Numerical simulation of DCM column-supported embankments using simplified plane-strain models

![CDATA[Three-dimensional numerical modelling based on the finite difference or finite element method used in solving geotechnical problems, which involve soil consolidation and complex material models, require large amount of computer memory and computational time. Alternatively, simplified numerical models of three-dimensional geotechnical problems can be used to predict the performance, while achieving high analysis efficiency. When simplified modelling of a column-supported embankment is concerned, two-dimensional plane-strain models can be easily adopted since an embankment is a relatively long structure compared to its width. Different methods have proposed in the literature, to convert a column-supported embankment into a two-dimensional plane-strain model. Some of the popular two-dimensional idealization methods are; conversion based on equivalent properties (EP) and conversion based on equivalent area (EA). However, no study has investigated the suitability of these simplified models to analyze deep cement mixed (DCM) column improved embankments, where yielding of DCM columns is of concern. The post yield softening behavior of DCM columns affects the load transfer mechanism and consolidation behavior. Therefore, a simplified model that can capture this behavior should be selected carefully to obtain reliable results. This study investigates the performance of a DCM column improved embankment, which experienced post yield softening behavior, using two-dimensional plane-strain models simplified based on EP and EA approaches. The numerical models were developed using ABAQUS/standard finite element program. Strain softening behavior of the DCM soil was incorporated using a special material model, which is an extended version of the Mohr Coulomb failure criterion. The simplified model results were compared with the field measured data in terms of settlements, lateral deformations and pore water pressures. The results demonstrate that the final predictions were significantly varied between the two idealization techniques.]]