Uncertainty Quantification of Steady-State Seepage Through Earth-fill Dams by Random Finite Element Method and Multivariate Adaptive Regression Splines

Abstract

This paper aims to investigate the effects of uncertainty in soil characteristics and dam geometry on seepage flow using the hybrid Multivariate Adaptive Regression Splines (MARS) and Monte Carlo Method (MCM). A computer program based on Darcy flow is developed in the Fortran language to calculate the discharge flow. After validating the numerical FORTRAN code with experimental outputs, firstly, the Deterministic Finite Element Method (DFEM) was used to obtain Seepage Exit Discharge (SED) in Steady State Condition (SSC), and MCM was used for probabilistic analysis to account for uncertainty in random parameters. The program monitored Pore Water Pressure (PWP) changes and integrated them into the time/space domains. To ensure minimal error, the results of the models were compared by Standard Error Calculation (SEC). The research also introduced a new component to compare the seepage flow resulting from the analysis of models in a dimensionless manner called the Effective Discharge MARSplines (EDM). In the present research, the combination of Machine Learning (ML) and MCM algorithms was used in an innovative way for Random Finite Element Method (RFEM) calculations. The results of the research indicate that a 17.9% increase in the Hd/Hu ratio in the deterministic analysis results in a 29.3% decrease in EDM, while in the probabilistic analysis, a similar increase leads to a 19.02% decrease in EDM. Upon comparing deterministic and stochastic models, it can be concluded that deterministic analysis is more accurate and exhibits less error when compared to the probabilistic model.

Keywords

• fortran programming
• seepage exit discharge (sed)
• two-phase soil
• stochastic analysis
• computational fluid dynamics