Abstract: When determining the parameters of stone column composite foundation, the simplified method of superposition of gravel and undisturbed soil parameters according to the proportion of plane area is often used in engineering, but the application conditions are seldom focused on. Based on the soft foundation treatment project for the deep overburden layer of the high earth-rockfill dam of the Nyabarongo II Hydropower Station in Rwanda, the triaxial consolidated-drained test numerical simulation of composite soft clay samples with gravel core under different area replacement rates was carried out using the PLAXIS finite element platform, and the rationalization of the numerical simulation scheme was verified by laboratory triaxial tests. The pile-soil interaction mechanism of stone column improved soft soil composite foundation was analyzed and the calculation parameters of hardening soil model were obtained. These parameters were applied to analyze the deformations of the dam foundation, compared with the traditional parameter- superposition method and the stone-column-wall method. The study shows that the parameters determined by the numerical composite sample method are reasonable and the error is negligible to analyze the settlement of the composite foundation of the earth-rock dam. However, the traditional parameter-superposition method underestimates the settlement of composite foundation, and it is only suitable for the conditions of low stress level and high area replacement rate, and will also overestimate the strength parameters of the stone column improved soft soil composite foundation. The two-dimensional finite element analysis of dam foundation deformation using the parameters of numerical composite sample method shows that it is feasible to adopt different replacement rates of stone columns to reinforce the foundation according to the different heights of the dam.

Key words: stone column, soft soil foundation, triaxial test, finite element, parameter-superposition method, hardening soil model