Measurement-While-Drilling Based Estimation of Dynamic Penetrometer Values Using Decision Trees and Random Forests

Abstract

Machine learning is a branch of artificial intelligence (AI) that consists of the application of various algorithms to obtain information from large data sets. These algorithms are especially useful to solve nonlinear problems that appear frequently in some engineering fields. Geotechnical engineering presents situations with complex relationships of multiple variables, making it an ideal field for the application of machine learning techniques. Thus, these techniques have already been applied with a certain degree of success to determine such things as soil parameters, admissible load, settlement, or slope stability. Moreover, dynamic penetrometers are a very common type of test in geotechnical studies, and, in many cases, they are used to design the foundation solution. In addition, its continuous nature allows us to know the variations of the terrain profile. The objective of this study was to correlate the drilling parameters of deep foundation machinery (Measurement-While-Drilling, MWD) with the number of blows of the dynamic penetrometer test. Therefore, the drilling logs could be equated with said tests, providing information that can be easily interpreted by a geotechnical engineer and that would allow the validation of the design hypotheses. Decision trees and random forest algorithms have been used for this purpose. The ability of these algorithms to replicate the complex relationships between drilling parameters and terrain characteristics has allowed obtaining a reliable reproduction of the penetrometric profile of the traversed soil.