Numerical Sensitivity Study Compared to Trend of Experiments for LEAP-UCD-2017

Abstract

This paper describes the numerical sensitivity study requested prior to the December 2017 LEAP workshop. Several but not all of the simulation teams participated in this sensitivity study. The results of the sensitivity study are used to begin to map out the simulation response surfaces that relate residual displacement to PGAeff and relative density. The simulation response surfaces are compared to the corresponding response surfaces determined by nonlinear regression of the centrifuge test data. The definition of the experimental response surface allows a means to objectively reduce the influence of outliers in the experiment dataset. The residuals between the experiments and the regression surface are used to quantify the uncertainty associated with experiment-experiment variability. Some metrics for assessing the comparison between simulations and experiments are explored; it is suggested that differences in the logarithm of displacement are more meaningful than arithmetic differences. As expected, some models predicted the average displacement well and some predicted triggering of liquefaction and the shape of the response function better than others. LEAP-UCD-2017 is not a final assessment of simulation procedures; instead, the results can be used to improve simulation specifications and calibration procedures and as a stimulus for more careful review of simulation results before they are submitted.