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

[peters]

Ok I see. Just to clarify, by VBE they mean one diode drop, that is the base-emitter voltage of the internal high side transistors, and there are 3 of them, normally 3*0.6V (but it depends on the temperature).
But later in the datasheet in the electrical specifications table at the "High Level Output Voltage" they say it is VCC-2.9V for the minimum at 650uA, so this is the lowest.

 

[Arlo1]

I have started working on my next controller design this one more or less follows all design rules and should work well. Hoping to be testing desat next week.
https://endless-sphere.com/forums/viewtopic.php?f=30&t=76025

 

[Arlo1]

Ok in testing on Game changer today I found the clamp was not doing a lot in the configuration I had it. So I tried a jumper from the clamp output of the driver to the gate and skipped the boost transistor for the clamp and it worked better. I suspected something with the clamp on the leaf inverter and I think I just found it. More testing to come.

 

[Arlo1]

For testing with game changer. I found I maxed out my 950 amp current sensor at 1026 amps. So to double check. I pulled one of the current sensors out of the leaf inverter as I had new revised boards to use to build up so I needed to pull the sensors out anyways. As I pulled the current sensor out of the middle board the pins came off the hall effect chip so that might have been a reason I had such a rough running system under load and might have been part of my failures.


I added Shottky diodes to all the input and output signals to try to keep things in check. The most important Shottky diode is the one from the gate to the gate 15+ supply positive! That will replace the double zener in series with a backwards normal diode I used.

Moving forward....

 

[Arlo1]

Ok I built up the new power stage boards for the leaf inverter. They seem like all the little bugs are worked out except I have 1 problem left.

When I run a double pulse test with any amount of current ~200amp or more the opposite side of the phase latches on. I have the input tied to 0v to keep it off and the input of the driver is commanded off but once I run a couple pulses thought the high side of the phase the low side latches on and its being commanded on from the driver chip it self. Its a ACPL333J I have no idea what's causing this. I disconnected the Traction battery and once it was powered down I took the cap off to have a look around while keeping the low voltage supplies powered up and found the diver in question is super hot.. 150 degC + I didn't have the thermal cam and my temp gun could not see it. So I measured it and on the input it was at 0v but the output was 11.8v.... So Its a problem internally being tripped by the freewheeling currents in the diode of the IGBT/or a voltage spike/swing.

I used 2 differential probes to see this in real time and watched both high side and low side gates.....
This is a problem I have had intermittently since day 1 with this driver so I think it must be something I missed in the data sheet or something I have never seen...

Once I get this 1 last problem I think its go time.

 

[Lebowski]

I'm not a big fan of these drivers... too expensive, need 6 of them, am not sure but i think you cannot use a simple bootstrap supply. For my kA project I am again going for the simple 2186's. One thing that these are good for is that they are tolerant to, if i remember correctly, the output being pulled negative due to inductance in the power circuitry. Have a look at the application noter from International Rectifier regarding this... Oh, and i will have desat detection in my kA project, but built differently, none of this 250uA in a cap etc but plain old digital

 

[Arlo1]

I'm not a big fan of these drivers... too expensive, need 6 of them, am not sure but i think you cannot use a simple bootstrap supply. For my kA project I am again going for the simple 2186's. One thing that these are good for is that they are tolerant to, if i remember correctly, the output being pulled negative due to inductance in the power circuitry.

Your choice.


I added clamp shottky diodes to the outputs and the inputs to protect the driver from this.

I will try to double check it to make sure its correct.

 

[HighHopes]

your topology is correct for the power level you are at. the opto-isolation is absolutely imperative, bootstrap is not appropriate at all. at power levels <5kW probably bootsrap can offer some cost advantage if you willing to live with reduced performance (i am doing this for my low power DC drive), but at your leaf inverter power level you have to do it the way you are doing it. except for the bug you have. ha. only way it can get so hot like that is to have a short-circuit. that's hard to do in the typical gate drive because when gate drive IC turns ON it charges the igbt gate cap and then the current drops to zero.

is your pull down gate/emitter resistor actually 4.7k (or there abouts)? you should be able to measure this with ohm meter. do you have external gate/emitter cap? if so, is the cap ok? is the soldering good for both or did you accidently short out the pads?

or maybe your igbt is damaged on the gate somehow and it no longer acts like a capactior.. easy enough to test, remove the igbt and put on an external cap of about equivalent to the igbt gate cap. apply some switching to input of gate drive IC and see if it acts normal. if it does, then next is to investigate the switching waveform and try to figure out how it is the gate got destroyed.

if it still has a problem then the gate drive IC has died some how. here it is hard to die, my guess would be a problem with the layout (traces too close, or not respecting creapage) or solder blob. its possible that switching your IGBT at 20kHz is too much load for this gate drive IC and it suffered thermal failure. easy enough to verify because if this is the case then the other gate drive ICs should be really hot too. its possible that your gate resistor value is too low which causes you to ask too high peak current from gate drive IC but i doubt you have this problem as you already how to size gate drive components.

if none of the tests reveal a problem then it is something intermittent that reveals itself under operation. here it is most likely something to do with reference of the gate drive IC (either the input or output) where reference is drooping low causing the "low" input signal to suddenly appear like a high voltage. like if your low signal was 0.1V with respect to ground/reference then the voltage difference is 0.1V and this is OFF signal. now keep the pwm signal the same at 0.1 but suddenly lower the reference on a negative going transient by 3.2V and now the difference between pwm and reference is 3.3V, that's an ON.. even though your PWM command is only 0.1V, at the gate driver with its crappy reference it looks suddenly like an ON. this sort of thing can happen on the input side or output side.

 

[HighHopes]

hmmm.. i should have read your pots a little better. i read it again, it sounds like you are having a lot of miller effect happening. at your power level you will expect some, like 4V worth, but this is normal. its why big IGBTs like leaf inverters should be turned OFF at -8V. so that when the adjacent mosfet turns ON causing miller effect of the other, you have LOTS of margin to absorb a transient. when you turn OFF at -8V you don't need to use the clamp feature of the gate drive IC. but .. having said that.. if it is miller effect, which it sounds like, the result is that the IGBT turns ON when it should be OFF (shoot-through might happen) the result is not really that the gate drive IC heats up and fails.. so.. hmmm...

can you post your schematic? maybe i can spot a problem.

 

[Arlo1]

hmmm.. i should have read your pots a little better. i read it again, it sounds like you are having a lot of miller effect happening. at your power level you will expect some, like 4V worth, but this is normal. its why big IGBTs like leaf inverters should be turned OFF at -8V. so that when the adjacent mosfet turns ON causing miller effect of the other, you have LOTS of margin to absorb a transient. when you turn OFF at -8V you don't need to use the clamp feature of the gate drive IC. but .. having said that.. if it is miller effect, which it sounds like, the result is that the IGBT turns ON when it should be OFF (shoot-through might happen) the result is not really that the gate drive IC heats up and fails.. so.. hmmm...

can you post your schematic? maybe i can spot a problem.

Its not miller effect.

I hooked a diff probe to the high and low gates at the same time. The low stays tight. I have some scope shots saved I will post tomorrow.
I am using -5 with a clamp. I will post the current sch in a bit.

I read though your first response and I don't think its any of those things.
I have had this problem with all of the drivers and I used new drivers on each build and this is the 3rd set of revised driver boards.
This only happens on the output. I scoped the input and its fine. (at 0v)
Once it happens the output stays on with all input signals at 0v.
It only happens when pulsing current.

Thanks for all you help. And support. Its always exciting to have Lebowski to tell me I'm doing it all wrong.

If you haven't seen I have over 1ka in testing with game changer and this driver has proven it self in other builds but I have not tested for this problem on game changer yet.

 

[Arlo1]

I'm not a big fan of these drivers... too expensive, need 6 of them,

BTW why don't you price out the fets and caps alone for what you are planning for over 1ka and tell me how much you want to cheep out on drivers that might save the expensive parts.

 

[HighHopes]

agree, its not anything of the things i listed. im trying to figure this out.. thinking out loud.

can you remove your clamp circuit (R5, Q5, D6) and just let your driver turn IGBT off at-5V? does this help?

R3, R1 are shown as optional. are they populated?
C15 is listed as DNP. did u populate it?

R19, try increase this to 1k ohm. this resistor limits the transient amps that will be pulled out of gate driver IC, this could actually be something related to your problem. you may have to change value of C6 to get same time constant.

your turn ON caps C29, C35 this is good. your turn OFF cap, why you have 3x4.7 instead of just two to keep symmetrical? C3 looks like unnecessary extra. don't think this means anything, but it is something that bugged me.

Schottky 5/6 are located in a strange spot. did you really have voltage transients at Vout??

R18 seems high, you sure 10k is acceptable?

this is for leaf inverter, driving IGBT no? this should be turned ON at 15V, not 12V (12 is for mosfets).

 

[zombiess]

Your mysterious glitch is probably ground bounce at turn on/off as it can cause spikes. This is one of the bumps I haven't fully explained the mechanism of yet.

Physically short the high side G-E together. Place one lead of the differential probe at this point and the other lead on the phase output. Repeat your double pulse test and you'll most likely see the ground bouncing around. This is a current driven phenomenon which is difficult to isolate without using a rogowski coil.

Top graph shows the differential between the phase output and the high side which I have shorted.

*edit* Now I'm thinking that the cause of these spikes is commutation of current through the leg inductance after looking at the schematic I posted. Seems kinda obvious. Most of the time these spike are really short, but large ringing at turn off can cause the ground reference to get pretty wavy.

Anyways, it's just one more thing you can check out since I haven't seen anyone mention it yet.

 

[Arlo1]

agree, its not anything of the things i listed. im trying to figure this out.. thinking out loud.

can you remove your clamp circuit (R5, Q5, D6) and just let your driver turn IGBT off at-5V? does this help? This is a good thing for you to point out.
IN this current build I jumped the clamp Q5 to let the driver hold the igbt off better and It seems worse. So maybe removing the clamp will make it better. I will try this.

R3, R1 are shown as optional. are they populated? Yes this is needed so the driver can see the voltage and know when to activate the clamp I think
C15 is listed as DNP. did u populate it? Yes I have a 100nf there

R19, try increase this to 1k ohm. this resistor limits the transient amps that will be pulled out of gate driver IC, this could actually be something related to your problem. you may have to change value of C6 to get same time constant. OH snap thanks for pointing that out. That's a good one I never thought of...

your turn ON caps C29, C35 this is good. your turn OFF cap, why you have 3x4.7 instead of just two to keep symmetrical? C3 looks like unnecessary extra. don't think this means anything, but it is something that bugged me. 7C was worried about a cap on the clamp fet so C3 goes next to the clamp fet

Schottky 5/6 are located in a strange spot. did you really have voltage transients at Vout?? I have 6 schottkies per driver two protecting the signal in 2 protecting the signal out and 2 protecting the signal to the gate of the igbt

R18 seems high, you sure 10k is acceptable? I will edit this sorry it is a 4.7k I just forgot to edit the sch

this is for leaf inverter, driving IGBT no? this should be turned ON at 15V, not 12V (12 is for mosfets).

Yes I am using 15v for the on and -5.6 for the off.
I will edit this as well.

 

[Arlo1]

Your mysterious glitch is probably ground bounce at turn on/off as it can cause spikes. This is one of the bumps I haven't fully explained the mechanism of yet.

Physically short the high side G-E together. Place one lead of the differential probe at this point and the other lead on the phase output. Repeat your double pulse test and you'll most likely see the ground bouncing around. This is a current driven phenomenon which is difficult to isolate without using a rogowski coil.

Top graph shows the differential between the phase output and the high side which I have shorted.

*edit* Now I'm thinking that the cause of these spikes is commutation of current through the leg inductance after looking at the schematic I posted. Seems kinda obvious. Most of the time these spike are really short, but large ringing at turn off can cause the ground reference to get pretty wavy.

Anyways, it's just one more thing you can check out since I haven't seen anyone mention it yet.

Thanks for the response. The ground bounce is something I was looking at and that's why each driver has 6 shottky diodes on it. 3 are to stop the negative form going below the -5.6v rail from bouncing away form the pin on the driver. 3 are to stop the positives from going over the 15v rail.
But it is something I have been looking at. And the only way I can think to make that better is to slow the turn off. Now having said that. I tried to make things better with this build up of power stage boards and made it worse. I will try HHs suggestions then I will come back to this.

On the last build I had it working decent most of the time then I would get a random desat every once in a while. Now I get it all the time. The layout has not changed inductance wise and the switching is still the same with the same gate resistors. So my guess is its not caused by ground bounce as much as current being pulled thought the opposite driver from the one being pulsed. I'm lucky I made it worse because now I need to debug it and learn form it and make sure its 100% before driving the car.


You keep mentioning a rogowski coil. Where did you find yours? I just got a new scope and a second diff probe. Got a DS1054Z and hacked it to the DS1104Z and got a 25mhz and a 65mzh diff probe. I will get another Diff probe soon. But With this particular problem I wanted to see more then 2 things at a time.
I can use a non diff probe for the current or something not sensitive. Also the new scope has a much higher sample frequency then the old one.
Its the best bank for buck for <$1100 us.
As for the rogowski coil its for measuring current right? How big is it? I am not sure I can use one on this and current in a circuit is the same all though the circuit.

 

[zombiess]

I don't think you can stop the cause of the issue I'm referring to with circuitry. It's not that the ground goes below -5.6V, but rather the -5.6v becomes +4V for a very breif period causing turn on. I've observed these spikes even when a clamped negative or shorted. It's a fight against layout and device inductance.

Using a spice program is really helpful to troubleshoot these issues. It's just difficult to build accurate model.

I purchased my coils from Athena energy. I think they were around $800ea. Mine read up to 200A and fit through the legs of a to247 and could be squeezed in a to220.

There is also a learning curve in using the coils accurately.

 

[Arlo1]

I don't think you can stop the cause of the issue I'm referring to with circuitry. It's not that the ground goes below -5.6V, but rather the -5.6v becomes +4V for a very breif period causing turn on. I've observed these spikes even when a clamped negative or shorted. It's a fight against layout and device inductance.

I did some deep digging a couple months ago and the shottky diodes are supposed to help this. If the -5.6 becomes +4 the diode will conduct all of the power back to the -5.6 trying to hold it at -5.6 (+ the low voltage drop of the shottky). Non the less if the problem was ground bounce and I added shottky diodes to help it and it got worse then its not likely the problem. But It can be R100 on the desat. I need to look at that and the other things HH suggested first then come back to this. I will try to set up a probe measuring the -5.6 at the driver and at the cap at the same time if needed.

Non the less Will it turn on and stay latched on ??? What is happening to my low side driver (when high side is pulsed) is the low side driver stays off then after something like a few seconds after a high side pulse with some current the low side randomly turns it self on and gets hot. (something internal).

Its not a faulty driver from what I can tell because its happened many times on different drivers.

 

[HighHopes]

can you snap a picture of your setup?

 

[Arlo1]

can you snap a picture of your setup?

I will get pics/video soon. Of the whole layout.

Its all at my shop but here is a screen shot of the current driver board.

 

[HighHopes]

yes, i remember pouring over that for hours... could have been teh prevoius one you said that was the same, but anyway. when i see it now, 2 things jumped out at me:

1. the ground pour looks too close to your lug connection. when you put a screw and/or washer on this lug does it touch the ground pour and maybe scratch away the solder mask as well?

2. the component looks very close to the lug..

 

[Arlo1]

So with 2 things to look at I chose the R100 as the most logical first choice to experiment with.

I changed R100 to 1k ohm like HH suggested and I run multiple double pulse tests with no latch up.


I feel this is solved.


I looked at the rise times and its seems to be the same. I will look harder as this and I will verify desat works like it needs to before trying to run the motor.

But thank you HH I would have overlooked R100 and been scratching my head for a long time.

I did think it was something to do with the freewheeling currents and how that effects the driver. A 500 ohm resistor or something might work as well but I think it will be ok to leave it at 1000ohms.
From the data sheet this is all it said.

DESAT Pin Protection Resistor
The freewheeling of flyback diodes connected across the IGBTs can have large instantaneous forward voltage transients which greatly exceed the nominal forward voltage of the diode. This may result in a large negative voltage spike on the DESAT pin which will draw substantial current out of the driver if protection is not used. To limit this current to levels that will not damage the driver IC, a 100 ohm resistor should be inserted in series with the DESAT diode. The added resistance will not alter the DESAT threshold or the DESAT blanking time.

The nominal blanking time is calculated in terms of external capacitance (CBLANK), FAULT threshold voltage (VDESAT), and DESAT charge current (ICHG) as tBLANK = CBLANK x VDESAT / ICHG. The nominal blanking time with the recommended 100pF capacitor is 100pF * 6.5 V / 240 μA = 2.7 μsec.

If there is 6.5v applied to a 1000 ohm resistor my math says it should allow up to .0065a or 6500uA so it should be ok with a 1k resistor and my scope showed about the same reaction time.

[youtube]yhbLNOL3N30[/youtube]

 

[Arlo1]

yes, i remember pouring over that for hours... could have been teh prevoius one you said that was the same, but anyway. when i see it now, 2 things jumped out at me:

1. the ground pour looks too close to your lug connection. when you put a screw and/or washer on this lug does it touch the ground pour and maybe scratch away the solder mask as well?

2. the component looks very close to the lug..

I think these are ok I checked again and the clearance is greater then it looks on the file.

I have been back and forth on the clearance but I'm trying to keep it tight but safe lol

 

[HighHopes]

huh.. honestly it seemed you were having ground bounce problem. i just through that R100 out there cause it crossed my mind and i like to think outloud when i review this stuff cause you never know what someone will pick up on. seems that practice worked out this time

ps. you owe me a BEvERage

 

[Arlo1]

ps. you owe me a BEvERage

Pony keg or a regular size?

When you come out to the west coast I will hook you up.

 

[Arlo1]

huh.. honestly it seemed you were having ground bounce problem. i just through that R100 out there cause it crossed my mind and i like to think outloud when i review this stuff cause you never know what someone will pick up on. seems that practice worked out this time

This makes sense though because I have a bit more inductance then I want (big components and trying to retrofit them to a pre-existing heat sink) and I am switching pretty fast.

SO if there was less current or less inductance or slower switching it would have been ok with a lower R100 value.
I will run it soon and make sure all is ok. But so far so good. I am 99.99999% sure this was my problem the whole time and this has randomly plagued me since day 1 about 2 years ago. The last build just made things switch a touch faster with the new layout and it became more apparent and less random.

I did 40+ double pulse tests yesterday and so far so good. Before changing R100 to R1000 it would latch up after about 1 double pulse under load.

I will try to post screen shots when I figure out how to get the new software for the scope to work.