Scholarlink

<p>Improving the use of recycled polymers without causing substantial loss of mechanical performance remains a central issue in sustainable additive manufacturing. For recycled ABS processed by fused deposition modeling (FDM), this study proposes a data-driven workflow that links designed experiments with machine-learning prediction and evolutionary optimization. A Taguchi L16 array was used to examine five variables, namely recycled material ratio, layer thickness, raster angle, nozzle temperature, and printing speed. Based on 1024 Taguchi-expanded data points, five regression models—multiple linear regression (MLR), second-order polynomial regression (SOPR), decision tree regression (DTR), random forest regression (RFR), and support vector regression (SVR)—were assessed in terms of predictive ability and model transparency. Among them, RFR and SVR produced the most accurate predictions, with <i>R</i><sup>2</sup> values above 0.96. The trained models were integrated with an NSGA-based optimization framework to explore the balance between mechanical behavior and recycled ABS utilization. The optimized parameter sets indicate that recycled ABS content can reach 70% with only limited deterioration in performance. Overall, the proposed framework provides a practical route for intelligent decision-making in sustainable polymer processing and offers a general approach for cleaner, more circular additive manufacturing.</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

Machine learning-based multi-objective optimization for sustainable fused deposition modeling using recycled ABS

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Pei-Yu Hung,
  • Meng-Ting Wu,
  • Cheng-Hsien Wu

摘要

Improving the use of recycled polymers without causing substantial loss of mechanical performance remains a central issue in sustainable additive manufacturing. For recycled ABS processed by fused deposition modeling (FDM), this study proposes a data-driven workflow that links designed experiments with machine-learning prediction and evolutionary optimization. A Taguchi L16 array was used to examine five variables, namely recycled material ratio, layer thickness, raster angle, nozzle temperature, and printing speed. Based on 1024 Taguchi-expanded data points, five regression models—multiple linear regression (MLR), second-order polynomial regression (SOPR), decision tree regression (DTR), random forest regression (RFR), and support vector regression (SVR)—were assessed in terms of predictive ability and model transparency. Among them, RFR and SVR produced the most accurate predictions, with R2 values above 0.96. The trained models were integrated with an NSGA-based optimization framework to explore the balance between mechanical behavior and recycled ABS utilization. The optimized parameter sets indicate that recycled ABS content can reach 70% with only limited deterioration in performance. Overall, the proposed framework provides a practical route for intelligent decision-making in sustainable polymer processing and offers a general approach for cleaner, more circular additive manufacturing.