Scholarlink

<p>Low-pressure MEMS piezoresistive pressure sensors commonly employ structured diaphragms as their primary pressure-sensing element, and the diaphragm geometry plays a decisive role in determining overall sensor performance. Designing such structures typically relies on simulation-based optimization, which can be computationally expensive when a large design space is explored. To improve design efficiency, this study proposes a data-driven surrogate-assisted optimization framework that integrates the CatBoost regression model with the Differential Evolution (DE) algorithm. Finite element simulation data are first generated for different parameter combinations within the predefined diaphragm configuration. A CatBoost surrogate model is then trained to predict the structural response of the diaphragm, with geometric parameters as inputs and the maximum stress and deflection as outputs. The trained surrogate model is subsequently embedded within a DE-based optimization process to efficiently search the design space while satisfying the deflection constraint. The results demonstrate that the proposed framework can effectively identify parameter combinations that achieve high stress levels under the specified constraints, while significantly reducing the need for repeated finite element simulations during the optimization process. In addition, SHAP (SHapley Additive exPlanations) analysis is employed to provide interpretable insights into the influence and interaction of structural parameters on diaphragm performance. The proposed CatBoost-DE framework therefore offers an efficient and interpretable data-driven workflow for the structural optimization of low-pressure MEMS piezoresistive pressure sensors.</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

An efficient and interpretable machine learning framework for structured diaphragm optimization in MEMS piezoresistive pressure sensors

  • 期刊论文

  • 发表日期：

    2026年7月18日

查看全文
  • Jinhua Shao,
  • Yilin Zhang,
  • Wenlong Lv,
  • Shitao Chen,
  • Jianmao Li,
  • Huiqiong Xue,
  • Weibing Wang

摘要

Low-pressure MEMS piezoresistive pressure sensors commonly employ structured diaphragms as their primary pressure-sensing element, and the diaphragm geometry plays a decisive role in determining overall sensor performance. Designing such structures typically relies on simulation-based optimization, which can be computationally expensive when a large design space is explored. To improve design efficiency, this study proposes a data-driven surrogate-assisted optimization framework that integrates the CatBoost regression model with the Differential Evolution (DE) algorithm. Finite element simulation data are first generated for different parameter combinations within the predefined diaphragm configuration. A CatBoost surrogate model is then trained to predict the structural response of the diaphragm, with geometric parameters as inputs and the maximum stress and deflection as outputs. The trained surrogate model is subsequently embedded within a DE-based optimization process to efficiently search the design space while satisfying the deflection constraint. The results demonstrate that the proposed framework can effectively identify parameter combinations that achieve high stress levels under the specified constraints, while significantly reducing the need for repeated finite element simulations during the optimization process. In addition, SHAP (SHapley Additive exPlanations) analysis is employed to provide interpretable insights into the influence and interaction of structural parameters on diaphragm performance. The proposed CatBoost-DE framework therefore offers an efficient and interpretable data-driven workflow for the structural optimization of low-pressure MEMS piezoresistive pressure sensors.