Predicting Shear Wave Velocity from Cone Penetration Test Data Using ‎Machine Learning: A Case Study on Sensitive Soft lacustrine Clays

1. Home
2. Publications
3. Online Library
4. Predicting Shear Wave Velocity from Cone Penetration Test Data Using ‎Machine Learning: A Case Study on Sensitive Soft lacustrine Clays

Shear wave velocity (V_s) is a vital parameter for assessing subsurface conditions, soil stiffness, and seismic site response. Despite the comparably rapid testing technique and data recording, an accurate estimation of V_s is challenging due to the complexity and time-consuming nature for determining the arrival time of the shear wave in field and laboratory testing. Field measurements of V_s capture the soil's intact condition through seismic testing, whereas laboratory measurements are often less representative due to sample disturbances, especially in sensitive soft clays. Traditional empirical correlations between cone penetration tests with pore pressure measurement (CPTu) data and V_s often show variability due to site-specific conditions and soil heterogeneity. This study investigates the potential of machine learning (ML) models to enhance the accuracy of predicted values of V_s from CPTu data, providing a robust alternative to commonly accepted correlations. A comprehensive dataset was compiled from a sensitive soft clay deposit, including CPTu measurements as input data and corresponding V_s from cross-hole seismic testing (CH). ML algorithms, specifically Random Forests (RFs) and Multivariate Linear Regression (MVLR), were used to train predictive models from this dataset. The predictive capabilities of the ML models were rigorously compared with established empirical CPTu-V_s correlations. This study examines the interdependency of input parameters and selects the most relevant ones using performance metrics. Preliminary results show that ML models significantly outperform traditional CPTu-V_s correlations, exhibiting lower prediction errors and higher consistency across diverse datasets. The presented ML models and their accuracy can be particularly applied for challenging conditions such as offshore environments where in addition to the time-consuming work to detect V_s, expensive operations and technical limitations are to be considered. In conclusion, this study demonstrates the potential of ML to provide highly adaptable and precise method for predicting V_s from CPTu data, addressing the limitations of conventional approaches.

5th International Symposium on Frontiers in Offshore Geotechnics (ISFOG2025)

5 - Data Analytics and Machine Learning