2019 Volume 75 Issue 4 Pages I_494-I_505
The behavior of soil-structure systems during earthquakes is commonly complicated and non-linear. One way to investigate this complex behavior is centrifuge model tests or effective stress analyses. Both techniques have been under intense investigation since 1970s, and their reliability is thought to be improving because of accumulated scientific knowledge. However, a practical process for the validation of analytical procedures has not yet been established, in particular for liquefaction phenomena, as commonly recognized among geotechnical engineering community. This may be because the validation process is conventionally carried out through a comparison with a one-time experiment and the degree of coincidence between experimental and analytical results is hard to evaluate in a quantitative way: the prediction accuracy of seismic ground response strongly depends on experimenters and analysts, and does not generally consider the variation in experimental and analytical results. To overcome the problem and improve the reliability of the prediction techniques, we carried out in this study multiple centrifuge model tests on the seismic behavior of liquefiable sloping ground, targeting a unique prototype, and subsequently performed effective stress analyses for each test result. In comparison of experimental and/or analytical results, uncertainty quantification (UQ) framework was introduced to estimate the variations and validate the computational model: an error measure called EARTH, in which a total error is given as a sum of phase, magnitude, and slope errors, was used in this study to quantify discrepancy between time histories of experimental or analysis results (e.g., response acceleration, excess pore water pressure). By directly comparing the time histories using the error measure, the variation in numerical simulations was assessed in a quantitative manner, as well as that in centrifuge experiments; this suggests that the validity of computational techniques can be evaluated in an objective way by considering both experimental and analytical variations.