Data-driven collaborative safety evaluation for seepage reliability of embankments considering spatial variability

Abstract

Seepage safety analysis in real embankments considering hydraulic parameters’ spatial variability is crucial. However, random seepage analysis in embankments is often lack of efficiency and accuracy for evaluation of low failure incidents based on complex random fields and fine finite element models. To improve both analytical efficiency and accuracy, a data-driven collaborative safety evaluation framework that integrates the Optimized Linear Estimation Method (OLEM) with Refine Subset Simulation (RSS) has been proposed. Firstly, the random field of the uncertain parameters of the soil–water characteristic curve (SWCC) is efficiently discretized using OLEM based on the results of deterministic analysis, and a sensitivity analysis is performed accordingly. Then, considering various combinations of spatial variability degrees of hydraulic parameters, RSS is applied to the rough mesh model for random finite element analysis, while fine element cases are generated using the Response Conditioning Method (RCM) for further collaborative safety evaluation. Finally, the accuracy of the method is validated by comparing it with the Monte Carlo Simulation (MCS) based on the same 3D rough model, and the relevant seepage safety analysis is conducted. The findings reveal that the coefficient of variation (COV) and autocorrelation distance of the permeability coefficient Ks in the filling section significantly impact the dam’s seepage safety. Despite strong spatial variability in parameters, the probability of seepage failure remains below 10^-3. The RSS method reduces computation time to 1/10 compared to MCS at a 10^-3 failure level.