Scholarlink

<p>The prediction of mechanical properties in architected cellular materials remains challenging due to complex structure–property relationships. The study compares five machine learning regression algorithms, Random Forest, XGBoost, Decision Tree, SVR, and kNN, in a detailed performance analysis. The objective is to predict the maximum bending stress in bio-based sandwich structures. A dataset comprising 108 samples spanning nine distinct core geometries was extracted from literature, encompassing both experimental three-point bending tests and numerical simulations of polylactic acid/flax fiber composites. Six input features were employed: relative density, geometry type, core tensile modulus, core compression modulus, core Poisson’s ratio, and replicate number. The dataset was partitioned into training (80%) and testing (20%) subsets, with hyperparameter optimization performed via GridSearchCV with fivefold cross-validation. Feature importance analysis identified tensile modulus as the dominant predictor, while sensitivity analysis revealed Poisson’s ratio as the most influential parameter. Pareto front optimization demonstrated that Random Forest achieves optimal balance between accuracy and robustness, whereas XGBoost offers maximum accuracy with increased hyperparameter sensitivity. Finally, geometry-specific optimization revealed that auxetic structures maintain the highest bending stress across the density range, while chiral geometries exhibit lower strength. Overall, the study proposes practical design recommendations, suggesting relative densities between 12 and 15% to enhance structural efficiency, and provides a reliable framework for the data-driven design of bio-composite sandwich structures.</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 Prediction and Optimization of Mechanical Properties in Cellular Materials: A Comparative Study of Multiple Regression Algorithms

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • Khawla Essassi,
  • Anas Bouguecha,
  • Mohamed Haddar

摘要

The prediction of mechanical properties in architected cellular materials remains challenging due to complex structure–property relationships. The study compares five machine learning regression algorithms, Random Forest, XGBoost, Decision Tree, SVR, and kNN, in a detailed performance analysis. The objective is to predict the maximum bending stress in bio-based sandwich structures. A dataset comprising 108 samples spanning nine distinct core geometries was extracted from literature, encompassing both experimental three-point bending tests and numerical simulations of polylactic acid/flax fiber composites. Six input features were employed: relative density, geometry type, core tensile modulus, core compression modulus, core Poisson’s ratio, and replicate number. The dataset was partitioned into training (80%) and testing (20%) subsets, with hyperparameter optimization performed via GridSearchCV with fivefold cross-validation. Feature importance analysis identified tensile modulus as the dominant predictor, while sensitivity analysis revealed Poisson’s ratio as the most influential parameter. Pareto front optimization demonstrated that Random Forest achieves optimal balance between accuracy and robustness, whereas XGBoost offers maximum accuracy with increased hyperparameter sensitivity. Finally, geometry-specific optimization revealed that auxetic structures maintain the highest bending stress across the density range, while chiral geometries exhibit lower strength. Overall, the study proposes practical design recommendations, suggesting relative densities between 12 and 15% to enhance structural efficiency, and provides a reliable framework for the data-driven design of bio-composite sandwich structures.