Application of kernel k-means and kernel x-means clustering to obtain soil classes from cone penetration test data

L.O Carvalho,D.B Ribeiro

doi:10.28927/sr.434607

Abstract

Application of kernel k-means and kernel x-means clustering to obtain soil classes from cone penetration test data

Highlights

The more commonly used soil classification standard is the Unified Soil Classification System, which is based on granulometry and plasticity
To produce the results presented within this section, the kernel k-means algorithm was applied. k was varied from 7 to 10, using the Full dataset and all cone penetration test (CPT) original measurements: z (m), qc (MPa), fs and u2
Each value is a percentage of soil samples that were assigned to a clustering class and to a specific Influenced by soil granulometry (ISG) or Focused on soil behavior (FSB) class

Summary

Introduction

The more commonly used soil classification standard is the Unified Soil Classification System, which is based on granulometry and plasticity. It has disadvantages like the difficulty of extracting undisturbed samples and the time delay required to get the results. One important issue concerning this classification is its connection to soil behavior in detriment of soil granulometry. In this context, pioneer work proposing soil classification from CPT data focused only soil granulometry (Begemann, 1965), following studies stated that soil behavior should guide class definitions for being related to the soil load-bearing capacity (Douglas & Olsen, 1981). A new friction ratio-based chart was later proposed, changing the circular curves of Robertson (1990) by hyperbolic ones (Schneider et al, 2012). Robertson (2016) modified these charts, defining a fully behavioral classification, including the dilative and contractive behaviors for each of the three soil types

Objectives

Methods

Results

Conclusion