#77🚀 AI Vs. Analytical Models in Magnetics
SuperPowers for Electrical Engineers 🦸🏻♂️🦸🏼♀️
👋 Hello, friends! Dr. Molina here 👨🔧
Welcome to Dr.Molina Newsletter, where I break down Magnetic design for you every week, in
4 minutesthe time needed.
Have you taken a look at our new web?
During the XVII century, most scientific chemical discoveries sought a different thing. In 1675, Henning Brand, a German scientist, collected 50 bins of urine at his home, expecting to create gold. He was convinced that by distilling urine, he would find gold.
He didn´t find gold, but something happened. The substance started shining and sometimes burst into flames when it was exposed to the light!
After a long story (and a strong smell at home 🤢), he discovered phosphorus.
I don´t know if future discoveries in power electronics will arrive after looking for a very different thing, but definitively, we have to keep an open mind.
In the research of Magnetics, Frenetic is investing millions in building a Scientific Team with proper equipment.
After APEC, I like to review all the conference publications, and I want to share something about the Magnetics track.
There is a University and a Researcher to follow.
The University to watch is the Princeton Department of Power Electronics, with Minjie Chen leading the team, putting a lot of effort into building an Open Data Base of Core losses (Magnet) to apply Artificial Intelligence to the core loss modeling. Diego Serrano (another Spaniard engineer) explained a Neural Network for predicting the BH loop in a lecture during the Magnetic track.
Who is the researcher in Magnetics to follow?
The researcher to follow is Thomas Guillod. Thomas is following a very interesting road around magnetics. His first publications were related to general magnetic design, and suddenly in 2020, he surprised everyone with an article about AI.
He referenced Frenetic in this article (Thank you Thomas, you have been the only one referencing us, even, we wrote the first article about this topic 4 years ago in Bodo Power)
He created a completed open algorithm for designing an inductor using an ANN trained with FEM simulations. At that time, he was at ETH Zurich.
Later (I don´t know precisely when), he moved to Princeton and started collaborating with Minjie Chen in his research group. He appeared as a co-author in every publication about AI in core loss modeling.
It looked like he was starting to become an AI researcher for Magnetics.
But…nop. This year at APEC, he presented an exciting work.
I enjoyed the article a lot. Let me summarize it for you:
The current models of Core losses are based on the Steitzmet equation. He said they were not good enough and explained why:
Relaxation losses are not considered
iGSE is linear with the frequency (in logarithmic scale)
The following picture is part of the article, and it shows the error of the different models with respect to the measurements over a toroid.
Therefore, he presented iGCC, a method that uses measurement data to be accurate, whereas classic methods show a clear error. He even explained that iGSE is a particular solution of iGCC. The iGCC method improves the relation between the measurements and the predictions in similar conditions, concretely, at 5% better than iGSE for the 95th percentile.
He used the data for building an Analytical Model instead of using an Artificial Neural Network (ANN) Model.
The elephant of the room here is:
Are Analytical Models better than ANN for core losses?
In any case, I’m excited about the publications of Thomas Guillod; I think he will be a very important researcher in the Science of Magnetics, and I would love to collaborate with him at some point.
I started this edition by talking about random discoveries in chemical science. I wonder what will be the next big thing in Magnetics: I bet somebody will find something he won´t be looking for, and it will take a long time to really apply it in the industry. What do you think? Is the power electronics industry ready for discoveries out of the typical forums?
Next week is our New Lab Opening in Madrid. It’s a dream come true for me. Building a Scientific Laboratory for Magnetics is a big first step.
If you missed….some of the most widely read articles in the Dr.Molina Magnetics Newsletter: