AI-Supported probabilistic steady-state and transient seepage analysis of an embankment dam with clay core: A case study on Büyükcekmece Dam

Abstract

The integration of the finite element method with probabilistic approaches accounting for uncertainties has become a widely adopted strategy for accurately modelling seepage (Q) in embankment dams. However, this approach is time-intensive, prompting this study to conduct a probabilistic analysis encompassing both steady-state and transient seepage to demonstrate the efficiency of using artificial neural networks (ANN) for accurately estimating Q values, thereby significantly reducing the reliance on extensive finite element simulations. In this context, first, the variability in the permeability of the clay core of embankment dams was investigated. Permeability was modeled as a random variable using two different approaches: First, as a log-normally distributed random variable for the entire clay core; and second, using random field theory, as a spatially correlated log-normally distributed random variable in horizontal layers. The data obtained were analyzed with artificial neural networks (ANN) and artificial bee colony (ABC) to ensure the consistency of the analyses. In all steady-state analyses, the lowest discharge was obtained when spatial variability was considered. Finally, the trained ANN-ABC network was utilized to estimate Q values for 0.75 COV (k_s), a scenario for which finite element analyses were not conducted. This approach demonstrated the efficacy of using ANN to obtain Q values efficiently, thereby eliminating the need for labour-intensive finite element analyses.