Scholarlink

<p>The axial flux permanent magnet machines (AFPMs) have attracted significant attention as a promising candidate for next-generation electric machines due to their high efficiency and power density. However, three-dimensional finite element analysis (3D FEA), which is essential for the optimal design of AFPMs, is limited by its excessive computational cost. To address this limitation, the quasi-3D analysis is often employed as a faster alternative, but it inevitably compromises prediction accuracy. In this study, a multi-fidelity surrogate modeling framework based on co-kriging is proposed to effectively integrate a small set of high-fidelity (HF) data with a large set of low-fidelity (LF) data. The proposed model is applied to the optimal design of AFPMs and validated using 3D FEA, demonstrating remarkably high predictive accuracy and efficiency. Furthermore, the advantages of the proposed method over other single-fidelity and multi-fidelity machine learning approaches are compared and discussed. This methodology provides an effective solution to the trade-off between computational cost and accuracy, offering a valuable alternative for the optimal design of complex engineering systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient multi-objective design optimization of axial-flux permanent magnet machines for drone propulsion system using multi-fidelity gaussian process

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Jae-Kyung Ryu,
  • Chan-Woo Kim,
  • Seong-Won Jeong,
  • Min-Ro Park,
  • Soo-Hwan Park

摘要

The axial flux permanent magnet machines (AFPMs) have attracted significant attention as a promising candidate for next-generation electric machines due to their high efficiency and power density. However, three-dimensional finite element analysis (3D FEA), which is essential for the optimal design of AFPMs, is limited by its excessive computational cost. To address this limitation, the quasi-3D analysis is often employed as a faster alternative, but it inevitably compromises prediction accuracy. In this study, a multi-fidelity surrogate modeling framework based on co-kriging is proposed to effectively integrate a small set of high-fidelity (HF) data with a large set of low-fidelity (LF) data. The proposed model is applied to the optimal design of AFPMs and validated using 3D FEA, demonstrating remarkably high predictive accuracy and efficiency. Furthermore, the advantages of the proposed method over other single-fidelity and multi-fidelity machine learning approaches are compared and discussed. This methodology provides an effective solution to the trade-off between computational cost and accuracy, offering a valuable alternative for the optimal design of complex engineering systems.