Machine Learning Applications for Site Characterization Based on CPT Data

Abstract

In this study, supervised machine learning techniques and more specifically, artificial neural network (ANN) models based on multi-layer perceptron (MLP) algorithm have been adopted for soil stratigraphy interpretation and estimation of soil shear wave velocity profiles based on CPT data. Supervised machine learning algorithms analyze a set of labeled training data consisting of a set of input data and desired output values, and produce an inferred function which can be used for predictions from given input data. For soil stratigraphy interpretation, supervised learning ANN using classification techniques were trained using backpropagation based on selected project-specific CPTs for which soil stratigraphy was manually interpreted. Similarly, for prediction of soil shear wave velocity profiles, supervised learning ANN using regression techniques were trained using direct soil shear wave velocity measurements from selected project-specific seismic CPTs. This application was implemented for a large project in the Netherlands involving the design of levees extending approximately 15 kilometers. The available data included approximately 300 raw CPTs and 11 seismic CPTs. Integrating machine learning techniques in the interpretation of CPT data allowed for reliable and rapid interpretation of subsurface conditions and estimation of soil properties for design.