Conference Agenda

This scientific tutorial explores the emerging role of Artificial Intelligence (AI) techniques in revolutionizing power electronics engineering processes. The presentation begins with an overview of the AI hype cycle, highlighting key technologies influencing various levels of power electronics design and optimization. The tutorial then delves into three main topics.

First, it compares sampling and optimization approaches for large-scale simulations, discussing the limitations of random and genetic algorithm methods in high-dimensional spaces. The presentation introduces AI-motivated searching strategies, particularly focusing on Continuous Active Random Search (CARS), which offers an adaptable trade-off inspired by Reinforcement Learning (RL) techniques.

Second, the tutorial examines inter- and extrapolation of engineering data using both established and novel neural network architectures. It showcases the MagNet Challenge, demonstrating accurate prediction of magnetic losses, and compares standard fully connected Neural Networks (NN) with Kolmogorov-Arnold Networks (KAN) for power electronics circuit simulations. The discussion addresses challenges in modeling highly non-linear behaviors and precise prediction of realistic operation points in resonant converter topologies.

Lastly, the tutorial explores Physics-Informed Neural Networks (PINN) for transformer design, highlighting their ability to predict relevant power electronic quantities for arbitrary 2D transformer geometries without relying on empirical data. This approach opens new possibilities for on-the-fly optimizations in power electronics design purely driven by the fundamental physical equations.

Through these topics, the tutorial provides a comprehensive overview of how AI is transforming power electronics engineering, offering insights into advanced modeling, optimization, and design techniques.