Dynamic Strength Prediction of Brittle Engineering Materials via Stacked Multi-Model Ensemble Learning and...

Area of Science:

• Geotechnical Engineering
• Materials Science
• Computational Mechanics

Background:

• Accurate prediction of dynamic compressive strength in brittle materials is crucial for underground engineering and safety.
• Existing models may lack accuracy and interpretability for complex material behaviors.

Purpose of the Study:

• To develop an enhanced machine learning framework for predicting dynamic compressive strength.
• To improve model accuracy and interpretability using ensemble learning and SHAP.
• To identify key factors influencing material dynamic strength.

Main Methods:

• A novel stacking ensemble learning framework was developed.
• Six regression models (KNN, RF, GBDT, LightGBM, XGBoost, MLPNN) were trained and evaluated.
• SHapley Additive exPlanations (SHAP) were used for model interpretability.

Main Results:

• The proposed stacking model significantly outperformed individual base models in prediction accuracy, stability, and generalization.
• Strain rate was identified as the dominant factor influencing dynamic strength prediction.
• Static strength, P-wave velocity, and bulk density were found to be significant contributors.

Conclusions:

• The developed framework offers a robust and interpretable tool for predicting brittle material dynamic strength.
• The findings provide valuable insights into the mechanical behavior of materials under dynamic loading.
• This approach supports enhanced safety assessments and design in underground engineering.