This paper summarizes the main results and contributions of the MagNet Challenge 2023, an open-source research initiative for data-driven modeling of power magnetic materials. The MagNet Challenge has (1) advanced the stateof-the-art in power magnetics modeling; (2) set up examples for fostering an open-source and transparent research community; (3) developed useful guidelines and practical rules for conducting data-driven research in power electronics; and (4) provided a fair performance benchmark leading to insights on the most promising future research directions. The competition yielded a collection of publicly disclosed software algorithms and tools designed to capture the distinct loss characteristics of power magnetic materials, which are mostly open-sourced. We have attempted to bridge power electronics domain knowledge with state-of-the-art advancements in artificial intelligence, machine learning, pattern recognition, and signal processing. The MagNet Challenge has greatly improved the accuracy and reduced the size of data-driven power magnetic material models. The models and tools created for various materials were meticulously documented and shared within the broader power electronics community.

Predicting core losses is a difficult task for designing magnetic components. Under triangular current waveforms the classical approaches work with duty cycles close to 0.5 but are inaccurate for high and low duty cycles. In this work the composite waveform hypothesis is tested, using data from the MagNet database for N87 material at 25 ºC (3312 data points). Limitations of the composite waveform hypothesis are discussed and a technique to expand the loss space is presented. The new approach based in an extended loss space is validated against the classical improved Generalized Steinmetz Equation, reducing the root mean square and 95 percentile errors by 5.85 and 6.96 times, allowing to predict core losses accurately even at high and low duty cycles. A direct correlation between the expanded loss space and the Steinmetz parameters from the improved Generalized Steinmetz Equation is proven, allowing to represent the core losses as functions of these parameters and generating the composite improved Generalized Steinmetz Equation. The proposed methodology for core loss prediction is validated making use of all data available from the MagNet database (10 materials, 4 temperatures, 59423 data points), achieving better results than the commonly used improved Generalized Steinmetz Equation.

This paper focuses on the design of a charging unit for an electric vehicle fast charging station. With this purpose, in first place, different solutions that exist for fast charging stations are described through a brief introduction. Then, partial power processing architectures are introduced and proposed as attractive strategies to improve the performance of this type of applications. Furthermore, through a series of simulations, it is observed that partial power processing based converters obtain reduced processed power ratio and efficiency results compared to conventional full power converters. So, with the aim of verifying the conclusions obtained through the simulations, two downscaled prototypes are assembled and tested. Finally, it is concluded that, in case galvanic isolation is not required for the charging unit converter, partial power converters are smaller and more efficient alternatives than conventional full power converters.

Prediction of Litz wire losses is a challenging endeavor in the design of high-frequency transformers. Although many analytical approaches of varying complexity can be found in the literature, classical 1-D-based models are the norm in the analysis of wide design spaces. Even then, an important degree of uncertainty exists around the accuracy of the 1-D models. After a critical review of these models and an analysis of their application in Litz wire loss prediction, a new model is proposed, which tries to address some of the concerns and limitations of the existing approaches. The models are evaluated against 2-D finite-element simulations in many different conditions to ensure that the models correctly consider the impact of different parameters. The experimental results are presented to demonstrate the accuracy of the new approach at very high frequencies and high amounts of strands using four prototypes with different Litz wires, designed to correctly match the conditions inherent to any 1-D model.

Achieving the optimal design of power converters requires a deep understanding of the system's dissipation elements to meet the desired performance and safety standards. Once the power converter is designed, it is of key importance to estimate the actual power losses in the real setup, in order to redesign the power converter or monitor and control the semiconductor power losses. With that purpose, calorimetric techniques have outperformed electrical methods. However, they come with mechanical limitations and depend on analytical electrothermal equivalent circuits. These models are highly topology and technology dependent, often resulting in simplistic representations that underestimate thermal effects or complex sets of differential equations. To overcome these challenges, we present a novel post-design automatic method for characterizing semiconductor power losses through its converter thermal dynamics. Our method is rooted in an optimization program that identifies the optimal discrete-time linear model according to a set of power versus temperature profiles. The proposed approach ensures the accurate identification and integration of desired modeling requirements. The methodology is applicable to any power converter topology, and the derived linear model enables the use of standard control theory techniques for monitorization and control. Experiments with a real power converter validate the proposal's versatility and accuracy.

This paper discusses the possibility of achieving partial power processing with non-isolated DC–DC topologies. To this end, partial power converter architectures are seen as an interesting solution for reducing the power processed by the converter. We observed via simulations that single-inductor non-isolated topologies cannot achieve partial power processing since the obtained current and voltage waveforms were the same as those found in a full-power converter. However, when using double inductor non-isolated topologies, reduced current and improved efficiencies were achieved. In order to confirm the results obtained from the simulations, single- and double-inductor topologies were tested experimentally. Finally, it was concluded that a double-inductor non-isolated topology can improve its performance by using partial power processing.