Scholarlink

<p>To increase the ratio of manufactured recycled poly(ethylene terephthalate) (rPET) bottle, the control of ISBM (Injection Stretch Blow Moulding) process must account for varying mechanical and thermal properties of mechanically recycled PET. Calibration and optimization of the process have been successfully realized in past works, but they use costly PDE-based models. Therefore, they are difficult to use online for applications where the process parameters need to be regularly adjusted, i.e. whenever a new batch of preforms made of recycled PET is considered. To address this, a meta-algorithm is proposed to replace the PDE-based digital twin. A gaussian process regression is trained offline using the PDE-based digital twin results in the process’s variabilities range. Using the internal pressure history of a test ISBM operation, our strategy allows us to obtain calibrated results without solving PDEs online. To show the capability of the methodology, a simplified process and its associated parametric uncertainties are enunciated. Finite element simulations of the ISBM process where the properties follow a multivariate Gaussian distribution are used to realize the Gaussian process regression. The&#xa0;quality of the digital twin’s predictions is assessed. Then, an illustrative example is presented, demonstrating the use of the digital twin predictions to optimize the thickness distribution of a bottle following the blowing process.</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

AI-accelerated digital twinning for the online PDE-based control of ISBM processes with recycled PET

  • 期刊论文

  • 发表日期：

    2025年11月3日

查看全文
  • William W. Han,
  • Pierre Kerfriden,
  • Laurianne Viora,
  • Christelle Combeaud,
  • Jean-Luc Bouvard,
  • Sabine Cantournet

摘要

To increase the ratio of manufactured recycled poly(ethylene terephthalate) (rPET) bottle, the control of ISBM (Injection Stretch Blow Moulding) process must account for varying mechanical and thermal properties of mechanically recycled PET. Calibration and optimization of the process have been successfully realized in past works, but they use costly PDE-based models. Therefore, they are difficult to use online for applications where the process parameters need to be regularly adjusted, i.e. whenever a new batch of preforms made of recycled PET is considered. To address this, a meta-algorithm is proposed to replace the PDE-based digital twin. A gaussian process regression is trained offline using the PDE-based digital twin results in the process’s variabilities range. Using the internal pressure history of a test ISBM operation, our strategy allows us to obtain calibrated results without solving PDEs online. To show the capability of the methodology, a simplified process and its associated parametric uncertainties are enunciated. Finite element simulations of the ISBM process where the properties follow a multivariate Gaussian distribution are used to realize the Gaussian process regression. The quality of the digital twin’s predictions is assessed. Then, an illustrative example is presented, demonstrating the use of the digital twin predictions to optimize the thickness distribution of a bottle following the blowing process.