wind powered FET destroyer.

[letERblow]

I have built a wind powered FET destroyer.

Wind turbine -
   16 ft Dia
         80 ft tower
   Neo magnet air gap
   3 Phase
   24 overlaped coils per phase
   1v per RPM
   Individual rectified phases, 3ea bridge rectifiers, 600V PIV

Controller -
   micro computer 
   PWM outputs  starting at 80 RPM, full output at 240 RPM ( 240 V )
   Opto coupled through 80ft wire run to
   2 4420 FET drivers
   2 banks of 4 paralleled FETs, Lo side N-channel
      mounted to 1/8" x 4" x 5" water cooled copper plate

normal Load -
   4500 watt electric water heater element (11 ohm)

additional load -
   4500 watt dryer heating element (11 ohm)


FETs are mounted within 4 feet of water heating element and 15 ft to space heater. FET drivers are 2" from FETs.

The FETs are shorting out from Drain to Source, no smoke or blow outs. Things will be running along fine, some times for months then one will short and then the other 3 in the bank will soon follow. The failures do NOT seem to follow any pattern of high loads or gusty winds. In my initial design I tried 480V DC SSRs but they would blow almost instantly. So I thought it must be a PIV problem and used individual FETs with 600 or 900 V VDS, tried a few different types looking for high VDS and avalanche. Some times the FETs would not show a short with an ohm meter but as soon a some voltage was applied it turned fully on. 

Some FETs tried
9N90 -       900v, 9A,  36A pulsed, 1.2ohm
10N90A -     900v, 10A, 40A pulsed, 1.2ohm
IRFP460 -    500v, 18A, 73A pulsed, 0.27ohm
FQA8N100C -  1000v, 8A, 32A pulsed, 1.2ohm
STB15NM60ND -600v, 14A, 56A pulsed, 0.29ohm

11N90s lasted the longest.

My genny has been down for a while for some winding repair issues and when I recently got it back in the air the FETs blew after a few hours. I had some 9N90s left and put one in and increased the input capacitance (C1) by another 100 Uf and also added 940 uF across the heater.  It blew instantly with barely enough wind to get 100 volts. Blew a couple more thinking it must be a bad FET.

I am no electronics designer but can copy other applications and have learned much from this site.
Could there be something in the gate drive? Maybe need a TSV diode on the gate?
I am at a loss trying to figure out what is wrong. Any ideas appreciated.

[phil b]

The first thing I'd suspect would be heat is killing your components. The bridges can make a lot of heat. Try isolating them on their own heat sinks.

[DamonHD]

I would also put some fast reverse EMF diodes on your loads as they probably have a fair bit of inductance...

Rgds

Damon

[letERblow]

Thanks Guys

The heat sink plate is water cooled and a blown FET never feels even slightly warmer than ambient. In my latest trials they have blown almost instantly with the first PWM pulse.

The FETs have internal reverse diodes but I installed external ones as well on this latest go round.

[OperaHouse]

First off never put a cap across the heater, that will make it look like a dead short to the FET.  I like the idea of adding capacitance just after the bridge rectifier.  The mill is inductive and you could be getting a pretty big kick just after the the FET turns off.  I would like to see multiple caps there which would lower the capacitor banks internal resistance.  What is the PWM frequency you are using and do you do anything in software to limit short on or off times?  I think the problem is over voltage.  Your software may not be fast enough to respond to a very quick increase in voltage.  A larger capacitor bank may be the answer to that.  I looks like a great application.  Don't get discouraged, there is an answer.  I would also put a small resistance, maybe 5 ohms in series with the cap across the FET.

[DamonHD]

Do you have reverse diodes across the *loads* rather than across the FETs?

Rgds

Damon

[OperaHouse]

I would also post what you are using for that diode.  Though I fear the "resistive" inductance is not the one to worry about.  In my water heating project I purposely avoided using a FET driver chip.  You could try increasing the FET gate resistance from 10 ohms to more than 500 ohms to soften the transition if your PWM is less than 500hz.   

Another spike protection method is to feed a large capacitor from the FET with a high speed diode.  The capacitor will have a 3K resistor to common that will drain off some of the charge.  Spikes over the normal max voltage will be shunted to the cap and dissipated by the resistor.

[joestue]

Move the connection point of the 0.047uf capacitor from the drain and connect it to the high side of the resistor, which is the + connection point of the bridge rectifiers.
Place any junk diode across the load resistors, rated for about double the maximum voltage you expect to see from the turbine.
You can use the body diode of a mosfet here if you don't have any diodes, short the gate to the source to keep it off.
--use the 1000 volt mosfets you've got if you have any left.

circuit should look like this:
http://johansense.com/bulk/pwm_resistor_2.png
ignore for the moment all the inductors, they are parasitic components and you can't get rid of them.

The connection leads from the motor run cap, mosfet, and diode should be as short as practically possible, to minimize inductance.

can you explain how you've wired up the 4420 drivers?

[Ungrounded Lightning Rod]

First off never put a cap across the heater, that will make it look like a dead short to the FET.

And double-never put a cap across a switch, without at least a small resistor in series with the cap.

Those .047uf caps across the FETs look like the culprit to me.  Every time you turn a FET off, its cap quickly charges up to 48V.  That stores 1/2 CV^2 joules in it.  Then when the FET turns back on, ALL that energy is dissipated in the FET, selectively in the parts of it that turn on first.

1/2 * .000000047 * 48 * 48 = may be only 54 microjoules.  But you're doing pluse-width modulation so you dissipate that much in the turn-on transient of every cycle and have a lot of cycles per second.

Adding a small resistance, of some reasonable multiple of the on-resistance of the FET, cuts the current spike down drastically and dissipates most of the energy in the resistance, rather than the FET.

But why are the caps there in the first place?  If you're trying to limit the voltage across the FET, use a diode to somewhere appropriate (like the opposite power rail) or a varistor.

[joestue]

could be avalanche energy failures but I've pushed those irfp460 fets pretty hard.

I disagree that the 47nF caps are blowing the fets, not at turn on anyways--that's only 10 times the internal capacitance for the ifrp460. while the fets are rated for instantaneous pulse of xx amps.. that's on a thermal time constant of the bond wires and silicon die. The 54 microjoules, or 200 microjoules at 100 volts at 500hz is .1 watts. Many IGBTs for similar applications have 1-3 millijoules of loss at both turn off and turn on.

I would speculate further about unknown unknowns but, until the circuit is fixed i can't duplicate these mosfet killing circuits that keep being posted on this site.
I melted an irfz44 fet off my circuit board through avalanche energy with the same schematic in my prior post (minus the supply side capacitor) and it survived.

the only speculation i would add is i think the gate is getting overvolted due to supply side inductance and a ground loop of some sort somewhere.
yes, putting a zener or tvs on the gate will fix the problem, but there shouldn't be a ground loop, and the supply should be properly terminated and bypassed.

[letERblow]

Thanks for the insight and things to try.

Opera House & DamonHD-
 Have not tried a reverse diode across the *load*, just across the FET. I can put 1/2 of a CSD20060  600V 20A Silicon Carbide Schottky Diode across the load.

PWM frequency is 405 Hz. Currently there is no limit on the shortest on or off time. PWM steps are integer values from 0 to 100. The software is tracking RPM signal which is dampened to avoid rapid changes in PWM outputs so I guess I have designed it in to have wide voltage swings. Maybe a lot more caps can save me or re think the software.

OperaHouse -  could you show a drawing of this method ? "Another spike protection method is to feed a large capacitor from the FET with a high speed diode. "

Joestue-  the 4420 driver has no external components other than a LED on the input and a big cap across the 12 v power that comes from a battery at the controller end ( 80 ft away). The opto coupler is also 80 ft away at the controller. The controller has its own 12v battery supply and all inputs are isolated. The other 12v battery powers input sensors like RPM, wind speed , thermostat and FET drivers. It's neg is common with the FET neg Rail.


Ungrounded -  The .047 caps are for noise suppression, Without them a noticeable hum is present. And I could see a noticeable reduction of spikes across the FET on the scope. "Every time you turn a FET off, it's cap quickly charges up to 48V." should this read charges up to the highest voltage of the generator or a spike, which would be a lot more joules?

I appreciate all the suggestions and will surly give them a test, just wish the wind and my available time would get their schedules synchronized. 

[Simen]

Take a look at user Frackers project he did a while back; http://gilks.ath.cx/~g8ecj/New_Turbine/new_controller.html, and pay attention to his fet driver circuit.

I did a similar project at the time, and learned a bit from him.

[OperaHouse]

You can learn a lot by looking at schematics if you ask yourself why they included that part.  The following is a variant that
can be found in some switching supplies. The capacitor will charge up to the nominal voltage through the high speed diode. After
it is charged the voltages are almost the same so it poses very little load.  The capacitor(s) form a high load to any Spike on
the FET.  These are normally a very short duration and a fairly high resistance will drain off this charge.  If you were in a
little room with a FET and one wire went out to a totally unknown load, this is what i would use to protect it. It does consume
a little power the parts are easier to find than a 500V 20W zener This can also be a very useful diagnostic tool.  If the drain
resistor is increased to a value over 200K, the capacitor over time will charge up close to the maximum voltage the FET will see.
If you have extra wires o your micro and a spare A/D input, I would write a little routine that records the maximum voltage on
this cap.  I would try adding multiple caps in parallel to at least get .5uF if you can't find an electrolytic to parallel with
them.

The 100 ohm .1uF RC network will get rid of any RF noise.  I like RC networks. Until you turn it into heat RF noise will just
bounce around in a system.  A capacitor just stuck across a load or FET can potentially cause extreme currents and destroy the
FET.  I wasn't concerned about that .047 doing that.  More likely the surge currents would destroy the cap over time.  Capacitors
are about the least perfect component compared to the theoretical ideal.  In the old days capacitors were much like a gum wrapper,
aluminum foil on a wax paper insulator.  Now it is a metalized plastic and the "welds" to the leads are not much more than a heat
crimp forming a molten glom.  Open welds are a common failure with repeated current surges.  Polyester is a common capacitor
insulator because it is cheap.  It has problems with high frequencies. You don't want it in a suit or capacitor.  Avoid MKT
capacitors.  Try to find MKP which are polypropylene if you are using garden variety.  Real snubber capacitors can handle several
amps continuous at 100khz but they were costing us several bucks each in quantity.  MKP capacitors can often be found as the AC
line filter in PC power supplies.  Your big capacitor(s) on the rectifiers probably have inductance at high frequencies and should
be supplemented with a couple .1uF capacitors.  he diode across the FET would be suitable to charge the capacitor and find another
diode to place on the load.

Switching the FETs that fast is just inviting trouble.  I would change the gate resistor to 470 ohm.

[letERblow]

Thanks to all for the suggestions. I've been scrounging through my collections and I found enough to try most of the suggestions except the big caps. Will look at EBay. Need to get busy modifying the circuit.

[Flux]

I suspect some of the others are in a better position to advise you, there have been a few good suggestions,

My feeling is that it may be over voltage spikes or it may be gate failure. You have long leads everywhere and that makes life difficult for fets. The heater leads are the main offenders but I worry about the long opto coupler leads.

Firstly the scheme is inherently a boost converter with the fets chopping the alternator winding inductance but you have included the big electrolytic across the bridges and this ought to stop that. Maybe more capacitors very directly to ground of the fets to reduce the inductance may help but what you have may be adequate if the circuit inductance is low enough.

As suggested I would change the capacitors across the fets for R - C snubbers the resistor dissipates the spike energy, without it the thing rings and may be no use at all.

Also I would mount a few pulse polypropylene capacitors (say 2 parallel .47 uf) directly on the fet grounds and feed these from the rectifier positive rail so you have a low impedance supply voltage right next to the fets, then fit clamp diodes to the drains to these capacitors. Any spike above this supply voltage will be conducted directly into this low impedance, keep leads as short as physically possible, you may have to do each fet bank individually to keep leads short enough.

The fet driver must be as close to the fet as possible and use twisted leads for the source and gate, put the gate resistor directly on the gate ( one per fet).  Again zener clamps directly across the gate to source one per fet will be absolutely necessary if you raise the gate resistor as suggested, with 10 ohms it ought not be necessary but won't do any harm. 15v zeners should be ok.

Don't do anything that will form a ground loop in the driver circuits, this is where your long opto coupler leads may be causing trouble.

Flux

[letERblow]

Flux
Thanks for the insight.

Could I impose on you do post a sketch of this filter connections I am a little confused and wouldn't want to get wrong?

"Also I would mount a few pulse polypropylene capacitors (say 2 parallel .47 uf) directly on the fet grounds and feed these from the rectifier positive rail so you have a low impedance supply voltage right next to the fets"

Thanks

[OperaHouse]

Taking a moment to reflect. I went back and read all of your posts on fieldlines and more than impressed with your project.
You are now on a quest to rid your system of gremlins of mythological proportions by hanging a bunch of parts on it.  This
just doesn't sit well with me.  In my UNO threads I have tried to demystify electronics.  After hearing your tale of a FET
eater, many will react as they did to JAWS and believe it really isn't safe to go in the water.

Certainly I have often said that FETs should be driven hard and fast, but speed kills too. Gut feeling is that 10 ohm
gate resistor is making your transitions just too fast and you are creating problems for yourself.  I have been driving my water heater FETs at their maximum rated voltage without any protective diode. As you well know that is not enough time to prove anything.  I use only an opto isolator and a thousand ohm resistor for the drive. This is far short of the 6A your driver chip is capable of.  My FETs are 53A and 55V.  These high current FETs have a gate capacitance if nearly 1700pF.  Three of them in parallel is about 5,000pF.  For simplicity the gates are all tied together. That is a lot to drive so the response is slow. If you are switching at 100khz, that is a problem and will cause overheating.  At 400hz that heating just isn't a problem.Slow the rise time and these spikes just can't be created.  Look at the diode waveform from my water heater thread.  This is actually the waveform driving an inductor of a buck converter at 50V.

I won't tell you what to do.  Education always costs and FETs are cheap.  I would change the FET gate series resistor to more than
a thousand ohms.  Then take your AM radio or whatever you were using to test for EMI and run the test again with your EMI cap removed.  Bet there is a big change.  If the FETS aren't getting hot, you don't have a problem with that value.  Chances are that placing a diode on the load will solve the basic problem but that is just a bandaid. Slowing the slew rate will just make the system more stable.

[Flux]

Not sure if this will work.

Flux

[joestue]

As flux provided another example, the diode across the load is critical,  It returns all the energy stored in the load's inductance into itself, which is why it is required, not optional.
Next is reducing the parasitic inductance and providing sufficient supply side capacitance to swamp the supply side inductance.

I think everyone reading this would have better success using an igbt half bridge, mounting an igbt snubber cap (2uf) across the terminals of the half bridge block. The 2uf cap is enough all by itself except for the worst of supply side inductance scenarios, 600volts at 2 uf is .36 joules, which would be 84 amps at 100uH.. and i think most folks here can get a battery bank wired up to provide much less than 100uH --and few are running 85 amp dump loads--which would be 4Kw at 48 volts.

[Flux]

I am glad the image worked, this is the first image I have posted since they changed the board software.

It was past bed time last night but seeing joestue's comment I should have added some explanation. To DC the idea is nothing more than connecting the diode across the load resistor but there is a difference under pulse conditions. With the diode across the heater connections the clamp is via the big 470uf capacitor and all the associated wiring, which may be a foot or more with a big loop.The critical inductance depends on the area enclosed within that loop.

My scheme has a loop consisting of the length of the pulse capacitor, a diode length and the length of the fet leads, this can be kept under 2" and by placing things correctly the loop area can be very small.

This is probably over kill at the frequencies used but easy to do.

When they first appeared I was wary of IGBTs because of the bipolar transistor second breakdown but the IGBT is not really the same animal and has proved better than mosfets in high power high voltage controllers. Certainly the power modules IGBT and mosfet are way superior to the small packages and seem more robust. They are a bit too expensive to use where the circuit is a bit suspicious so for now the fets may be the way to go. If you can prove the circuit and want to build more the modules may be a simpler and just as cost effective way. I have given up with TO220 packages and even the bigger to3p package is not very inspiring at high currents. The quoted figures of hundreds of amps are only possible in things like 5v computer power supplies, the package leads are just fuses to that current if maintained for a finite time.

Flux

[letERblow]

Flux-
Thanks for the sketch, I can follow a drawing but sometimes my interpretation of a narrative is not quite right.

I will give these suggestions a  try and take some scope shots to see which ones look the best. I will keep you posted but it may take me a while with other projects needing done before the snow comes.

[letERblow]

 
Taming the FET destroyer-

I have been running some tests over the last few months, wind and time permitting, trying to get a handle on what is going on with my controller.

First, easy and most logical, calm down the wild AC to DC with more capacitance across the bridge rectifier output. Added 6 ea 1000 uF of 450v electrolytic capacitors. Very weird side note here, see below.

put a 0.1 uF across the DC buss.

Put the RF filter from a computer power supply across the drain to source on the FETs.

Built the Fast diode, 0.47 uF 15 ohm filter suggested by Joestue & FLUX from drain to source.

put 1/2 of CSD20060  600V 20A Silicon Carbide Schottky Diode across the load.

modified FET driver board so it would be easy to select different gate drive resistors
10, 100, 470 ohm.

Took some 9N90 FETs 9 amp 900v with 470 ohm gate drive, thought I would test with one at a time at low wind /amps, it would be a good opportunity to look for spikes. As the genny came up to speed I was ready with scope to see the PWM start.
Volts built to 80v the PWM turned on with 1% and the FET shorted within 1 second didn't get a chance to see anything.

Blew a couple more instantly on successive tries.

I had a couple of IRFK3DC50 600v 24a half bridge hex packs I had been reluctant to try ( destroy) but decided to try one while waiting for some more 11N90Cs. Had all the filters on, 470 ohm gate drive. Another low wind event, volts rising, pwm started 1 -8%, controller was loading the genny, things were looking good. Could not see any spikes on scope across FET, signal was quite rounded on turn on and off, maybe 10% of pulse height and on time. ( scope will hold but doesn't
have storage or down load and I could not take a decent picture with the glare on the screen) After about 5 minutes of 0.5 to 1 amp flowing the FET was getting real hot. I didn't notice it in the specs but on the drawing it showed internal gate resistors so I removed mine and the FET pack ran cool. Could see no spikes, had sharper turn on and off, the RC filters stopped the 400 hz buzzing I used to hear.

I received some more 11N90s so decided to see how they would work with the new filters and higher gate resistors. Would try it with one FET, after all they are rated at 11 amps and it takes a very good wind to get that. With the IRFK3DC50 in the water heater side and a 11N90 on the space heater, 470 ohm gate resistor, same filters on each. Had the genny running with 1.5 to 2 amps going through the IRFK3DC50 and switched it over to the 11N90. It ran about 5 seconds then shorted out, did not get a chance to see anything on the scope.

Connected two 11N90s in parallel, individual 470 ohm gate resistors, ran for a few seconds and one FET shorted. Got to thinking, in parallel they may not turn on at the same time and one is taking all the surge, remembered OperaHouse tied the gates to common bus with one gate drive resistor.

Connected two more 11N90s in parallel, gates connected, one 470 ohm gate drive. Ran it for a few hours this way, could see no spikes, on and off edges were quite rounded. With 1 to 2 amps it started to get hot so reduced gate resistor to 100 ohms, the on off edges were squarer and the FETs ran cold. I let it run for several hours over the next few days less than 4 amps. One day the wind picked up with the opportunity to test at higher loads. Scope signals showed no spikes,a little jittery at the on and off corners. The amps picked up to 5 to 6 amps and one FET shorted after a few minutes.

Getting discouraged and tired of all the running back and forth testing I just took the other IRFK3DC50 and connected it to the space heater circuit. Have been running this way for two weeks, mostly less than 1000 watts but there have been a couple of high wind events pushing close to 10 amps at 220 volts, genny furling nicely at 20 mph.

Still not sure what was/is causing the problems, I can't help thinking it is the surge current, one of the paralled FETs turning on a fraction before the rest and taking all the surge. The 96 amp pulse current of the IRFK3DC50 verses 46 amps for the 11N90.

We hung a lot of parts on things without definite proof that they had a beneficial effect but in the end the system is working a lot better and the time and money has been spent. Question is, do I leave well enough alone or let my curiosity lead me to further testing, maybe later.

that's my story and I'm sticking to it! ( unless my wife tells me different )

Thanks again for all the help!


tale of capacitors
At first I left the existing 200 uF of the AC motor run capacitors in parallel with the new electrolytic s. The digital scope was hooked across the FET, volts and time settings set for the anticipated signals, and  my good the digital volt meter on the DC buss. There was a mild wind, genny putting out about 150 volts. As the PWM started to turn on the FET I was watching the scope to see if I could catch any spikes and he scope's screen filled up a solid black, the volt meter was beeping and the reading jumping all over. I quickly unhooked them and thought Oh crap have I destroyed my instruments. I tested them on a battery and they seemed to function properly. Each time I attempted to reconnect the volt
meter to the wind DC buss it would beep and freak out, actually I didn't even need to connect it just get the lead close to the buss. Tried a different DVM, same freaky readings. The scope would react the same just getting the probe close the the FET. Went outside and the genny was making some weird humming noise. There was some super high frequency being generated somehow. As it turned out I took the motor run capacitors off and everything returned to normal. Didn't dare put them back on to see if the problem returned, thought better leave well enough alone.

[OperaHouse]

Well, it looks like you have fallen down the rabbit hole and no one wants to follow you there.  All the canned solutions
haven't worked but there are clues to what is happening. The problem is obviously external of what your focus has been.
First thing to remember....Regardless of what the amp meter is saying, the FET is seeing full current with every pulse.
A couple of questions first.

Is your PWM rate 490Hz or have you changed it to something faster?

What is your software trip voltage that the PWM initiates at and how often does it sample?

Is the LED drive signal to the opto isolator isolated from the FET drive ground.  It should make a complete loop back
to the micro and not share any current path.

How is the FET driver power being supplied. What does it actually tie into on the microprocessor end. This could be a
potential current loop.  If you have an old power transformer with about a 10 ohm winding, that could be inserted in
this power lead.

People are enamored with infinite control. You mention a PWM OF 1%. Short duty cycles can cause nothing but problems. 
Never let the PWM operate at values under 20 or over 235.  If the PWM count goes under 20, set it to zero. If the PWM
count goes over 235, set it to 255. Keep the normal count the same, modify the number as it goes to the PWM pin as is
shown in the following code.  You have a lot of mass in your mill.  It is OK to let it bump along.
 
 PWM3drv = waterheatercount;                   // Set PWM output to count

 if (waterheatercount >= 235) PWM3drv = 255;   // Limit upper boundry of PWM

 if (waterheatercount <= 20) PWM3drv = 0;      // Limit Lower boundry of PWM,if PWM < 20) output zero               

 analogWrite(3, PWM3drv);                      // output protected values 
                                               
How about a picture of those high voltage capacitors.  As a general rule i don't like to have more than an amp of
current in a consumer grade capacitor.  Your FET will be doing 10-15A pulses.  The voltage with that current will also
be bouncing around. Capacitors are not perfect but if you have enough of them in parallel they will average out.

The fact that bringing leads near this and having instruments go crazy indicates some oscillation begins as a result
of some induced signal.  This problem can be solved.

[joestue]

Photos will help.

1% duty cycle at 490hz is 20 microseconds, and that can't be done with 1K ohm gate resistors.

Still not sure what was/is causing the problems, I can't help thinking it is the surge current, one of the paralled FETs turning on a fraction before the rest and taking all the surge. The 96 amp pulse current of the IRFK3DC50 verses 46 amps for the 11N90.

To switch this much current you need something quite a bit different than what we saw in the original photos, if you want to switch that much current you need a 3 layer copper sandwich of sheet metal and fiberglass-epoxy. you could actually use 2 oz circuit board, it might be possible with two layers.

what i recommend you do is size the resistor to pull 10 amps per fet if you want to use point to point wiring and the to-247 package, use 10-20 ohm gate resistors.

add more dump loads in parallel and program the microproccesor to output a different frequency to each dump load.. use prime numbers so they stay perpetually out of phase for a very long time.--or program them to be out of phase.

[Mary B]

I think you have nailed it. He has some feedback loop and is creating  a power oscillator... FET drive signals need to be carefully isolated. Or you get pickup from the output and it builds in a self sustaining oscillation until the FET goes poof.

The fact that bringing leads near this and having instruments go crazy indicates some oscillation begins as a result
of some induced signal.  This problem can be solved.

[Tight Yorky]

Happy new year everyone! 

Hi LetERblow,
Sorry to hear about your MOSFET destroyer, but I see you are hanging in there.
Although I may not be able to solve your problem directly, I may be able to give a bit of advice to assist.
With MOSFETs I have found that failure is not always associated with Source/Drain but can be Gate/Source.
To avoid excessive device failures look up your device's maximum tolerance for voltage across the connections and insert a zener across these.
Do not protect these zeners with current limiting resistors. They are sacrificial. Ie Their purpose is to be destroyed prior to the MOSFET. For example if your Gate/Source is 20V and you are driving with 12V, add a 15V zener. If a zener gets warm you have a problem.  A tip I was given by a college technician was to put a dab of wax on them. While testing, if you smell wax, switch off quick.
Do not put capacitors directly across the Drain/Source. If you need a snubber capacitor. Charge through a diode. Discharge through a resistor (to suit the MOSFET capacity).
Your space heater load may have a large inductance as many of these are spiral wound resistance wire (normally on steel supported porcelain). Both heaters may have a much lower initial resistance due to low temperature and be higher rating to that on the plate/sticker. These issues may be a factor with your MOSFET sizing.
Relating to the MOSFET driver, an excellent chip I have had success with is the FOD3182. This has the optocoupler and driver in a single device. All matched up, micro one side, power circuits the other. The chip is placed as close to the power transistor as possible.
Hope the above assists.
Good Luck.
T Yorky

[OperaHouse]

"Looks easy.  Just try going a couple rounds with this water heater."
Firesign Theater, Future Fair

I've been waiting 40 years to use that line.  I've done it with 60V (a few volts above the rating of the FET), no flyback diode, no RC network to get rid of RF, with only an opto isolator for drive with a 1000 ohm gate resistor.  No warmth at all to indicate the FET was operating.  A AM radio had to be held within 6 inches of the wiring in order to pick up any noise.  Of course I eliminated any short pulses in software and the 1000 ohm gate resistance with 3 FET capacitance close to 2000pF made turn off so slow it was hard to develop any spike.   Just look at this other control they are selling rated at 10A 200V.  Those caps ought to explode in a couple of years, but it works.

The point is what has been posted should work even if it isn't an optimal design.  All the canned solutions haven't done a thing.  That tells me it is how the circuit is connected.  If I was there the problem would likely be obvious and no test equipment would be required.  Some really detailed photographs and schematic of how everything is connected would be necessary.   First thing to do is change the software to eliminate short pulses.  THAT MUST BE DONE.  Had the FETs been driven with just the opto like I did, there wouldn't likely be a problem.  As it is the FET driver is likely amplifying a noise problem and driving the system into oscillation.

[letERblow]

Thanks for taking the time to review my problem. I will get the info gathered to post answers to questions. There has been no wind since Christmas day, good thing as its been -15F, extra wind chill would not be good.

Questions I can answer now

OperaHouse ;
                     PWM rate is still 405 Hz
                     I have a look up table RPM vs PWM out 20 RPM increments,
                          It has been set to start at  80  RPM (80v) =2% 100 RPM = 5%. etc with interpolation   between.
                          PWM out is % of 1023.
                          I can add boundary limits to the code.

The only time I even saw a hint of the HIGH Frequency oscillations was when the new electrolytic capacitors were paralleled with the existing motor run caps.


Joestue:  sorry to miss lead you with statement
"Still not sure what was/is causing the problems, I can't help thinking it is the surge current, one of the paralled FETs turning on a fraction before the rest and taking all the surge. The 96 amp pulse current of the IRFK3DC50 verses 46 amps for the 11N90."
These are the PULSE Current ratings from the specs on these FETs, not my load.  Maybe I need to measure the ohms of the analog amp meter shunt and put the scope across this and see if I can determine the pulse current the FETs are seeing.

I have tried 15v zeners from gate to source in the past and even once in this last go round.

thanks to all, we will beat this yet!

[letERblow]

Some more info in our quest.


   "Is your PWM rate 490Hz or have you changed it to something faster?

No it is still 405Hz
would faster PWM be better as at low PWM% gen has longer unloaded time to generate higher volts? available 901Hz

and  1807Hz

   "What is your software trip voltage that the PWM initiates at and how often does it

sample?

80 RPM / volts   sample about every 0.3 sec

   "Is the LED drive signal to the opto isolator isolated from the FET drive ground.  It should

make a complete    loop back to the micro and not share any current path.

Yes , the opto isolator LED is driven from the isolated uProcessor power supply . it is located at the uProcessor (

turbine control panel). Switched side has a 80 ft run to the FET driver. I have monitored this line and have seen

no unusual noise.

   "How is the FET driver power being supplied. What does it actually tie into on the

microprocessor end. This    could be a potential current loop. 

FET driver is powered from the analog circuits 12v battery located back at the turbine control panel ( 80 ft away)
3 pair twisted shielded pair cable. pair 1 = + - FET driver power. pair2 = PWM1 and PWM2 signal with return on pair

1 -. pair3 = water heater temperature switch and spare. shields grounded one end.

the new 450 v electrolytic s 5 ea 1000uf 2 ea 450 uF

Just a thought - more caps would adsorb more joules from a high voltage rise to dampen voltage changes but also has

more joules to instantly discharge through the FET. High amp pulse

The old motor run caps that were removed


FET board with IRFK3DC50 600v 24a. These have been working well.

FET driver board


Schematic of system


Thanks for all the help

[letERblow]

correct photo of new electrolytic caps


I realized the schematic is to small. will try to re size and re post

sorry

[joestue]

Something like this is what i meant by low inductance:
http://www.stevehv.4hv.org/DRSSTC-.5/bridge1.JPG

I can't find the diode you have connected across the load resistor in the photos.
the distance between the capacitor, diode, and mosfet has to be as low as possible, as in they need to be bolted to the heatsink right next to the mosfet, and connect one of those 1uf film caps that i think i see. directly to the fet and diode.
yes you can use wires to connect the other capacitor, but use zip cord or twist a number of pairs together.
and don't connect any capacitors directly across the mosfet.

The oscillation you reported when mixing the electrolytics with the motor run caps is likely due to the inductance of the wiring connecting them together.
It doesn't matter how many are connected, they only need to be able to handle the ripple current drawn by the load, which at 50% duty cycle is about equal to the load current.

[letERblow]

diode across load not in photo but at water heater element is  1/2 of CSD20060

the diode on the spike filter, next to small blue cap is LXA03T600

enlarged system schematic

[joestue]

i should have seen this error before, i appologize for that. hope you haven't blown up anything since.

the RCD snubber needs to be connected like this:


when the rcd snubber is connected to ground, that resistor passes current through the diode whenever the fet is turned off. --this will burn out the resistor in seconds.
It needs to be connected so that any voltage on the drain that is higher than the main dc bus gets dumped into the capacitor through the diode only when the fet turns off. and the resistor can be something like 15 ohms. The snubber resistor should NOT get warm at 409 hz pwm.

All of the components on that schematic need a minimum of inductance between them, with the exception of the load resistor. you can make that as inductive as you want, it does not matter.. that's what the flyback diode is for, and it needs to be installed close to the 10uf film cap (use whatever you have) and the mosfet, just a few inches from each other.