Machine-Learning Optimization of Recycled PETG Pellet Extrusion Using Response Surface Modeling and Multi-Objective Experimental Design

Authors

  • Ashik Mahmud Mechanical Specialist, KrisEnergy Ltd., Cumilla, Bangladesh; Author
  • Md Momen Shahriar Joarder Quality Assurance Engineer (QA Engineer), Walton Hi-Tech Industries Limited, Bangladesh; Author

Keywords:

Recycled PETG, Extrusion, Machine Learning, Response Surface Methodology, Multi-Objective Optimization

Abstract

Recycled polyethylene terephthalate glycol (pet) offers considerable potential for resource-efficient polymer manufacturing, although variability in recycled feedstock and nonlinear interactions among extrusion parameters can adversely affect mechanical properties, dimensional consistency, surface quality, productivity, and energy performance. This study quantitatively evaluated and optimized pet pellet extrusion through an integrated framework combining response surface methodology (RSM), machine-learning prediction, multi-objective optimization, and experimental confirmation. A total of 96 extrusion trials were planned, of which 88 were retained for analysis, representing a 91.7% retention rate and 98.6% overall data completeness. Barrel temperature, screw speed, feed rate, and die temperature were investigated as principal controllable parameters. Mean tensile strength was 47.86 ± 3.92 MPa, Young’s modulus was 2.08 ± 0.17 GA, throughput was 2.43 ± 0.16 kg/h, and extrudate diameter was 1.752 ± 0.036 mm. RSM models achieved R² values ranging from 0.901 to 0.958, with statistically significant main, interaction, and quadratic effects at p < 0.05 and nonsignificant lack-of-fit results. Among the evaluated predictive models, Boost demonstrated the strongest generalization, achieving a training R² of 0.982, five-fold cross-validation R² of 0.949 ± 0.016, independent holdout R² of 0.957, and overall MAPE of 3.18%. Multi-objective optimization identified an optimal compromise at 239.2 °C barrel temperature, 52.6 rpm screw speed, 2.86 kg/h feed rate, and 231.4 °C die temperature, producing a composite desirability of 0.931. Predicted tensile strength, throughput, surface roughness, and specific energy consumption were 53.87 MPa, 2.61 kg/h, 3.88 µm, and 0.438 kWh/kg, respectively. Five independent confirmation trials produced tensile strength of 53.42 ± 0.71 MPa and throughput of 2.58 ± 0.05 kg/h, with most prediction errors remaining below 2%. The findings demonstrated that integrating interpretable RSM with validated machine learning and multi-objective optimization provided a robust framework for balancing mechanical quality, dimensional stability, productivity, surface performance, and energy efficiency in recycled PETG extrusion.

Author Biographies

  • Ashik Mahmud, Mechanical Specialist, KrisEnergy Ltd., Cumilla, Bangladesh;

    B.Sc. in Mechanical Engineering, Khulna University of Engineering & Technology (KUET), Bangladesh;

  • Md Momen Shahriar Joarder, Quality Assurance Engineer (QA Engineer), Walton Hi-Tech Industries Limited, Bangladesh;

    Rajshahi University of Engineering and Technology (RUET), Bangladesh;

Downloads

Published

2022-12-03

How to Cite

Ashik Mahmud, & Md Momen Shahriar Joarder. (2022). Machine-Learning Optimization of Recycled PETG Pellet Extrusion Using Response Surface Modeling and Multi-Objective Experimental Design. Journal of Sustainable Development and Policy, 12(12), 83-133. https://jsdp-journal.org/index.php/jsdp/article/view/83