Conference Agenda

Automotive high-voltage inverters often devote a substantial fraction of their size and cost to electromagnetic interference (EMI) filters, making filter optimization a critical design challenge. This paper presents a new AI-based framework that comprehensively explores both conventional and innovative filter topologies, integrating multi-objective optimization for performance and cost. The proposed approach uses an exploration–exploitation strategy to generate a dataset of diverse filter architectures—ranging from simple LC configurations to multi-level filters with additional damping elements. This dataset is used to train a neural network capable of predicting each candidate topology’s attenuation profile and estimating power losses across inductive, capacitive, and damping components. By replacing costly circuit-level simulations with the trained surrogate model, an iterative design algorithm—implemented via an evolutionary optimizer—rapidly evaluates numerous candidate solutions. The resulting multi-objective optimization balances EMI performance with power loss, size, and cost metrics, enabling a holistic assessment of filter topologies. Demonstrated results indicate close to 250-times reductions in optimizations duration and deeper exploration of the design space, achieving 5 – 10 % better results, paving the way for more efficient and cost-effective EMI filters in automotive powertrains.

Accurate frequency response estimation is crucial in various engineering applications, including electrical circuits especially compact power electronic circuits. In EMC, numerical simulations are often computationally expensive, especially if they need to be run multiple times for optimization. Trained machine learning models offer a suitable solution. With such a model, millions of simulations can be run per day, enabling to perform multi-objective optimization even for complex physical models. However, these models are particularly accurate only at "well-behaved" areas of the transfer function but not at their resonances. This paper investigates a methodology which incorporates the systems physics to better resolve these resonances. Instead of predicting the frequency response on individual frequency samples this work aims to predict the coefficients of its complex-valued transfer function. The results show that the physics-informed approach can resolve the resonances of the transfer function better, but the accuracy of the predicted coefficients must be very high.

Waveguides provide a robust platform for high-bandwidth, low-latency, and interference-resistant communication systems. A promising variant, the non-radiating slotted waveguide (NRSW), enables mobile feeding structures without undesired radiation when properly designed. This work presents a novel electromagnetic model for NRSWs, systematically analyzing the relationship between geometric parameters and unintended radiation. The electromagnetic behavior of this system is described through a physical model and validated via full-wave simulations in Ansys HFSS.

To further optimize the design process, we develop and compare different simulation-based metamodeling strategies. One approach, which does not incorporate prior knowledge of the electromagnetic behavior, employs Ordinary Kriging and Universal Kriging. In contrast, a physics-based approach leverages insights from the physical model to enhance accuracy and efficiency.

The proposed physics-based metamodel achieves an error of only 0.0189 dB/m in the low-radiation region while requiring just 48 simulations needed to fully describe the behavior of NRSW regarding radiation.

In this paper, the equivalence of working volumes (WV) in reverberation chambers (RCs) is investigated by the regime of state-of-the-art sensitivity analysis (SA) techniques while inspecting the effect of changing the size of the device under test (DUT) and the stirrer position to the field uniformity (FU) in a frequency-dependent study close to the lowest usable frequency (LUF). The Sobol'indices as SA measures are evaluated at each stirrer step and frequency. For efficient calculation, state-of-the-art surrogate modeling techniques were utilized to substitute the full-wave simulation model depending on the characteristics of the WVs. The computational expenses of the problem are further reduced by using a decreased number of stirrer steps and frequencies, which are achieved by means of adaptive sampling techniques through kriging interpolation. Furthermore, the size of the experimental design (ED) set, i.e., the number of different configurations is controlled by performing convergence studies. This method is able to reconstruct the 2D sensitivity map (SM) of the configuration parameters as functions of the stirrer steps and the frequency with a fewer number of samples.