How to measure IGBT Losses? Thermal Issue?

[digi001]

I may have a potential thermal issue with 1 IGBT in a 6 pack configuration for 3 phase AC inverter. The IGBT has a self-shutdown thermal protection and it looks like it is faulting the device into shutdown temporarily. Not sure if its from Conduction losses or switching losses?

To measure losses of this device, can I just monitor voltage across Collector to Emiter (Forward voltage drop) and the current through the device?

If i multiply these two (Voltage and Current) I should have watts. How does this plot of watts relate to average power, or thermal shutdown of the device?

 

[FvM]

If i multiply these two (Voltage and Current) I should have watts. How does this plot of watts relate to average power, or thermal shutdown of the device?

total_losses = conduction_watts*duty_cycle + switching_energy_loss*frequency
temperature_rise = total_losses*thermal_resistance (junction-to-ambient)

Conduction losses can be read from the data sheet (if you know the current) and are the part of the calculation, that can be most easily derived.

 

[digi001]

Thanks,

HOWEVER, In this system the switching frequency is variable. It changes to spread the spectrum of switching components for EMI.

 

[mtwieg]

I may have a potential thermal issue with 1 IGBT in a 6 pack configuration for 3 phase AC inverter. The IGBT has a self-shutdown thermal protection and it looks like it is faulting the device into shutdown temporarily. Not sure if its from Conduction losses or switching losses?

How does an IGBT have thermal shutdown...? Unless you mean it's built into the control circuit. Does the 6 pack of an integrated NTC?

You can't really get a decent measurement of switching losses. I know people in industry who are able to actually plot current and voltage through each IGBT with enough bandwidth and memory to do a direct measurement, but it takes a ton of specialized equipment which I'm guessing you don't have.

First I would try a spreadsheet method where you look at one full cycle (that is a motor rotation cycle, not a PWM period) for one IGBT. Actually you should only need 90 degrees of a cycle, not sure for a 3phase inverter. Break it up into a bunch of discrete time points (you don't need one for every PWM period). You should know what the average leg current for that IGBT at each point, so you should know the turn on and turn off energy for the IGBT (if you read the datasheet closely), and the energy lost each PWM cycle due to conduction losses (since you also know duty cycle). Multiply all of these by your average PWM frequency and that should give you your average power loss.

This method sounds tedious, but was extremely useful to me once when I was designing a high power PFC supply. It turned out that power dissipation was about twice what I expected due to large recovery currents from my catch diodes. Also I realized that the thermal impedance of my IGBTs and thermal pads was higher than expected.

I've attached the last spreadsheet I did, which was with a f450r12ke3 IGBT six pack (but we didn't use it as a 3phase inverter, rather we were making an interleaved PFC boost converter, so you'll need to revise it somewhat). You can see how we did rough linear curve fits for the switching and conduction losses, taken from the device datasheets. Sheet 2 has the actual table which calculates the losses at each time step over 90 degrees of an AC cycle. Also includes estimated thermal impedance to give an estimated temperature rise.

If after doing the spreadsheet you're convinced that you're still getting more dissipation or temperature rise than expected, I would then take a scope to the system and see whether you're getting rise and fall times that match up with the IGBT's datasheet. Also consider that you may have poor thermal conduction.

 

[digi001]

How does an IGBT have thermal shutdown...? Unless you mean it's built into the control circuit. Does the 6 pack of an integrated NTC?

Sorry, yes it is an IPM that contains IGBTs with integrated features like thermal shutdown.

You can't really get a decent measurement of switching losses. I know people in industry who are able to actually plot current and voltage through each IGBT with enough bandwidth and memory to do a direct measurement, but it takes a ton of specialized equipment which I'm guessing you don't have.

Am i missing something? My scope is a tektronix DPO high bandwidth, high sampling frequency, high memory,4 channel. It can do instantaneous arithmetic. So i measure Vce and Current coming out of IGBT then multiply them on the scope to get watts and plot this? Right? If not I can pull the data out on a USB and do the analysis in Matlab.

 

[mtwieg]

Am i missing something? My scope is a tektronix DPO high bandwidth, high sampling frequency, high memory,4 channel. It can do instantaneous arithmetic. So i measure Vce and Current coming out of IGBT then multiply them on the scope to get watts and plot this? Right? If not I can pull the data out on a USB and do the analysis in Matlab.

The limitation isn't in the scope (so long as you're only doing one or two IGBTs at a time, these guys would do all of them), but rather the instrumentation for instrumentation. Generally you have a laminated bus optimized for low ESL and ESR, so there's no way to probe the collector current accurately without altering those parameters (which have lots of influence on switching losses). I was told that they made their own shunt current sense resistors out of folded aluminum sheet, and that they were only a few mm thick so they could be inserted directly between the bus bar and IGBT terminals without introducing significant ESR or ESL, while still getting a high bandwidth measurement with great accuracy. Magnetic measurements were never accurate enough for switching loss measurement, or so I'm told. Voltage measurement is easier; they just use HV differential probes.

 

[digi001]

I am measuring current using this: **broken link removed**

 

[mtwieg]

I am measuring current using this: **broken link removed**

So I'm assuming you have some length of wire or busbar between the DC bus and the DC terminal of whatever IGBT you're probing. The measurement will be correct, but only for that specific configuration (which is probably not optimal). Also the group delay of the voltage and current probes have to be matched closely (like 10ns if possible).

 

[digi001]

Opps yes I see what you mean now.

I was measuring current in the wrong spot.

That does make the measurement a whole lot tougher.

 

[mtwieg]

Yes, you have to measure in series with just one IGBT, which makes it difficult without compromising performance or accuracy. That's why I would first do a spread sheet just to see if theoretical calculations can explain the problem. If that doesn't give an answer then you'll have to resort to measurement, or trial and error solutions.

 

[FvM]

Sorry, yes it is an IPM that contains IGBTs with integrated features like thermal shutdown.

I'm sorry that you never mentioned an exact type.

I am measuring current using this:

To acquire the dynamic current of individual IGBTs in a module, a small Rogowski coil is typically better suited, e.g. like this
**broken link removed**

 

[mtwieg]

To acquire the dynamic current of individual IGBTs in a module, a small Rogowski coil is typically better suited, e.g. like this
**broken link removed**

We actually tried something like this at one point (the probe looked exactly like the one pictured there, anyways) and the rising and falling edges were always horribly distorted (very slow rise/fall times with weird nonlinearities). Not sure why; the industry friend who brought it in for us to use was really surprised by how poorly it worked as well. Just my $0.02.

 

[FvM]

the rising and falling edges were always horribly distorted (very slow rise/fall times with weird nonlinearities).

Presuming the real waveforms have been different, this sounds like a defective integrator amplifier. If you can accept the pure AC measuring capability, Rogowski coils are amost perfect current sensors. The PEM instruments I have seen yet are performing well.