Discussion Forum

Hi Johnny,

the impedance BC is only applicable to exterior boundaries. So you must remove those aluminum domains from the mf physics. This is allowed if the skin depth is small enough.

If you want to keep the Al domains in the model you can consider a transition BC which is allowed on interior boundaries. Of course you can model the eddy currents in detail, but then you have to mesh fine enough to resolve the skin depth with 4 - 5 elements. You can apply boundary layers in the mesh.

All these suggestions work in a Union.

Cheers Edgar

Hi Edgar,

Thanks for your reply, indeed the model works as a union with your suggestion, and that's how it's done in the library model, too. I don't care about keeping the Al domains. However, my model does not converge now, the Bicgstab solver stays forever at an error of ~1e3. Also, I'd like to know if there's any way to do this calculation in the time domain, or at least add some harmonics with different amplitudes and phases.

Thanks a lot.

Hi Johnny,

you could try a direct solver. Don't forget to add a Gauge Fixing node to the model then. And of course you can do a frequency sweep with complex input for the coil excitation as a function of the frequency. Of course you can do it in the time domain with proper meshing in the Al domains to resolve the skin depth.

Cheers Edgar

Hi Edgar,

Thanks for your help again, much appreciated!

I tried the direct solver and it solves the model really fast indeed. Is the Gauge Fixing for A-field really necessary though? I read that it is required for direct solvers in another thread too.

Without it, I get reasonable values for the eddy current losses in the aluminium frame (~60W total in a 10000W transformer). With it, the losses are somewhat lower in most aluminium parts, except from the bottom frame and especially its middle leg passing through the winding, where the losses are more than 2000W. That is not possible.

I suppose because of the high frequency I would need an abnormally fine mesh to resolve the skin depth, which would be impractical. Regarding the frequency sweep, what you mean is possibly setting the coil current as i(f)=A(f)<phi(f) and doing an ordinary frequency domain analysis for the desired harmonics? How would I implement it? I'm not sure I can simply add the eddy current losses of each component because of the different phase angles of the harmonics.

Cheers, Johnny.

In many cases the direct solvers just don't converge without the gauge fixing. Sometimes it works without but the A-field may look strange. I had been discussing this a couple of times but haven't got a conclusive answer when it is needed and when not. So I always use it with a direct solver. I would think you can add the losses from the components because all in the Al boundaries is linear, no?

In reality, the eddy current losses should have a periodic waveform like the coil current. I suppose the frequency domain analysis just provides the average value. I believe one shouldn't add these values, just like they shouldn't add the amplitudes of the current harmonics, since they have different amplitude and phase angles. That's why I have settled on the fundamental component for the time being.