Arlo's power stage Leaf controller runs and drives page 103

[Arlo1]

Ok.... I will try to see what else I think can be causing it.... Mean while it was SjwNz
who found this datasheet
http://www.avagotech.com/docs/AV02-0258EN
Its extra parts that need to be added, but looks like it adds the delay by keeping the Blanking cap discharged for XXuS and then releases the Cap for it to charge.

Hmmmm I have scoped the gate a few times and never saw anything enough to cause this but... The one other thing I added was the jumper from the actual emitter to the small emitter connection on the driver board. I wonder if some how that is giving me issues.

 

[Arlo1]

I have spent a lot of time searching but can't find data on common causes for IGBTs to fail shorted. Anyone have any data on that as what I've been told is IGBTs more often fail open. This is with 227v on the cap... with a 1200v rated part.

 

[HighHopes]

here's a thought.. with 4us transition time, that's a lot of time spent in the linear region. do u think you over heated?

 

[Arlo1]

here's a thought.. with 4us transition time, that's a lot of time spent in the linear region. do u think you over heated?

No because that is at high current and the failure was at low current but likely fast transiants

 

[HighHopes]

operating at low amps or high amps makes no difference to shoot-through and desat, as soon as shoot-through happens it is instantly extremely high amps. since you already have confidence that desat works by deouble pulse test results, then likelyhood that desat failed is unlikely. possible yes, but unlikely. desat tends to be highly robust by nature, course it needs to be given the environment it was designed to work in.

so what is special about low amp operation? two things really, there is virtually back-emf because presumably you are operating at low speed. so you put full voltage across the the AC output. there is no "subtracting" of this voltage due to existing bEMF as bEMF is about 0. this happens every time at start-up too, so its normal. anyway, you end up switching high voltage at low or zero motor rpm. high dV/dt, at its maximum which means noise. a level of noise that does not exist at high amp or high motor rpm. the other thign is very short PWM pulses. too short PWM. the opposite can happen at very high torque, too long PWM. there should be software to know what your max/min PWM pulses should be and limit it to that. so its not possible to drive at very veyr low RPM, sometimes, because software will not allow such small PWM pulses that can over heat a modeule; this is sometihing you won't likely find in the text books

 

[zombiess]

there should be software to know what your max/min PWM pulses should be and limit it to that. so its not possible to drive at very veyr low RPM, sometimes, because software will not allow such small PWM pulses that can over heat a modeule; this is sometihing you won't likely find in the text books

A little tangent off shoot here.

HighHopes, Where do you typically lock out short/runt pulses? I'm guessing it's somewhere around the G-E rise time + 10-20% since the pulse needs to be able to get past the Miller Plateau in order to reach saturation and not heat the module.

Back on topic:
Is this module subject to a latch up failure mode? Not sure what generation this IGBT Arlo1 has, but I've read that the parasitic NPN/PNP/Thrysistor comes into play when the gains > 1 due to heating. Is it possible that a dynamic latch up scenario can occur due to the high dV/dt at turn off?

This is an article I have previously read on the topic which came to mind.
http://www.eetimes.com/document.asp?doc_id=1273173&page_number=1

 

[HighHopes]

HighHopes, Where do you typically lock out short/runt pulses?

ya i probably shouldn't have mentioned this. its more of an advanced topic item because it is not critical to get something to work but it is important for developing a product. and talking such small details will distract from the good work being done on the bulk scale by u and arlo in the last year. anywho.. i design for 50/50 or 60/40 thermal split between switching and conduction loss at FLA. then i pick max/min PWM for the reason you said. i calculate what that should be though, basically i want tthe module to switch ON and that amount of time it takes to turn ON is the same amount of time that the minimum conduction period is. so minum PWM then is turn ON + conduction min, at FLA. you can argue pro/con on this method or to consider this fact (like there is also switch OFF that happens in same period) but it has been a good balance between protecting against a failure and not making min PWM too big, its a way that has worked for me. the point is at least you think about this type of failure mode, when you're drive is clipping (PID's output is saturated) and issuing many cycles of unusually max or min Pwm and now you have some way to protect against a failure that can happen in this strange scenario. saturate PID, i know you know what this means cause we talked about it. question that might come up though is why would anyone expect PID to saturate in the first place? just remember, we are talking about a high performance drive, not an industrial drive. its a high bandwidth system with large step response and small signal stability that is stressed to large amount every single moment its in operation. PID is likely saturate sometimes; well, sh#t happens

latchup... i hope not. that was designed out by IGBT manufacuters 10+ years ago and fuji was one of the first to do that

 

[zombiess]

It may be advanced, but it's worth a quick note as some of his screen shots showed quick succession of on/off pulses at turn off which could force commutation of ever increasing high current events. Full Load Amps + a stream of short pulses could = high Tj.

I know exactly what you mean about the PID :wink: Best not to derail this thread.

I'm not up to speed on the history of IGBTs at all, but I have read that latchup can be an issue on some devices (mainly older ones). Thought it was worth tossing it out there.

 

[Arlo1]

It may be advanced, but it's worth a quick note as some of his screen shots showed quick succession of on/off pulses at turn off which could force commutation of ever increasing high current events. Full Load Amps + a stream of short pulses could = high Tj.

That was me turning up the first pulse on the desat setup until It triggered Desat. On the double pulse tester I did not have it shut anything down as it was 1 double pulse test at a time.

When its running in the controller its shut down on the first desat event.

 

[Arlo1]

Ok... I manned up and shorted out the phase to positive then tested the low side switching into a dead short with 456v applied to it...
I hit a max in my tests of 1315 amps in 4.4uS before desat saved the day... DAMN near 600kw into a short for a split second.... All looks well so far and I will see what else I can test/debug asap.

[youtube]KvqZvjyaCDM[/youtube]

 

[HighHopes]

i'm glad u have a good handle on desat protection, its the one that will prevent a fire. but for the record, you're crazy

 

[Arlo1]

but for the record, you're crazy

Weird... I have herd that before.

 

[Lebowski]

all this bench testing is fun and dandy, but when are you going to put it in the car and do dyno testing ?

 

[Arlo1]

all this bench testing is fun and dandy, but when are you going to put it in the car and do dyno testing ?

I am just trying to debug the last problem. I had something kill a couple IGBTs And well... I just sent you some chips for v2.5 with field weakening

 

[Arlo1]

Ok. I blew up 2 more igbts. Today. The desat works during testing but when I try to run the motor during sensorless startup it almost always pops an IGBT. Its all over in less then a couple seconds I'm only running it on 115v for now and having 0 luck. What should I do to debug this without risking any more blown igbts. The igbts are always failing shorted.

 

[jmz]

Ok. I blew up 2 more igbts. Today. The desat works during testing but when I try to run the motor during sensorless startup it almost always pops an IGBT. Its all over in less then a couple seconds I'm only running it on 115v for now and having 0 luck. What should I do to debug this without risking any more blown igbts. The igbts are always failing shorted.

Test with a massive current limiting resistor on the DC bus? (and at low power!)

And then instrument the shit out of it, gates, Vce's, leg currents...

 

[Arlo1]

Ok. I blew up 2 more igbts. Today. The desat works during testing but when I try to run the motor during sensorless startup it almost always pops an IGBT. Its all over in less then a couple seconds I'm only running it on 115v for now and having 0 luck. What should I do to debug this without risking any more blown igbts. The igbts are always failing shorted.

Test with a massive current limiting resistor on the DC bus? (and at low power!)

And then instrument the shit out of it, gates, Vce's, leg currents...

Although I use a 100 amp DC breaker with a 450 amp fuse on the DC side
Limitiing the DC buss is not an option to save igbts. 1 amp at 115v DC can become 1000 amps on the AC side.

 

[jmz]

Sorry, I don't quite follow you there.

What is the source of the fault current? I meant putting a resistor right between the IGBTs and the main DC bus.

You are producing >1 kA on the AC side while only drawing 1 A from the bus?

 

[Arlo1]

Sorry, I don't quite follow you there.

What is the source of the fault current? I meant putting a resistor right between the IGBTs and the main DC bus.

You are producing >1 kA on the AC side while only drawing 1 A from the bus?

Yes google buck converter.

you can turn just .001 amps from the battery into 10000 amps on the ac side in the right setup
What I am getting at is any sort of current limit between the igbt and the DC buss will not work that is considered DC current . The only place some sort of current limit will work is on the AC side. Which is from the IGBT to the motor phases but putting something there is very dangerous.

Now this is the big part we need to look at where the problem lies. It might not be a current problem on the power legs of the igbts. It could be a voltage spike on the gates....
I need to look at ways to prove this safely and fix it.

 

[Arlo1]

So I found this link. http://www.eetimes.com/document.asp?doc_id=1273173&page_number=2 While reading up on "latchup"

The gains of the PNP and NPN transistors are set so that their sum is less than one. As temperature increases, the PNP and NPN gains increase, as well as the body region spreading resistance. Very high collector current can cause sufficient voltage drop across the body region to turn on the parasitic NPN transistor, and excessive localized heating of the die increases the parasitic transistor gains so their sum exceeds one. If this happens, the parasitic thyristor latches on, and the IGBT cannot be turned off by the gate and may be destroyed due to over-current heating. This is static latchup. High dv/dt during turn-off combined with excessive collector current can also effectively increase gains and turn on the parasitic NPN transistor. This is dynamic latchup, which is actually what limits the safe operating area since it can happen at a much lower collector current than static latchup, and it depends on the turn-off dv/dt. By staying within the maximum current and safe operating area ratings, static and dynamic latchup are avoided regardless of turn-off dv/dt. Note that turn-on and turn-off dv/dt, overshoot, and ringing can be set by an external gate resistor (as well as by stray inductance in the circuit layout).

Also found this. http://www.fujielectric.com/company/tech_archives/pdf/36-04/FER-36-04-166-1990.pdf
It talks about a ratio that want to keep at .7 npnnp and if it gets to 1 it will latch on. I am trying to understand this.
Here as well. http://www.microsemi.com/document-portal/doc_view/14696-igbt-tutorial

The gains of the PNP and NPN transistors are set so
that their sum is less than one. As temperature
increases, the PNP and NPN gains increase, as well as
the body region spreading resistance. Very high
collector current can cause sufficient voltage drop
across the body region to turn on the parasitic NPN
transistor, and excessive localized heating of the die
increases the parasitic transistor gains so their sum
exceeds one. If this happens, the parasitic thyristor
latches on, and the IGBT cannot be turned off by the
gate and may be destroyed due to over-current heating.
This is static latchup. High dv/dt during turn-off
combined with excessive collector current can also
effectively increase gains and turn on the parasitic
NPN transistor. This is dynamic latchup, which is
actually what limits the safe operating area since it can
happen at a much lower collector current than static
latchup, and it depends on the turn-off dv/dt. By
staying within the maximum current and safe operating
area ratings, static and dynamic latchup are avoided
regardless of turn-off dv/dt. Note that turn-on and
turn-off dv/dt, overshoot, and ringing can be set by an
external gate resistor (as well as by stray inductance in
the circuit layout).

Now I can say this, what's happening to me is in the first couple seconds of trying to run the motor at low current but the current spikes might be high as the big heavy motor is trying to shake back and fourth and doesn't know which way to spin. Its defiantly not heat causing it.

 

[Lebowski]

At the end of this old post is an explanation of what I think you should pay careful attention to:

https://endless-sphere.com/forums/viewtopic.php?f=30&t=55641&p=875398&hilit=.pdf#p875398

 

[Arlo1]

At the end of this old post is an explanation of what I think you should pay careful attention to:

https://endless-sphere.com/forums/viewtopic.php?f=30&t=55641&p=875398&hilit=.pdf#p875398

I know and until I prove you wrong you will tell me over and over the circuit inductance is causing it as you have for the last year.
The thing is the double pulse tests proves you wrong. As I can do 1000s of double pulse tests into dead shorts with 460v and 1200-1300a flowing though the igbt with no failures!
Something else Is going on because when I try to run the motor and only ONLY then I get failures. My thoughts are the motor is bouncing around and the controller doesn't know which phase to fire as its when I'm trying to calibrate things like the hall sensors but haven't got it spinning right yet.

 

[Lebowski]

At the end of this old post is an explanation of what I think you should pay careful attention to:

https://endless-sphere.com/forums/viewtopic.php?f=30&t=55641&p=875398&hilit=.pdf#p875398

I know and until I prove you wrong you will tell me over and over the circuit inductance is causing it as you have for the last year.
The thing is the double pulse tests proves you wrong. As I can do 1000s of double pulse tests into dead shorts with 460v and 1200-1300a flowing though the igbt with no failures!
Something else Is going on because when I try to run the motor and only ONLY then I get failures. My thoughts are the motor is bouncing around and the controller doesn't know which phase to fire as its when I'm trying to calibrate things like the hall sensors but haven't got it spinning right yet.

Ehm, no disrespect meant or anything, all in good humor etc, but, eh, so far you're doing a good job of proving me right....

A double pulse test is not the same as a continuous 5 kHz pulse train.

If you want to make it easier to run sensorless (as you only want this to calibrate the halls), what you can to is spin the
motor by hand before giving throttle. The controller will then get in sync with the spinning motor before you apply throttle (phase current)

 

[Arlo1]

latchup... i hope not. that was designed out by IGBT manufacuters 10+ years ago and fuji was one of the first to do that

hmmm. The 1mbi600u4b-120 and the 1mbi800u4b-120 are both GEN 5 parts so they should have the latchup problems designed out by this point. I think I read Gen 1 was the only Gen to have latchup problems.
I am trying to work though this so first question is... Is there no way to cause a latchup? Second question is what can cause a IGBT to permanently stay shorted between Collector and Emitter?

 

[Arlo1]

Ehm, no disrespect meant or anything, all in good humor etc, but, eh, so far you're doing a good job of proving me right....

A double pulse test is not the same as a continuous 5 kHz pulse train.
I get that but I can write code for a continuous 5khz pulse if you want and prove you wrong that way does that work for you?

If you want to make it easier to run sensorless (as you only want this to calibrate the halls), what you can to is spin the
motor by hand before giving throttle. The controller will then get in sync with the spinning motor before you apply throttle (phase current)

the motor is a SOB to spin by hand.... Maybe I can set the current limits low in Drive 2? What about this... I did discover with your chip there is a lot of times when the igbts get very short on and off pulses... Can we add limits for the shortest on and longest on pulse and shortest off pulse?