Dear PLECS forum,
I am trying to model a 2x parallel interleaved 3-level flying capacitor inverter with thermal models provided by EPC to model the switching losses (The topology I’m modelling is besides the point).
I noticed that when probing the custom part provided by EPC and feeding it through a periodic impulse average block, it gives a different result from using the built-in “switch loss calculator” block (even though the internal implementation should be the same as the former, and the averaging time is also set to the same). I have been diving deeper into this problem and start to think that there is a mistake in the determination of the switching losses in the EPC switch model.
Below you can see the differences in the results for a single switch (somewhere in my FCC model), one gives ~0.20 and the other ~0.29, which is a significant difference:
(The switches circled in red are not the actual switches that are probed; they only serve to illustrate what part I dragged into the probe and the “switch loss calculator” block.
What I noticed was the following:
The names of the parts do not correspond, even though I dragged exactly the same part into the probe and the switch loss calculator block.
This made me realise that the EPC custom switch part is just a masked subsystem, so I started digging into that. And there I found the MOSFET that the switch loss calculator is actually probing:
I verified this by dragging this underlying MOSFET from the EPC switch subsystem mask into the probe aswell:
Now, the method using the probe does give the same result as the switch loss calculator block.
The only question that remains now is: which value is correct?
(I couldn’t post more pictures in one post, so I split the post into two sections)
To try to find that out, I looked inside the EPC switch subsystem mask, and in there, they have created some process to determine the switching losses:
Two subsystems are responsible for this, let’s start with the one circled in red:
In here the cd losses and sw losses of the MOSFET are probed like normal. The sw losses are then fed into the “Switching loss separation” subsystem circled in green:
In here, they multiply the switching losses by the rising and falling edges of the switching signal. This is where I think the model is wrong. The makers of the model probably assumed that the switching losses retrieved from the probe are continuous, and they tried to make energy pulses out of it in this fashion. However, the switching losses that are probed are already energy pulses.
To verify this, I then scoped:
- The switching signal (scaled down to the level of the switching loss pulses),
- The switching loss pulses from the EPC switch model, and
- The switching loss pulses probed from the underlying MOSFET
The results can be seen here:
As can be seen, for some reason, some of the energy pulses generated in the underlying MOSFET model are lost during the rising- and falling-edge detection operation in the EPC model. Since the extra energy pules in the underlying mosfet model (in red) do allign with switching instances, I assume that this gives the correct result, whereas the EPC model gives a false (lower) result when probed.
Could someone please verify this assumption? Are the red pulses from the underlying MOSFET model indeed the ones I should use? If so, it might be good to contact EPC’s engineers so their GaN-FET thermal models for PLECs can be updated to resolve this issue.
Thank you in advance for taking the time to look at my comprehensive post.
Kind regards,
Timokuiz
