Abstract
Void ratio is a critical soil characteristic directly describing the degree of sand compaction. Though several empirical models for void ratio prediction were available in the literature, the prediction accuracy was restricted by various soil types and complex soil conditions. The existing literature unanimously agreed on the strong correlation between the shear-wave velocity and void ratio, but the former was rarely used for the prediction of the latter. In this study, a machine learning-based framework was developed using the XGBOOST model with shear-wave velocity as a major input to improve prediction accuracy. The XGBOOST model had its hyperparameters tuned by Bayesian Optimization and was trained with the database collected from dynamic laboratory tests. The comparison results demonstrated the superiority of the machine learning-based method over the existing empirical models in terms of prediction accuracy. Feature importance analysis also revealed the prediction of void ratio had a high sensitivity to the existence of shear-wave velocity. This study evidenced the validity of including shear-wave velocity as a major input for improving the prediction accuracy of void ratio. In addition, using shear-wave velocity as major input, the XGBOOST model enabled the investigation of the void ratio by non-intrusive field tests.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have