Scholarlink

<p>Electrohydrodynamic jet printing enables the generation of femtoliter-scale droplets, offering advantages such as additive manufacturing, fine patterning, and high process efficiency. With these advantages, electrohydrodynamic jet printing technology holds broad application potential for repairing complex electronic structures at micro- and nanoscales. This technology has entered industrial application. However, precise droplet volume control is required during repair, and conventional methods often fail to achieve both high efficiency and accuracy. In this work, a physics-guided closed-loop droplet volume control framework is proposed by integrating meniscus-based physical modeling with data-driven reinforcement learning. Key process parameters are reduced to physically interpretable equivalent parameters via physics-inspired dimensionality reduction, significantly simplifying the control space and enabling efficient learning. A reinforcement learning strategy is then employed to adaptively regulate these equivalent parameters, with a reward function directly reflecting droplet volume deviation and filling accuracy to ensure alignment with practical manufacturing objectives. The framework is experimentally validated on an industrial E-jet printing platform. Using a 1200 ppi pixel pit substrate as a representative case, the droplet filling rate is improved from 65.6% to 96.8%, with substantially enhanced adjustment efficiency compared to manual tuning.</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 reinforcement learning for micron-scale precision manufacturing in electrohydrodynamic jet processes

  • 期刊论文

  • 发表日期：

    2026年4月20日

查看全文
  • Ziwei Zhao,
  • Wei Chen,
  • Yangwen Zhao,
  • Jiacong Xiong,
  • Xin Li,
  • Pengcheng Yu,
  • Jiankui Chen

摘要

Electrohydrodynamic jet printing enables the generation of femtoliter-scale droplets, offering advantages such as additive manufacturing, fine patterning, and high process efficiency. With these advantages, electrohydrodynamic jet printing technology holds broad application potential for repairing complex electronic structures at micro- and nanoscales. This technology has entered industrial application. However, precise droplet volume control is required during repair, and conventional methods often fail to achieve both high efficiency and accuracy. In this work, a physics-guided closed-loop droplet volume control framework is proposed by integrating meniscus-based physical modeling with data-driven reinforcement learning. Key process parameters are reduced to physically interpretable equivalent parameters via physics-inspired dimensionality reduction, significantly simplifying the control space and enabling efficient learning. A reinforcement learning strategy is then employed to adaptively regulate these equivalent parameters, with a reward function directly reflecting droplet volume deviation and filling accuracy to ensure alignment with practical manufacturing objectives. The framework is experimentally validated on an industrial E-jet printing platform. Using a 1200 ppi pixel pit substrate as a representative case, the droplet filling rate is improved from 65.6% to 96.8%, with substantially enhanced adjustment efficiency compared to manual tuning.