Scholarlink

<p>The widespread adoption of wide-bandgap (WBG) power devices has raised converter switching frequencies into the megahertz range, substantially improving power density and efficiency but intensifying electromagnetic interference (EMI). Consequently, the integrated co-design of compact power stages and EMI filters has become essential. Traditional EMI filter design methods relying on iterative manual tuning or computationally intensive simulations, which struggle with the complex interplay of component parameters and layout-dependent parasitic effects. This study proposes an automated optimization framework integrating piecewise gaussian process regression (PGPR) with a genetic algorithm (GA). The PGPR surrogate model accurately captures the intricate frequency-dependent characteristics of EMI filters from sparse experimental data, while robustly quantifying predictive uncertainties. Leveraging this surrogate model, the GA efficiently searches discrete, standardized component libraries to identify the optimal, Bill-of-Materials-ready (BOM-ready) EMI filters. Validation across 75 filter configurations confirmed high predictive accuracy (R²=0.99) and precise uncertainty quantification. A detailed case study further demonstrates the practical effectiveness of the approach. An optimal EMI filter was autonomously generated, fabricated, and experimentally verified to meet stringent CISPR 32 class B standards, achieving excellent predictive agreement (RMSE of 0.62 dB). This work transforms EMI filter design from manual trial-and-error into a streamlined, automated, data-driven process, fully unlocking the potential of SiC/GaN converters.</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

PGPR-GA hybrid framework for automated EMI filter design

  • 期刊论文

  • 发表日期：

    2025年11月18日

查看全文
  • Yifan Shi,
  • Bing Chen,
  • Wei Yan,
  • Sirui Dai,
  • Meng Xia Zhou

摘要

The widespread adoption of wide-bandgap (WBG) power devices has raised converter switching frequencies into the megahertz range, substantially improving power density and efficiency but intensifying electromagnetic interference (EMI). Consequently, the integrated co-design of compact power stages and EMI filters has become essential. Traditional EMI filter design methods relying on iterative manual tuning or computationally intensive simulations, which struggle with the complex interplay of component parameters and layout-dependent parasitic effects. This study proposes an automated optimization framework integrating piecewise gaussian process regression (PGPR) with a genetic algorithm (GA). The PGPR surrogate model accurately captures the intricate frequency-dependent characteristics of EMI filters from sparse experimental data, while robustly quantifying predictive uncertainties. Leveraging this surrogate model, the GA efficiently searches discrete, standardized component libraries to identify the optimal, Bill-of-Materials-ready (BOM-ready) EMI filters. Validation across 75 filter configurations confirmed high predictive accuracy (R²=0.99) and precise uncertainty quantification. A detailed case study further demonstrates the practical effectiveness of the approach. An optimal EMI filter was autonomously generated, fabricated, and experimentally verified to meet stringent CISPR 32 class B standards, achieving excellent predictive agreement (RMSE of 0.62 dB). This work transforms EMI filter design from manual trial-and-error into a streamlined, automated, data-driven process, fully unlocking the potential of SiC/GaN converters.