How AI-algorithms and AI-hardware Accelerate the Design Process in Power Electronics

Based on a general overview of AI, we analyse current AI implementations and solutions in the application field of power electronics. We show how efficient algorithms on GPU hardware can accurately solve transient circuit simulations in fractions of a second and how ferrite losses can be estimated quickly and accurately using a data-based neural network. We provide insights into so-called Physics-Informed Neural Networks (PINNs), which combine physics, maths and AI to predict all relevant parameters of a transformer without simulation and measurement data, just using the corresponding fundamental physical equations.