Prediction of Rockfill Materials' Shear Strength Using Various Kernel Function-Based Regression Models-A Comparative...

Area of Science:

• Geotechnical Engineering
• Computational Mechanics
• Materials Science

Background:

• Safe and cost-effective design of embankment dams requires understanding the mechanical behavior of rockfill materials (RFMs).
• Characterizing RFMs, especially those with large particles (>500 mm), is challenging and expensive due to their complex shear strength properties.

Purpose of the Study:

• To investigate the efficacy of various kernel function-based Gaussian Process Regression (GPR) models for predicting the shear strength of RFMs.
• To identify the most effective GPR model for accurate shear strength estimation in RFMs.

Main Methods:

• Utilized 165 datasets from existing literature to train and test Gaussian Process Regression (GPR) models.
• Compared multiple GPR models employing different kernel functions, with a focus on the Pearson Universal Kernel (PUK).
• Evaluated model performance using metrics such as R-squared (R²), correlation coefficient (r), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE).

Main Results:

• The Pearson Universal Kernel (PUK) based GPR model achieved high accuracy in predicting RFM shear strength.
• During the training phase, the PUK model yielded R² = 0.9806 and r = 0.9903.
• In the testing phase, the PUK model maintained strong performance with R² = 0.9455 and r = 0.9724, demonstrating its robustness.

Conclusions:

• The GPR-PUK model is a feasible and effective tool for predicting the shear strength of rockfill materials.
• The model's accurate predictions support safer and more cost-effective designs for embankment dams.
• This approach offers a promising alternative to traditional, often costly, methods of RFM characterization.