Using Machine Learning to Predict Shear Wave Velocity

Cone penetration testing (CPT) is widely used for geotechnical characterization of subsurface materials. Shear wave velocity (Vs) measurements provide very useful information for engineering purposes and can usually be added to land-based CPT testing programs for a relatively minor increase in cost and time. However, performing Vs measurements from a floating platform on submerged materials such as impounded tailings and coal ash is a challenge due to the lack of available equipment required to generate shear waves from the mudline. To solve this problem, the authors developed a new model using machine learning techniques to estimate Vs based on conventional CPT measurements without direct Vs measurements. The model uses a random forest algorithm implemented in Python for solving this regression problem. The new model is shown to provide more accurate estimates of Vs than methods widely used currently. The technique also provides for the possibility of generating continuous Vs profiles with greater vertical resolution than currently obtained using discrete measurements.