Probabilistic Characterization of Site-Specific Inherent Variability of Undrained Shear Strength Using Both Indirect and Direct Measurements

Abstract

Determining geotechnical design parameters (e.g., undrained shear strength Su) is an essential step in the analysis and design of geotechnical structures at a particular site. The estimated design parameters are affected by the actual variability (i.e., inherent variability) of design parameters concerned, measurement errors due to imperfections in the measuring process, and transformation uncertainty associated with the transformation model linking measured parameters [e.g., standard penetration test (SPT) data] to design parameters (e.g., Su). These uncertainties are lumped together as the total variability of design parameters. It is the actual variability resulting from natural geological factors, not the total variability, that directly affects the underlying mechanism and actual response of geotechnical structures. Proper characterization of site-specific geotechnical inherent variability is, hence, pivotal to accurately estimating the actual responses of geotechnical structures at a site. This paper develops a Bayesian sequential updating (BSU) approach for probabilistic characterization of site-specific inherent variability of Su of clay. The proposed BSU approach systematically combines prior knowledge (e.g., engineering judgment and experience) and site-specific information from both indirect (e.g., SPT data) and direct (e.g., Su values from triaxial tests) measurements to inversely infer the inherent variability of Su at a particular site. It accounts, explicitly and quantitatively, for effects of measurement errors and transformation uncertainty on the probabilistic characterization of site-specific inherent variability of Su. The proposed approach is illustrated and validated using real-life and simulated data. It is shown that the proposed approach provides proper probabilistic characterization of site-specific inherent variability of Su based on available information from multiple sources. Sensitivity studies are also performed to explore effects of measurement errors on the performance of the proposed approach.