Polychronopoulos Nickolas D, Sarris Ioannis, Vlachopoulos John
Department of Mechanical Engineering, University of West Attica, Ancient Olive Grove Campus, Egaleo, 12241 Athens, Greece.
Department of Chemical Engineering, McMaster University, Hamilton, ON L8S 4L7, Canada.
Molecules. 2025 Apr 23;30(9):1879. doi: 10.3390/molecules30091879.
Achieving a uniform thickness and defect-free production in the flat die extrusion of polymer sheets and films is a major challenge. Dies are designed for one extrusion scenario, for a polymer grade with specified rheological behavior, and for a given throughput rate. The extrusion of different polymer grades and at different flow rates requires trial-and-error procedures. This study investigated the application of machine learning (ML) to provide guidance for the extrusion of sheets and films with a reduced thickness, non-uniformities, and without defects. A dataset of 200 cases was generated using computer simulation software for flat die extrusion. The dataset encompassed variations in die geometry by varying the gap under a restrictor, polymer rheological and thermophysical properties, and processing conditions, including throughput rate and temperatures. The dataset was used to train and evaluate the following three powerful machine learning (ML) algorithms: Random Forest (RF), XGBoost, and Support Vector Regression (SVR). The ML models were trained to predict thickness variations, pressure drops, and the lowest wall shear rate (targets). Using the SHapley Additive exPlanations (SHAP) analysis provided valuable insights into the influence of input features, highlighting the critical roles of polymer rheology, throughput rate, and the gap beneath the restrictor in determining targets. This ML-based methodology has the potential to reduce or even eliminate the use of trial and error procedures.
在聚合物片材和薄膜的平模挤出中实现均匀厚度和无缺陷生产是一项重大挑战。模具是针对一种挤出方案、具有特定流变行为的聚合物等级以及给定的产量设计的。不同聚合物等级和不同流速下的挤出需要反复试验的过程。本研究调查了机器学习（ML）的应用，以为厚度减小、无均匀性且无缺陷的片材和薄膜挤出提供指导。使用用于平模挤出的计算机模拟软件生成了一个包含200个案例的数据集。该数据集涵盖了通过改变限流器下方的间隙、聚合物流变和热物理性质以及加工条件（包括产量和温度）而产生的模具几何形状变化。该数据集用于训练和评估以下三种强大的机器学习（ML）算法：随机森林（RF）、XGBoost和支持向量回归（SVR）。训练ML模型以预测厚度变化、压降和最低壁面剪切速率（目标）。使用SHapley加性解释（SHAP）分析提供了关于输入特征影响的宝贵见解，突出了聚合物流变学、产量以及限流器下方间隙在确定目标方面的关键作用。这种基于ML的方法有可能减少甚至消除反复试验过程的使用。
