Scholarlink

<p>Cyclone flow is a technique to evaporate the liquid at room temperature. Efficient milliliter-to-microliter sample concentration at room temperature is expected for high-sensitivity analysis in biology and chemistry; however, it is hindered by the difficulty of quantifying the volume in real-time under the cyclone flow. This research targets a non-contact, vision-based system using a physics-guided machine learning (PGML) framework to precisely monitor and control this process under intense cyclone flow. The key is a physics-informed loss function that embeds the container’s geometric constraints into the neural network’s training, substantially enhancing model robustness and accuracy. Experimental results demonstrate the PGML model’s superiority, achieving a nearly 70% reduction in error compared to purely data-driven methods. The system shows a measurement error of just 1.2% and a coefficient of variation of 1.5% at a 20 µL target, meeting stringent bioassay requirements. This work establishes a powerful solution for automated and precisely quantitative sample concentration, promising to advance a wide range of analytical applications.</p> Graphical abstract <p></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

Physics-guided machine learning for real-time, non-contact quantification of liquid volume at micro litter under cyclone flow

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月22日

查看全文
  • Chenyu Zhou,
  • Ruying Wang,
  • Sangming Xu,
  • Roichi Ohta,
  • Hidekatsu Tazawa,
  • Kazuma Mawatari

摘要

Cyclone flow is a technique to evaporate the liquid at room temperature. Efficient milliliter-to-microliter sample concentration at room temperature is expected for high-sensitivity analysis in biology and chemistry; however, it is hindered by the difficulty of quantifying the volume in real-time under the cyclone flow. This research targets a non-contact, vision-based system using a physics-guided machine learning (PGML) framework to precisely monitor and control this process under intense cyclone flow. The key is a physics-informed loss function that embeds the container’s geometric constraints into the neural network’s training, substantially enhancing model robustness and accuracy. Experimental results demonstrate the PGML model’s superiority, achieving a nearly 70% reduction in error compared to purely data-driven methods. The system shows a measurement error of just 1.2% and a coefficient of variation of 1.5% at a 20 µL target, meeting stringent bioassay requirements. This work establishes a powerful solution for automated and precisely quantitative sample concentration, promising to advance a wide range of analytical applications.

Graphical abstract