App Notes
Design and Simulation of a 500W PSFB
In this article, we’ll explain how to design it using Frenetic AI and as bonus, we will talk about our patent for integrating the three magnetics in one.
We developed a machine learning-based model that integrates highly accurate data from numerical simulations with the speed of AI. Using tree-based models, as opposed to neural networks, our approach learns the underlying physics of foil winding losses, enabling rapid and precise predictions across a wide range of operational conditions and design features, and including the presence of an air gap. This solution significantly reduces computation time while maintaining high accuracy, enhancing the iterative design process in power electronics, and offering a practical and efficient alternative to traditional methods.
If computational methods for physical modeling were fast, we would be all set: numerical predictions (e.g., Finite Element Analysis (FEA or FEM)) would hit high accuracy in a practical timeframe. But they are slow, and alternatives are needed. Especially for an iterative process like magnetic design. Traditional modeling approaches in power electronics have significant limitations: (i) numerical methods like FEM offer high accuracy but are slow, (ii) analytical methods are accurate and help understanding but are often limited to specific cases and are usually hard to develop, and (iii) empirical methods, while simple, are limited in scope and not versatile when aiming for generalization. Machine learning, on the other hand, is fast and improves when fed with more and more data.
This publication aims to leverage the best of both worlds, i.e., ML and numerical approaches, on the example of foil winding losses, a big challenge in power electronics design, especially in highfrequency applications. There are few analytical models for foil winding losses, and these are often restricted to scenarios without an air gap [1][2] or where the foil has the height of the core window height [3]. Even though these are typical cases, in many situations, finite element modeling (FEM) remains the only viable option. We invest time in generating highly accurate data through numerical simulations. Then an ML model learns the underlying physics and is now enabled to make highly accurate predictions much quicker. Like this, rapid and accurate predictions are achieved, improving the iterative design process. Our novel work uses those well-established machine learning models to solve the problem of foil losses in magnetics with the presence of an air gap.
In this article, we’ll explain how to design it using Frenetic AI and as bonus, we will talk about our patent for integrating the three magnetics in one.
In this article, we’ll explain how to design it using Frenetic AI and as bonus, we will talk about our patent for integrating the three magnetics in one.
.png)
What if you could design custom magnetics automatically, right after creating your circuit? Now you can.
What if you could design custom magnetics automatically, right after creating your circuit? Now you can.
This session, led by Nicola Rosano, is perfect for engineers new to LLC transformer design who want to learn the fundamentals and streamline their design process using Frenetic Simulator.
This session, led by Nicola Rosano, is perfect for engineers new to LLC transformer design who want to learn the fundamentals and streamline their design process using Frenetic Simulator.
Reach out now and see what Frenetic can do for you.
