Shukai Wang

This article presents and applies the concept of time-domain foundation models (FM) to analyzing nonlinear soft ferrite material under transient excitations. Key principles of a time-domain foundational model are defined and elaborated. A neural network (NN) based foundation model is implemented to characterize different materials under a wide range of transient excitations. The model is backed by MagNetX – a new extension of the MagNet database, which includes extensive measurement data of quasi-steady-state PWM waveforms with minor loop excitations. Different from traditional neural networks that implement neural networks as datasheets, the foundation model utilizes time-domain arbitrary length memory from the $B$-$H$ excitations, and predicts arbitrary future steps based on the respective input. The foundation model is simple, robust and flexible, and precisely characterizes the material $B$-$H$ characteristics during transients.

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.

This letter presents a simplified dc-bias injection circuit utilizing a voltage mirror transformer configured in a unique way for improved equipment applicability and measurement accuracy. A mirror transformer can minimize the impact of the reflected actuation voltages on the dc source and improve the quality of the dc-bias injection. By combining the dc-bias and ac actuation current, the traditional three-winding dc-bias measurement setup was simplified with improved accuracy and reduced equipment complexity. The dc-bias injection circuit was implemented as a part of a highly automated data acquisition system designed for fast and accurate characterization of power magnetic materials under a wide range of operating conditions.

This paper presents the MagNet-AI platform as an online platform to demonstrate the “Neural Network as Datasheet” concept for $B$ – $H$ loop modeling and material recommendation of power magnetics across wide operation range. Instead of directly presenting the measured characteristics of magnetic core materials as time sequences, we employ a neural network to capture the $B$ – $H$ loop mapping relationships of magnetic materials under different excitation waveforms at different temperatures and dc-bias. LSTM and Transformer based neural network models are developed, verified, and compared. The neural network can be used to rapidly predict hysteresis loops and core losses under different operating conditions, compare materials, and recommend materials for design. The neural network model is also proved effective in reconstructing the raw measurement while accurately maintaining the magnetic characteristics, enabling rapid material evaluation and comparison.

This article applies machine learning to power magnetics modeling. We first introduce an open-source database—MagNet—which hosts a large amount of experimentally measured excitation data for many materials across a variety of operating conditions, consisting of more than 500 000 data points in its current state. The processes for data acquisition and data quality control are explained. We then demonstrate a few neural network-based power magnetics modeling tools for modeling the core losses and $B$ – $H$ loops. The neural network allows multiple factors that may influence the magnetic characteristics to be modeled in a unified framework, where the nonlinear behaviors are captured with high accuracy and high generality. Neural network models are found to be effective in compressing the measurement data and predicting the material characteristics, paving the way for “neural networks as datasheets” to assist power magnetics design. Transfer learning is applied to the training of neural network models to further reduce the data size requirement while maintaining sufficient model accuracy.

This article motivates the development of sophisticated data-driven models for power magnetic material characteristics. Core losses and hysteresis loops are critical information in the design process of power magnetics, yet the physics behind them is not fully understood or directly applicable. Both losses and hysteresis loops change for each magnetic material and depend heavily on electrical operating conditions (e.g., waveform, frequency, amplitude, and dc bias), mechanical properties (e.g., pressure and vibration), temperature, and geometry of the magnetic components, and in a nonlinear and coupled fashion. Understanding the complex and intertwined relationship these factors have on core loss is important for the development of accurate models and their applicability and limitations. Existing studies on power magnetics are usually developed based on a small amount of data and do not reveal the full magnetic behavior across a wide range of operating conditions. In this article, based on a recently developed large-scale open-source database— MagNet —the core losses and hysteresis loops of Mn–Zn ferrites are analyzed over a wide range of amplitudes, frequencies, waveform shapes, dc bias levels, and temperatures, to quantify the complexity of modeling magnetic core losses, amplitude permeability, and hysteresis loops and provide guidelines for modeling power magnetics with data-driven methods.

Borophene has important application value, boron nanomaterials doped with transition metal have wondrous structures and chemical bonding. However, little attention was paid to the boron nanowires (NWs). Inspired by the novel metal boron clusters Ln₂B_n⁻ (Ln = La, Pr, Tb, n = 7–9) adopting inverse sandwich configuration, we examined Sc₂B₈ and Y₂B₈ clusters in such novel structure and found that they are the global minima and show good stability. Thus, based on the novel structural moiety and first-principles calculations, we connected the inverse sandwich clusters into one-dimensional (1D) nanowires by sharing B−B bridges between adjacent clusters, and the 1D-Sc₄B₂₄ and 1D-Y₂B₁₂ were reached after structural relaxation. The two nanowires were identified to be stable in thermodynamical, dynamical and thermal aspects. Both nanowires are nonmagnetic, the 1D-Sc₄B₂₄ NW is a direct-bandgap semiconductor, while the 1D-Y₂B₁₂ NW shows metallic feature. Our theoretical results revealed that the inverse sandwich structure is the most energy-favored configuration for transition metal borides Sc₂B₈ and Y₂B₈, and the inverse sandwich motif can be extended to 1D nanowires, providing useful guidance for designing novel boron-based nanowires with diverse electronic properties.