文献类型：期刊文献

英文题名：A machine learning-driven digital twin framework for mold optimization and safety assessment in highly filled viscous materials extrusion

作者：Zhou, Zihan[1];Zhuang, Xin[1];Xu, Chenglong[1];Wang, Yu[1];Li, Guo[1];Wang, Jiangning[2];Xie, Linsheng[1]

机构：[1]East China Univ Sci & Technol, Shanghai 200237, Peoples R China;[2]Xian Modern Chem Res Inst, Xian 710000, Peoples R China

年份：2026

卷号：174

起止页码：1206

外文期刊名：JOURNAL OF MANUFACTURING PROCESSES

收录：;EI(收录号：20263221276142);Scopus(收录号：2-s2.0-105046821822);WOS:【SCI-EXPANDED(收录号:WOS:001848602800001)】；

语种：英文

外文关键词：Highly filled viscous materials; Single-screw extrusion; Machine learning; Computational fluid dynamics; Structure optimization

摘要：The single-screw continuous extrusion of highly filled viscous propellants is constrained by flow instabilities and safety risks, making empirical design hazardous. This study proposes a physics-informed, data-driven framework coupling computational fluid dynamics (CFD), physical experiments, and machine learning (ML) to optimize the extruder structural geometry and establish a safe processing envelope. CFD simulations revealed that extrusion quality relies on the screw's pressure-building ability and the die's radial velocity gradients. Structural optimization demonstrated that reducing the screw internal taper (from 1:10 to 1:18) significantly enhances the volumetric squeezing effect. Concurrently, extending the die's sizing and pre-sizing sections acts as a fluidic buffer that reorients radial momentum into stable axial flow. This effectively mitigates unstable flow induced by excessive radial velocity disparities, thereby improving the exit-velocity nonuniformity coefficient from 1.83 to an optimal 1.21. These numerical findings were validated via extrudate swell experiments, showing high dimensional consistency with a minimal error of 7.88%. To overcome the computational cost of iterative CFD evaluations, a Random Forest-based model was constructed to map the complex nonlinear relationships between structural parameters and critical safety targets (maximum pressure and shear rate). Feature importance analysis proved that the die geometry (sizing and pre-sizing sections) dictated approximately 85% of the predictive variance, identifying them as the dominant parameters controlling viscous dissipation and pressure accumulation. Ultimately, this comprehensive framework shifts from trial-and-error to predictive digital opti-mization, providing a robust design methodology for the safe and high-quality processing of complex highly viscous non-Newtonian fluids.