOW components and heavy filtering. I'll have to look at reducing the filtering some and upgrading the comopnents to higher speed logic (they currently have CMOS4000 for the high voltage rating....ZZZZZzzzzzzzz).

thanks

Your vehicle frame must be earthed for safety reasons

It's very difficult to use a transformer for a wide rangeing mark/ space ratio, due to the need to maintain a constant volt/second product.

Probably cheaper to make the power section as a replaceable module.

?What you need to

Well you cant do that with a simple transformer drive circuit. How would the drive voltage recover to full voltage?

Actually, if you are talking a battery vehicle and hope to meet UL583 then the frame cannot be attached to battery potential (it must float). That implies the case must float. And being a vehicle the wheels pretty well insulate the frame from 'earth'.

Now UL583 is an industrial tractor standard but it's pretty basic. Although the high voltages suggest the OP may be running somthing more like a pantograph fed line in which case the requirements may be different.

Robert

Don't use transformers. There are lots of optoisolators on the market with totem-pole outputs, specially designed to drive IGBTs. Some will source a couple of amps peak.

Have you considered "intelligent power modules", 6- or 7-pack IGBT modules with built-in drive and protection. 3 phases and brake. Low power drive and no problems regarding input capacitance. One IPM, six optocouplers, and 4 DC-DC converters, job done. Will even stand two devices in the same leg going on together (for a while at any rate). Switching times of a few hundred nanoseconds. You should be able to get a 15 amp device for way less than a hundred bucks.

Firstly, figure out how hard you can thump your IGBTs.

Its a transient adiabatic thermal analysis problem. Moderate- and larger-sized IGBTs (eg anything that has screw terminals) can typically eat 10us of desaturation.

When an IGBT switches into a short, it "desats" - Vce pulls out of saturation, and inevitably rises to full DC bus voltage (thus assumes the "short" is a lot lower impedance than the IGBT. In which case any arguments about your current limiter leaving the IGBT in its "linear mode" are moot.

Via the magic of IGBTs, they also limit the fault current. for 15V Vgs, this is typically 10x rated current, so a 600A IGBT limits at about 6kA.

The IGBT then sucks about (10 x Irated) x Vdc_bus, until you turn it off (or the die melts).

say a 600A IGBT with an 800V DC bus, thats 6000A x 800V = 4.8MW. In 10us thats 48J of energy dumped directly into the die.

Its an adiabatic problem (or nearly so) because the die is bonded to a large lump of metal, with a thermal time constant well in excess of 10us.

knowing your IGBT rated current, DC bus volts and die size (along with the density and specific heat of Si), you can easily calculate how hot it will get in that time or, conversely, how long you have before it gets "too hot".

"too hot" is a somewhat vague term - make sure its well below 200C,

Powerex (among others) give curves of dT-vs-no of cycles. Its a very steep curve on a log-log graph. 30K dT will do hundreds of millions of cycles; 100K dT will do ten or so.

I have personally proven this, by making a drive desat (which tripped it) then resetting it and pressing start. When done rapidly (c.f. thermal time constant of attached heatsink), the device Tj ratchets up and *kaboom*, after only a few hits. When done slowly, I've managed about 100 desats (70A 400V drive) before it failed. fun stuff, best done after friday drinks :)

The desat time should be as long as possible to prevent false trips, but short enough to ensure the unit doesnt fail. A smart person ensures their software runs a thermal model which tracks the die temperature during fault conditions, and prevents users from ratcheting up Tj as described above. Of course said model works during normal operation, too (many do the latter but not the former).

module-based Tj sensors measure baseplate temperature, and so wont give Tj; if you do a crude Rjb model to then give Tj, it will be completely wrong during desat-type events.

Dont forget that when you turn a desatting IGBT off, its collector current is HUGE so turn-off over-voltage spike V = LdI/dt is correspondingly huge. I have worked on an entire family of drives that, in the event of a desat, turned off the IGBTs and then killed them with the resultant over-voltage spike. oops.

one solution here is to reduce Vgs during fault conditions, as Vgs controls Idesat. Your theoretical current limiter basically does this, except it wont work as drawn, a-la John Larkins comments - think how high the current will get! then look at the peak power in Rsense, it'll probably fail too.

A better solution is to combine the "add a little L" suggestion with fast DC Current Transformers sensing the actual output current, and hardware current limit (dont forget to constrain the resultant Fswitch) and trip circuitry.

With a good DCCT and a reasonably fast current trip circuit, you dont actually need desat circuitry to control/survive an external cable fault.

I have personally done desat tests on many drives, and with Itrip hardware its actually very hard to make a drive desat - the current trip circuit tends to ghet there first, due even to wiring inductances. In several cases I have had to use broad, flat Al strips to short out one IGBT, as ~300mm of 4mm^2 wire was inductive enough to allow the Itrip circuit to operate.

And a modest amount of Lout (10s of uH) can be enough to let the current limit hardware actually control the fault current. With a decent SW loop time, the controller itself ought to be able to directly control the current into a dead short (BDTD). Danfoss always used to give demos of this at trade shows - they put a 500V/us slew-limiting filter at the output of their drives.

Of course a desat is still required, as due to thermal cycling (or any of a variety of other faults) it is still possible to turn both transistors in a half-bridge on.

I also second the "transformers suck for IGBTs" comment. you really need a gatedrive bandwidth that extends down to DC.....can be done with a transformer, but need to think harder ;)

HTH

Cheers Terry