Prediction of mixed-mode I/II fracture toughness of rock-concrete bi-material disc with interface crack: Interpretable NRBO-XGBoost-SHAP model and experimental validation

Key Points

• The aim is to predict the effective fracture toughness of rock-concrete bi-materials using an interpretable machine learning model.
• Developed an NRBO-XGBoost-SHAP model to predict effective fracture toughness (K eff ) using data from CSTBD testing.
• Compared seven machine learning models to identify the most effective for predicting K eff .
• Conducted experimental validation with a new dataset to substantiate the model's reliability.
• The NRBO-XGBoost model outperformed six other machine learning models in predicting K eff.
• Integration of SHAP analysis enhanced model interpretability related to classical fracture mechanics.
• Experimental validation confirmed the NRBO-XGBoost model's predictions with high accuracy.