Scholarlink

<p>Accurate forecasting of tensile properties is important for efficient design of eco-friendly composites. We present a proof-of-concept ensemble workflow to predict tensile strength and tensile modulus of hybrid polypropylene (PP) composites reinforced with long flax fiber bundles, basalt fibers (BF), and rice husk powder (RHP). A lab-scale dataset (<i>n</i> = 65) was generated under standardized testing. Preprocessing (Savitzky–Golay denoising, feature standardization) preceded Optuna-tuned support vector regression (SVR) and XGBoost, whose predictions were combined via a stacked linear meta-learner. Under ten-fold cross-validation, the ensemble achieved R<sup>2</sup> = 0.881 (RMSE = 0.639 GPa) for modulus and R<sup>2</sup> = 0.907 (RMSE = 1.569&#xa0;MPa) for strength. The framework is a computationally efficient complement to simulation-based analyses for early-stage screening within the explored domain. This is a proof-of-concept study based on a small, single-lab dataset (<i>n</i> = 65) without external validation; future work will enlarge the dataset across laboratories, extend to additional properties, and incorporate independent validation. Within the explored domain, the workflow yields actionable composition windows (e.g., 30–35 wt% BF, 4–6 wt% RHP, 3–5 flax plies) that balance stiffness and strength.</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

Proof-of-concept ensemble machine-learning forecasting of tensile properties in hybrid polypropylene composites reinforced with flax, basalt, and rice husk powder

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月20日

查看全文
  • Thanh Mai Nguyen Tran,
  • Duy Tran Quang,
  • Prabhakar M.N.,
  • Xiem Nguyen Thang,
  • Jung-il Song

摘要

Accurate forecasting of tensile properties is important for efficient design of eco-friendly composites. We present a proof-of-concept ensemble workflow to predict tensile strength and tensile modulus of hybrid polypropylene (PP) composites reinforced with long flax fiber bundles, basalt fibers (BF), and rice husk powder (RHP). A lab-scale dataset (n = 65) was generated under standardized testing. Preprocessing (Savitzky–Golay denoising, feature standardization) preceded Optuna-tuned support vector regression (SVR) and XGBoost, whose predictions were combined via a stacked linear meta-learner. Under ten-fold cross-validation, the ensemble achieved R2 = 0.881 (RMSE = 0.639 GPa) for modulus and R2 = 0.907 (RMSE = 1.569 MPa) for strength. The framework is a computationally efficient complement to simulation-based analyses for early-stage screening within the explored domain. This is a proof-of-concept study based on a small, single-lab dataset (n = 65) without external validation; future work will enlarge the dataset across laboratories, extend to additional properties, and incorporate independent validation. Within the explored domain, the workflow yields actionable composition windows (e.g., 30–35 wt% BF, 4–6 wt% RHP, 3–5 flax plies) that balance stiffness and strength.