Data-Driven Modelling of Polyethylene Recycling under High-Temperature Extrusion.

Area of Science:

• Polymer Science and Engineering
• Materials Science
• Chemical Engineering

Background:

• Controlled thermo-mechanical degradation of polyethylene (PE) is crucial for recycling applications.
• Reactive extrusion processes require accurate data-based modeling for predicting material properties.
• High-density polyethylene (HDPE) and ultra-high molecular weight polyethylene (UHMWPE) were investigated.

Purpose of the Study:

• To investigate the extrusion process for controlled degradation of PE.
• To develop data-based models for predicting rheological behavior and molecular characteristics.
• To compare different modeling techniques, including numerical methods and machine learning.

Main Methods:

• Extrusion of HDPE and UHMWPE under varying conditions (temperature, flow rate, screw speed).
• Development of a numerical method using the Carreau-Yasuda model and Cox-Merz law for rheological prediction.
• Application of inverse rheology and Size Exclusion Chromatography (SEC) for molecular weight distribution analysis.
• Implementation of machine learning techniques: Support Vector Machine Regression (SVR) and sparsed Proper Generalized Decomposition (sPGD).

Main Results:

• Extrusion successfully decreased molecular weight, with weight average molecular weights comparable across numerical, inverse rheology, and SEC methods.
• Inverse rheology accurately predicted molecular weight distributions for HDPE but showed inaccuracies for UHMWPE.
• Machine learning models (SVR and sPGD) effectively predicted process outputs and material characteristics, with sPGD outperforming SVR for zero-shear viscosity.
• Classical process simulation software (Ludovic®) faced limitations due to unknown rheo-kinetic laws.

Conclusions:

• Controlled extrusion is a viable method for polyethylene degradation in recycling.
• Data-based modeling, particularly machine learning (sPGD), offers effective prediction of material properties.
• Further research into rheo-kinetic laws is needed for improved classical process simulations.
• Machine learning shows significant promise for optimizing reactive extrusion processes with limited data.