Application of the soil parameter random field in the 3D random finite element analysis of Guanyinyan composite dam

Due to the inhomogeneous of dam materials, effect of the compaction environment and experi-mental data errors, the soil properties of the Guanyinyan composite dam present obvious spatial variability, resulting in the occurrence of obvious discordant settlements and cracks during the impounding period. The traditional finite element method (FEM) using inversion parameters that are modified by monitoring displacement back-analysis can neither simulate spatial variability in soil parameters nor predict dam safety before dam displacements occur. In this research, soil parameter random fields are developed considering spatial variability, and a three-dimensional random FEM (RFEM) analysis of the Guanyinyan composite dam is applied. According to soil dry density samples and experimental results, the autocorrelation of the same soil parameter in different positions in dam is determined. The cross-autocorrelation of different soil parameters is conducted using the Nataf transformation. The RFEM analysis shows that the mean values of maximum settlement, the mean values of maximum deformation gradient and area ratio with high hydraulic fracture probability exceeding 0.3 are close to those calculated by inversion pa-rameters. The cumulative frequencies of maximum settlement and maximum deformation gradient show 51% and 33% probabilities of exceeding the inversion calculation results. The area ratio and distribution range of elements with hydraulic failure probabilities greater than 0.2 are significantly larger than those calculated by the inversion parameters. The traditional certain security indexes, which ignore spatial variability of soil parameters, have a certain probability to overestimate the security of the dam. The uncertain security index considering spatial variability of soil parameters can provide a more reasonable and accurate reference for dam safety evalu-ation before water impounding.