Applying machine learning for soil classification in geotechnical engineering based on laboratory soil test data

Abstract

An engineering geological unit/soil layer is a homogeneous volume of soil and rock with the same name and physical and mechanical characteristics that vary without regularity, or if the physical and mechanical characteristics vary with a regularity then this regularity can be ignored when statistical conditions are satisfied. Classificating soil layers is very important to provide reliable data for calculating, designing the foundations, and choosing reasonable construction solutions that contribute to saving the costs of building constructions. Currently, the classification of soil layers is often carried out manually based on soil classification standards of Vietnam (TCVN), America (ASTM), Britain (BS), etc. Because the work is carried out manually, errors are inevitable, especially when working with big data sources. To reduce errors when classification of soil layers as well as detect anomalies in the soil stratigraphy and save time when synthesizing large amounts of data to build a geotechnical database for a region, this article aims to apply Machine learning to automatically classify soil layers based on the use of three clustering algorithms such as K-Means, Gaussian Mixture Model (GMM), and Mean Shift. The input data set is taken from the results of soil testing of 437 soil samples taken from District 1 and District 8 in Ho Chi Minh City. The results show that the automatic classification program gives results of soil layers that closely match the results of the manual method. Furthermore, the automatic program can divide soil stratigraphy into very detailed units, which helps detect anomalies in geotechnical engineering.