Dynamic Fracture Strength Prediction of HPFRC Using a Feature-Weighted Linear Ensemble Approach

Area of Science:

• Materials Science and Engineering
• Civil Engineering
• Computational Mechanics

Background:

• High-Performance Fiber-Reinforced Concrete (HPFRC) is crucial for structures under extreme loads due to its durability and crack resistance.
• Existing methods for measuring HPFRC dynamic fracture strength are costly and lack standardized protocols.
• Current data-driven models often fail to balance prediction accuracy with physical interpretability.

Purpose of the Study:

• To develop a highly accurate and interpretable method for predicting the Mode I dynamic fracture strength of HPFRC under high strain rates.
• To address the limitations of current experimental and computational approaches for HPFRC fracture behavior analysis.

Main Methods:

• Construction of a comprehensive database with 161 sets of HPFRC high-strain-rate test data.
• Identification of key modeling variables using correlation analysis and error-driven feature selection.
• Development and comparison of ensemble models (Feature-Weighted Linear Ensemble - FWL, Voting) using six base machine learning algorithms (KNN, RF, SVR, LGBM, XGBoost, MLPNN).
• Application of SHAP and LIME for global and local model interpretability analysis.

Main Results:

• The FWL model achieved superior predictive performance (R² = 0.908, RMSE = 2.632) on the test set, outperforming individual models and the Voting ensemble.
• Strain rate and fiber volume fraction were identified as the dominant factors influencing dynamic fracture strength.
• Strain rate exhibits a highly nonlinear response mechanism across different ranges, impacting HPFRC fracture behavior.

Conclusions:

• The developed FWL-based prediction framework offers a robust, accurate, and interpretable approach for HPFRC dynamic fracture strength.
• This integrated method provides valuable insights into HPFRC fracture mechanisms under high-strain-rate conditions.
• The study presents a novel and effective solution for predicting the complex fracture behavior of HPFRC.