A High-Generalizability Machine Learning Framework for Analyzing the Homogenized Properties of Short Fiber-Reinforced...

Area of Science:

• Materials Science
• Computational Mechanics
• Machine Learning

Background:

• Predicting mechanical properties of short fiber-reinforced polymer composites is crucial for material design.
• Accurate and efficient modeling methods are needed to overcome the limitations of traditional experimental and simulation approaches.

Purpose of the Study:

• To develop a highly generalizable machine learning framework for predicting homogenized mechanical properties of short fiber-reinforced polymer composites.
• To assess the accuracy, efficiency, interpretability, and generalizability of the proposed ensemble machine learning (EML) model.

Main Methods:

• An ensemble machine learning (EML) model using a stacking algorithm with Extra Trees (ET), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting (LGBM) base models.
• Integration of a two-step homogenization micromechanical model with finite element simulations to generate a high-quality ground-truth dataset.
• Performance evaluation using R-squared values, SHapley Additive exPlanations (SHAP) for feature importance, and comparison with experimental data and high-fidelity models.

Main Results:

• The EML model achieved high prediction accuracy with R2 values of 0.988 (train) and 0.952 (test).
• SHAP analysis identified matrix/fiber Young's modulus and fiber content as key factors, with fiber orientation dominating anisotropy.
• The EML model demonstrated strong generalization on experimental data and significantly reduced computational costs compared to high-fidelity models while maintaining accuracy.

Conclusions:

• The developed EML model provides an accurate, efficient, and generalizable approach for predicting the mechanical properties of short fiber-reinforced polymer composites.
• The framework offers a valuable tool for material design and optimization, outperforming traditional methods in terms of computational efficiency.
• Understanding key influencing factors through SHAP analysis enhances the interpretability and reliability of the machine learning predictions.