Reducing shear strength uncertainties in clays by multivariate correlations

Abstract

Quantifications of uncertainties in soil shear strengths, including undrained shear strength of clay, are essential for geotechnical reliability-based design. In particular, how to reduce the uncertainties in undrained shear strengths based on all available information by correlation is a practical research subject, given the considerable cost of a typical site investigation. Although it is simple to reduce the uncertainties by correlation when the information is one dimensional (or univariate), it is quite challenging to reduce the uncertainties by using multivariate information through multiple correlations. This study proposes a systematic way of achieving multivariate correlations on undrained shear strengths. A set of simplified equations are obtained through Bayesian analysis for the purpose of reducing uncertainties: the inputs to the equations are the results of in situ or laboratory tests and the outputs are the updated mean values and coefficients of variation (c.o.v.s) of the undrained shear strengths. Two case studies are used to demonstrate the consistency of the proposed simplified equations. Results show that uncertainties in undrained shear strengths can be effectively reduced by incorporating multivariate information. Given that reliability-based design can justify more economical design with reduced uncertainties, the proposed equations essentially link the value of more and better tests directly to final design savings. © 2010 NRC Canada.