How we build an Artificial Intelligence for magnetic components
I have been thinking about creating the future of Power Electronics for more than 10 years and I have failed constantly, but this is part of the process and remember, the only way to know that you are really finding a problem to solve and product market fit, is by understanding the feeling of not solving a real problem.
When you are not solving a problem, the conversation with the potential users is very easy, the potential user is kind and says a very interesting suggestion, including some congratulations in the middle of the conversation.
When you are solving a problem, the user is hard, is angry and the conversation can be very dysfunctional and confusing because the problem is huge. And this is what happens when you are talking about magnetic design, manufacturing, and testing, especially in the automotive industry.
My friend Francisco Berlanga used to say — “Building real innovative technologies is a very tough process”. You can figure out how tough creating a AI for Magnetic design could be.
The most important part of an AI is creating a dataset of measurements. It is not very glamourous, but this is the way we teach our AI about the physical properties of the windings and the magnetic cores.
Today, we are including a lot of data related to Litz wires, especially for Frequencies in the range of 100 kHz and 300 kHz. In the Automotive industry, most of the designs are a combination of Litz wires and foils and we need to include a lot of data for different coils, shapes…and especially, winding configurations.
The losses in transformers and inductors for the automotive sector are critical and the most difficult part of the estimate is the Rac resistance of the windings. Most designers spend a huge amount of hours doing FEM simulations, probably, some of them will say, it took them just a few hours. Of course, you can use your human experience and select directly 2–3 options and optimize it manually, however, in the automotive industry, iterations are very expensive and slow and the risks for the TIER1 companies of being out of a business are too high to take. This is the reason, why at Frenetic we try to minimize the error in the predictions with technology.
You need a lot of talented engineers not only programming and thinking but sometimes they need to get out of the office and go to the lab and build magnetics properly. I want to say thank you to all the team that spent every day a lot of hours building these magnetics (Angel, Carlos, Joyu, Nelson, Edu, Belén and Pablo).
It’s as simple as this: use Frenetic to ensure the success of your business.
