DIY Electric Car Forums banner

Open Revolt with IGBT driver blew IGBTs

46K views 100 replies 14 participants last post by  Stiive 
#1 ·
Hey guys, for the past few months I have been putting together an Open Revolt controller with a VLA500 IGBT driver and 3 Fuji 2MBI300L-060 IGBTs. I've read about a few other people doing the same thing and I've read the documentation on driving IGBTs and minimizing inductance. However it seems that I've done something wrong because today we tested the controller and 2 or 3 of the IGBTs blew. We first tested it with no load or battery hooked up, just to test the IGBT driver. The signal looked fine, with the low Base-emitter signal being -8v, and the high being 16v. We then hooked it up to a 12v battery and a 30-ohm burner, and it ramped the current up and down just fine from 0 to .5 amps. I did not notice anything weird about the gate drive signal on my oscilloscope. Then we hooked up our 144v traction pack and when we opened the throttle up, a high whine and then a pop was heard, and current went up to about 7 amps through the burner.

Here are some pictures if anyone can spot any stupid mistakes I might have made. The gate resistor on each IGBT is 5.6 ohms, as per the IGBT spec sheet. Unfortunately, I didn't bother to implement any voltage clamping diodes on the gate wires. Could that have been my mistake?






Any advice or comments are appreciated! Thanks!
 
See less See more
4
#44 ·
Well yesterday we got some fancy new snubbers and a big electrolytic cap! Per Jack Bauer's advice I installed a 33 ohm gate resistor on the IGBT. With a resistive load and no snubber, there's basically no spiking and the miller plateaus are very clear. I guess that is a good indication that this is working ok.

Then we hooked the starter motor back up and added the new snubber. Unfortunately it still spikes to 20-30 volts upon turn off. As Jack Bauer pointed out, the IGBT did get warm. We were doing about 30 amps peak, with the motor.. We then hooked up 24v to see how big the spikes got and they jumped to 50-70 volts. It seems that the freewheel diodes are not doing their jobs very fast, or at all. However, the spikes are much smaller and we can now see the oscillations in the voltage, instead of the turn off and turn on being a huge spike up followed by a curve down. We will post oscilloscope pics soon to illustrate what I'm talking about.

I guess I expected the new snubber and high gate resistor to solve our problem, but at least we're making progress.
 
#45 ·
Sometimes the spikes can actually be noise being picked up on the oscilloscope. A good way to see what the spikes really are is to put a diode from B+ to another capacitor + (like 330uF or whatever you feel like using), then connect B- on the controller to B- on the new capacitor. It will act like a peak detector. Or a crab trap! Little electron crabs go into the capacitor, and they want to leave, but they can't! The stupid diode won't let them. Then you boil them for 20 minutes in ocean water, and eat them.
 
#46 ·
One easy way to see if the scope is picking up noise rather than a real signal is to connect the probe to the same point as its ground reference on the device. It should not show anything significant. Otherwise, what you see is probably inductively coupled or transients in the grounds. Sometimes a full differential reading is better, using two identical probes, and using the scope's A-B mode. Sometimes you can adjust one of the gains for minimum noise.

Sounds like good progress. Turn-off transients can be a bear! ;)
 
#48 · (Edited)
Today we put on a 2200uF electrolytic capacitor and used that in conjunction with the new 5uF snubber. Here is the result



This is with one IGBT, one electrolytic and one snubber cap, inductive load, 33R gate resistor, 8khz switching circuit. Purple line is ground and green is signal. 10 volts per vertical division and 20us per horizontal division. The battery pack that it's switching is about 24v.

Does anyone have any idea why the IGBT is switching off twice in the time it's supposed to switch on once??? and we have these huge capacitors and gate resistor but the spikes are still large and there is this big curve down after the turn off? I connected the oscilloscope probes to the same point on the ground and there were only small ripples in the signal. I don't think this is enough to create the voltage spikes that we see. I will do the diode/cap test next, but even if it's reading too high this waveform makes no sense :confused:

Perhaps there is something very wrong with the Freewheel diodes... I guess I know what I should do now
a. augment/replace freewheel diodes??
b. do diode/cap test to verify oscilloscope results.

Any input is appreciated as we are stumped
 
#49 ·
I assume the waveform is the voltage from the bottom IGBT collector to GND? What is the gate signal under the same conditions? It appears that there are two fairly solid turn-ons, but one is about 20 uSec while the other is about 30 uSec. And these have 5 volts drop, so if you are running 50 amps that's 250 watts! Then there are the second turn-ons 6-12 uSec later, but there is a much higher voltage drop of 20-15 volts, which is an even higher amount of power.

You may be exceeding the safe operating area (SOA) of the IGBT, which can happen when there is a very high power dissipation for even a short time. It is generally destructive, but maybe you've been lucky so far.
http://en.wikipedia.org/wiki/Safe_operating_area
http://en.wikipedia.org/wiki/IGBT
http://www.electronics.dit.ie/staff/ypanarin/Lecture Notes/K235-1/2 Transistor-Thyristor.pdf

Apparently older devices were more susceptible to second breakdown, so if you have IGBTs from old drives that may be a problem

It would be very helpful to show your complete schematic, especially the base drive components, FWDs, and any snubbers or capacitors or inductors you have in the circuit.

If you have a very high snubber capacitance across the IGBT, the turn-off may cause a transient due to the motor inductance which could result in a breakdown of the IGBT, especially if the gate drive is too "soft". Thus the IGBT will act as part of the snubber and it will dissipate the inductive energy rather than the FWD. And the FWD also might be too slow, or defective.

So if you can show exactly what you have, I might be able to help.

Oh, and another clue is that the saturation voltage seems to be decreasing with time, which also hints at slow gate drive. If the IGBT is turned on full, the voltage should start very low and then rise as the inductance saturates. And it should be no more than 1 or 2 volts. If this is to be used for a motor that will draw 500 amps, you want to limit the peak dissipation to maybe 500W. If the IGBT has a 5V drop on a little starter motor, it will surely blow up on an automotive traction motor.
 
#51 ·
That helps. But I don't see the freewheeling diode. And a proper snubber is a resistor and capacitor in series, across the motor terminals, or even better might be across the IGBT. Do you have part numbers for the IGBTs? And the specifications for the driver circuit. I'd like to see the gate drive waveform as well. And I think you said the motor was a 12V starter motor, and I think you said you had 30A peak. Is that the peak of the PWM pulses, or the maximum RMS? :confused:

Also, your symbol for the battery is reversed. The wide line is positive, at least in my experience. And previously you said the battery pack was 24V, but the schematic shows 12V. :confused:

I found another good reference about IGBTs and their characteristics and optimum drive techniques:
http://www.irf.com/technical-info/appnotes/an-990.pdf
 
#52 ·
Gate is tied to emitter on the other IGBT to keep it off and act as FWD.

Woops on the battery :p

We are performing experiments with both 12V and 24V sources to try to understand this.
The amp numbers are RMS.

The IGBTs are Fuji 2MBI300L-060 (datasheet attached).

The gate waveform has been consistent, looks good as far as we can tell and is pictured in one of the above posts.

Thanks

View attachment 2MBI300L-060.pdf
 
#53 ·
I assume this is the gate drive? It looks like the peak gate voltage is almost 25 volts, while the absolute maximum is 20, so you should be around 15V. And this waveform looks like about 120 uSec out of 160 for a duty cycle of 75% and a frquency of 6.25 kHz.



But your more recent screen shots of the C-E voltage look more like about 20% duty cycle at 8 kHz, and the saturation voltage looks way to high, especially if you are running only 30 amps RMS. But at 20% duty cycle that's 150 amps peak.

You really should not be seeing such high spikes if the FWD of the upper IGBT is working properly. And the other transition about 30 uSec after turn-off may be some additional inductive ringing or noise that is getting to the gate and causing the IGBT to turn on again.

If you can measure the actual current through the IGBT then it might be more clear as to what is going on. And it may help to put about a 1k resistor and a 15V zener from the base to emitter. If you want to keep the -5V turn off drive then you will also need a 5V zener in series.

Have you tried again with a resistive load? IIRC even that was not quite right. You should get a very clean waveform and very low VceSat. Until you do, and until you provide the correct base drive, you should not attempt an inductive load.
 
#54 ·
Ya, simple little scope that's hard to decipher: the 0V on that trace is the small purple triangle to the right. The turn on is ~ +15V and the turn off is ~ -8V.

The duty cycle is unfortunately all over the place with our pics. Apologies.

We did do the resistive load and thought it looked pretty good.

Here's a better view of the recent progression: 12V, resistive load (low current level), 30R gate resistor, old 3uF snubbers
Technology Oscilloscope Electronic device Display device


Same 12V and old snubbers but with starter motor in place of resistive load. The screwy double-turn-offs (or is it ons?) kick in. The voltage spikes are understated in this trace capture. They were worse.
Oscilloscope Electronics Technology Electronic device Display device


Same 12V and starter motor but with new 5uF snubbers. Lower spikes on turn-off, flatter off-voltage but still the screwy double-kick. And upping to 24V seriously messed it up as in the set of pics posted a couple of days ago.
Oscilloscope Electronics Technology Electronic device Display device
 
#55 ·
Leads me to think either the frequency is too high for the diode, it can't recover in time, or something is just wrong with the diode regardless. Those IGBTs are supposed to work at 300 amps 600 volts, no way they'd work like this at that power level, something is wrong

Could you try 1 ot 2 kHz or is that not adjustable with the PWM controller that you're using?
 
#56 · (Edited)
Is the first picture with the 15V signal the gate drive or the C-E waveform? The gate drive should be rock solid and should not spike or sag as it appears to. Are you sure it is DC coupled, and the probe is properly compensated?

You really should have a dual trace scope for this. And I don't understand why the PWM duty cycle (and frequency) are not constant. It's important to know if the IGBT stays fully ON during the high gate drive. And when it is off, the voltage should quickly reach the battery voltage and stay there. Is the battery voltage solid? Also if there is a large inertial load on the starter, it may act as a generator, but only if it is a PM motor and not series wound. However I have no experience with these motors and controllers so I'm just tossing ideas into the mix. ;)

Also, do you have the other components for the VLA500 IGBT driver, as shown in the following?:
http://www.pwrx.com/pwrx/docs/bg2a_application_note.pdf

I see that you are using the P&S controller, which IIRC has been modified for IGBTs. I found a schematic on their website:
http://ecomodder.com/wiki/index.php...r_Series_1000_DC_motor_controller_.28Rev2A.29

It would be helpful to see a modified schematic showing the changes that were made for the IGBTs.
 
#57 ·
The PWM duty cycle is not constant because it is a constant-current controller, so it is adjusting the PWM to get the motor current to match the throttle input. The first picture with the 15v signal is the C-E waveform in swoozle's latest post.

Unfortunately I don't have the programming know-how to change the firmware to a 2khz drive signal. I agree there is something about the diode. I don't think we are out of the safe operating area, because we are measuring the current and voltage across the IGBT and neither are near the maximum rated. The saturation voltage is definitely something to look into, as a symptom that will tell us more about our problem. Our gate drive is very slow right now because of the 30 ohm resistor. However, with the 5.6 ohm resistor, we seemed to be having the same double-switching issues.

I think the gate drive is fine but I will check again. You are right about the resistive load though, the voltage graph was surprisingly spiky for very small current and voltage.

Looking at the diode reverse recovery time, it says 300ns for 900a/us di/dt and -10v Vge. So this should switch more than fast enough for our uses. Unless it's not switching fast enough because Vge is 0v for the side of the IGBT that we're using as the diode, because G and E are directly connected?
 
#58 ·
Maybe the problem is with the current/torque mode control. An unloaded series wound motor is very difficult to control, because there will be very little current to work with. Can you put some sort of load on the motor shaft, like a fan? Can you take a scope reading on the output of the LEM current sensor? Are you monitoring the speed of the motor? And how are you setting the throttle?
 
#59 ·
For low current testing of P&S controller i took LEM off the busbar and routed a few turns of cable trough it - it reads actual current multiplied trough loops number, it makes PID more stable with low power levels; I used 12V 50W lightbulb, then 80W wiper motor and 160W radiator fan, the last one was quite easy to control trough ~3/4 of throttle range.
 
#60 ·
We have tested the output signal of the controller and of the IGBT driver and determined that the strange double-turn off is not coming from the controller! It is coming from the IGBT driver. We are not sure why the IGBT driver is doing this though. I have placed the oscilloscope on the 15v supply and it is as solid as a rock. I am thinking it my be because I disabled the short-circuit protection so I will re-enable it and see if that gets us anywhere. This is very strange behavior from the VLA500. The presence of the motor doesn't matter, it just always does this at low duty cycles, when you first twist the throttle. It even does the double-turn off with no load.
 
#61 ·
OK, we got the double-tap taken care of. It was a couple of things, one of which is a weird interaction between the netgain hall throttle and the revolt board. Whenever the revolt 5V power is connected to the throttle, the 5V line takes on a 8khz roughly-sinusoidal 4-6V pattern which is then superimposed on the throttle output. Needless to say the VLA500 doesn't like that. Couldn't figure out the root cause so I took care of it with an off-board 5V regulator to supply the throttle. Problem solved.

The one that remains and is giving us fits is this long tail on turn-off. With a 30R gate resistor the switch time should be around 2us. The FWD reverse recovery time is supposed to be .3us. And yet the tail looks like the FWD isn't working at all over 40 or 50 us. What the heck? This is only 12V and 60 or 70 amps.

The behavior is the same using another IGBT of the same model (Fuji 2MBI300L) as well as another different IGBT (2MBI200N, same as in the EMW charger). I can't help but think it must be something obvious but we aren't finding it. Ideas are very welcome.
 
#62 ·
My original controller used a pair of 1mbi800 fuji parts and worked fine even with a unipolar 12v gate drive signal. Next thing to do is start dropping down the gate resistor and see if the the spike magnitude increases with switching speed.
 
#63 ·
Next thing to do is start dropping down the gate resistor and see if the the spike magnitude increases with switching speed.
Yes, it does. I dropped back to 10R with the same setup and the spikes got worse.

The gate on the FWD is just grounded. If that switch is turning on (as has been mentioned in other threads), would it exhibit this behavior? Doesn't seem like it, but I'm barely a novice here. Would a negative applied voltage on the FWD gate help by making sure the FWD switch stays off?
 
#64 ·
Trying to analyze this... If you are getting faster, higher spikes at turn-off, that shows that the IGBT is working properly. The magnitude of the spike should not be much higher than the bus voltage if the upper IGBT as FWD is working properly. A properly designed RC snubber across the motor (or across the lower IGBT) should reduce that spike and the ringing, although it may eat some power.

But your scope trace awhile ago also shows a decaying voltage that seems to start at a voltage substantially higher than the nominal DC bus voltage, and this cannot be if the upper FWD is working. The motor connection between the two IGBTs cannot go any higher than one diode drop above B+. So, either the diode is defective, or your bus voltage is not solid. If you are isolating the B+ from the battery with a diode, then it could rise to a much higher voltage, which would then bleed off through the bus capacitors. This could also be caused if the motor was generating during the OFF cycle. I don't think a series DC motor will do this, but if your starter motor is a PM type, it may do so. In that case, you may need much larger capacitors from B+ to B-. And you should be able to see this effect by scoping the B+ line.
 
#65 ·
I have not heard of anyone putting snubbers across the load or freewheel diodes. Are you sure that this is a common practice? I think that the next step may be to attach some external diodes directly to the motor and try to stamp out the spike at the source.
 
#66 ·
The snubber is used to absorb the transient that happens when current through an inductance is interrupted. Once the freewheeling diode conducts, it is no longer needed, but without an RC snubber there can be a very high, fast spike that can damage the IGBTs. I found a lot of information and posted some links on Stiive's thread:
http://www.diyelectriccar.com/forums/showthread.php/3-phase-dtc-svm-induction-motor-74151p18.html

Particularly useful might be the following:
http://www.cde.com/tech/design.pdf

It does show snubbers across the DC voltage supply but only when there is series inductance from the original source (batteries). Even lengths of wire have inductance, and the insulation of the windings of the motor have distributed capacitance which can "ring", which is shown by the high frequency damped oscillations. If you do not have a really good scope, it is possible that you are not seeing all of the transients, and what you don't see can damage the semiconductors or degrade the insulation.

But the main problem appears to be the decaying voltage with a rather long time constant that occurs at turn-off. This indicates a much larger source of stored energy, which could be a large inductance, or, more likely, the rotational inertia of the motor armature which is acting as a generator. However, if the voltage as shown on the scope is more than a couple volts above the B+ voltage, either the FWD is open, or there is little capacitance on the B+ line. That is easy to check with a scope.
 
#67 ·
How to check B+ line capacitance? And yes we seem to be sure that the FWD is simply not working. So I believe the next step is to put an oscilloscope across the FWD and see what is going on with that, and if it's not working, put some external diodes directly across the motor.
 
#68 ·
The first thing I would do is put the scope from B- to B+, right at the IGBTs. It should be just about rock solid, with maybe just one or two volts of ripple depending on the drive for the motor. If you see variations similar to what you see when the scope is from the motor (-) to B-, then the capacitance is too small. How much capacitance do you have? IMHO it should be at least 1000uF, and of course it needs to be rated for the maximum B+ voltage.

If you want to measure the actual capacitance, you can just connect the batteries to charge it up to the pack voltage. Then you can disconnect the batteries and put a resistor across the capacitor (it should have a bleeder anyway), and measure the time for the voltage to drop to 37% of the full charge. This is the RC time constant:
http://en.wikipedia.org/wiki/RC_time_constant

So since T=RC, C=T/R. If you have 100 volts and with a 10,000 ohm resistor it drops to 37 volts in 10 seconds, the capacitance is 1,000 uF.
 
#69 ·
I remember now that we have done a measurement of B+ to B-. We only did it once and I don't remember under what circumstances. I do remember that it was very wavy, with a sinusoidal thing going on starting at each turn-off.

The thing with the increased capacitance is - we tried a 400v 2200uf cap and it didn't damp much better than our 380uf 400v cap. Installing a small snubber close to the IGBT terminal made more of a difference than using a larger main bus capacitor.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top