Reliability analysis of seepage stability of core-wall rockfill dam based on stochastic response surface method

The finite element method and stochastic response surface method are combined to analyze the reliability of seepage stability in Shuangjiangkou core-wall rockfill dam project. The seepage flow analysis of the core-wall rockfill dam and the reliability analysis can be conducted separately within the framework of reliability analysis based on the stochastic response surface method. The nodes with the larger hydraulic gradient are selected to establish seepage stability function using the finite element method and stochastic response surface method. The failure probabilities of seepage failure at each node are then calculated and the maximum failure probability is taken as the failure probability of the core-wall. The relationship between the hydraulic gradient of the nodes with maximum failure probability and the hydraulic conductivity of the core-wall and the alluvial deposits, the relationship between the hydraulic gradient of the nodes with maximum failure probability and the upper water level, and the effects of the hydraulic conductivity of core and the variation of upper water level on failure probability of seepage failure of the core-wall are analyzed. The results show that third order Hermite expansion is able to ensure good precision with acceptable time consumption. The seepage at maximum failure probability node in Shuangjiangkou core-wall dam is closely related to upper water level, but has a weak negative correlation with the hydraulic conductivity of the core-wall and is less significant with the hydraulic conductivity of alluvial deposits. On the other hand, as the coefficient of variation of the upper water level increasing, the failure probability increases drastically, but this effect is less significant with regard to the hydraulic conductivity of the core-wall. The results provide a further evidence for readily application of the stochastic response surface method to practical engineering.