Assessment of small strain modulus in soil using advanced computational models.

Area of Science:

• Geotechnical Engineering
• Soil Mechanics
• Computational Geosciences

Background:

• The small-strain shear modulus (Gmax) is vital for seismic site response and foundation design.
• Gmax is influenced by soil properties like uniformity coefficient, void ratio, particle size, and confining stress.

Purpose of the Study:

• To develop a comprehensive Gmax database for soils.
• To propose and evaluate advanced computational models for Gmax prediction.

Main Methods:

• Established a Gmax database.
• Developed and compared three advanced computational models (including XGBoost).
• Utilized nine performance indicators, including four novel metrics, for rigorous model evaluation.

Main Results:

• The XGBoost model significantly outperformed other models.
• All models achieved high Gmax prediction accuracy (R² > 0.9, RMSE < 30, MAE < 25).
• Proposed models showed superior performance in IOS, IOA, a20-index, and PI metrics compared to empirical methods.

Conclusions:

• Advanced computational models, particularly XGBoost, offer superior Gmax prediction accuracy and generalization ability.
• These models provide a more reliable alternative to traditional empirical formulas for geotechnical applications.
• The findings contribute to improved seismic hazard assessment and foundation engineering practices.