Scholarlink

<p>Additive manufacturing, particularly fused deposition modelling (FDM), enables the production of complex geometries but still suffers from inconsistent surface quality and mechanical performance due to strong nonlinear interactions among process parameters. These challenges highlight the need for robust predictive models and data-efficient optimization strategies. Therefore, this study addresses the improvement of predictive accuracy and process optimization for polyethylene terephthalate glycol (PETG)-based FDM using a hybrid methodology that integrates I-optimal design of experiments (DOE), response surface methodology (RSM), and machine learning (extreme gradient boosting (XGBoost), support vector regression (SVR) and random forest (RF)). A total of 52 experimental runs were conducted, varying eight parameters (layer height (LH), print speed (PS), raster angle (RA), number of contours (NC), part orientation (PO), infill density (ID), nozzle temperature (NT), bed temperature (BT)). Surface roughness (Ra, Rz) and impact strength (IS) were selected as responses. RSM quadratic models achieved excellent fits (R² &gt; 0.99), while XGBoost demonstrated superior predictive performance with cross-validated test R² values of 99.35% (Ra), 99.52% (Rz), and 99.22% (IS). In contrast, SVR and RF showed reduced generalization. Analysis of variance revealed LH, PO, and ID as critical contributors to surface roughness and impact strength. Multi-objective optimization using desirability functions and grey relational analysis (GRA) yielded optimal settings (LH = 0.1&#xa0;mm, PS = 40&#xa0;mm/s, RA = 45°, NC = 6, PO = 0°, ID = 70%, BT = 80&#xa0;°C, NT = 240&#xa0;°C). Validation experiments confirmed the predictive reliability of RSM and XGBoost, with percentage errors below 5% across all responses. The proposed framework demonstrates a data-efficient, accurate, and transferable methodology for enhancing both surface and mechanical performance in PETG-based FDM, contributing to more reliable and sustainable additive manufacturing practices.</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

Integrated statistical-machine learning framework for optimizing surface roughness and impact strength in PETG FDM

  • 期刊论文

  • 发表日期：

    2026年3月12日

查看全文
  • Muhammad Abas,
  • Imran Khan,
  • Sahar Noor,
  • Fareed Ahmad

摘要

Additive manufacturing, particularly fused deposition modelling (FDM), enables the production of complex geometries but still suffers from inconsistent surface quality and mechanical performance due to strong nonlinear interactions among process parameters. These challenges highlight the need for robust predictive models and data-efficient optimization strategies. Therefore, this study addresses the improvement of predictive accuracy and process optimization for polyethylene terephthalate glycol (PETG)-based FDM using a hybrid methodology that integrates I-optimal design of experiments (DOE), response surface methodology (RSM), and machine learning (extreme gradient boosting (XGBoost), support vector regression (SVR) and random forest (RF)). A total of 52 experimental runs were conducted, varying eight parameters (layer height (LH), print speed (PS), raster angle (RA), number of contours (NC), part orientation (PO), infill density (ID), nozzle temperature (NT), bed temperature (BT)). Surface roughness (Ra, Rz) and impact strength (IS) were selected as responses. RSM quadratic models achieved excellent fits (R² > 0.99), while XGBoost demonstrated superior predictive performance with cross-validated test R² values of 99.35% (Ra), 99.52% (Rz), and 99.22% (IS). In contrast, SVR and RF showed reduced generalization. Analysis of variance revealed LH, PO, and ID as critical contributors to surface roughness and impact strength. Multi-objective optimization using desirability functions and grey relational analysis (GRA) yielded optimal settings (LH = 0.1 mm, PS = 40 mm/s, RA = 45°, NC = 6, PO = 0°, ID = 70%, BT = 80 °C, NT = 240 °C). Validation experiments confirmed the predictive reliability of RSM and XGBoost, with percentage errors below 5% across all responses. The proposed framework demonstrates a data-efficient, accurate, and transferable methodology for enhancing both surface and mechanical performance in PETG-based FDM, contributing to more reliable and sustainable additive manufacturing practices.