Characterization of model uncertainties for drilled shafts under undrained axial loading

Abstract

This invited paper presents a rigorous examination of model statistics for drilled shafts under undrained axial loading. To apply a model factor as an independent random variable on the predictive capacity equation, it is desirable to verify that it is not varying systematically with some underlying factors. The measured and computed capacities appear as the numerator and denominator, respectively, in the standard definition of the model factor and are the most obvious factors to check for such statistical dependencies. The Spearman rank correlation is a simple and useful method to perform this check. Even though the adhesion factor typically is expressed as a function of the undrained shear strength, this approach only is marginally acceptable from a rank correlation perspective. For undrained compression tip resistance, a generalized model factor is recommended because the existing definition produces strong statistical dependencies with the measured and computed alues.