Characterization of Model Uncertainty for Cantilever Deflections in Undrained Clay

This work presents a critical evaluation of the model factor for the mobilized strength design (MSD) method for cantilever deflection in undrained soft to medium-stiff clay. The model factor is characterized in two parts. A correction factor $\eta$ is first defined as the ratio between the wall-top deflection computed from the FEM and its corresponding value computed from the MSD. The hardening soil with a small-strain model is used in FEM. The statistics of $\eta$ are evaluated using 82 numerical simulations. Because $\eta$ is not random, and therefore cannot be modeled directly as a random variable, it is decomposed as a product of a systematic part ($f$) and a random part (${\eta }^{\ast }$). The random part is modeled as a lognormal random variable with a mean of 1.01 and a coefficient of variation (COV) of 0.18. The model factor for FEM (${\epsilon }_{\mathrm{FEM}}$) is next defined as the ratio between the measured wall-top deflection and the corresponding FEM result. The statistics of ${\epsilon }_{\mathrm{FEM}}$ are evaluated using 45 field cases. The ratio ${\epsilon }_{\mathrm{FEM}}$ is random and can be modeled directly as a lognormal random variable with a mean of 1.01 and a COV of 0.21. The model factor for MSD (${\epsilon }_{\mathrm{MSD}}$) is finally derived from the product of $\eta$ and ${\epsilon }_{\mathrm{FEM}}$. This proposed approach is validated by 14 centrifuge tests. The importance of including the proposed model factor ${\epsilon }_{\mathrm{MSD}}$ in reliability analysis is illustrated using a field example.