Probabilistic assessment of reinforced soil wall performance using response surface method

Abstract

The paper demonstrates the use of the response surface method (RSM) to carry out probabilistic assessment of selected performance features of geosynthetic-reinforced segmental retaining walls under operational conditions. The method substantially reduces the number of Monte Carlo simulations required to carry out probabilistic analysis of numerical models with a large number of problem parameters. A numerical model is verified using performance data from three physical full-scale laboratory walls, and is then used to generate a large synthetic database of numerical results for maximum wall facing deformation, maximum reinforcement connection strains and maximum reinforcement strains in the backfill. Closed-form solutions (performance functions) are formulated using a full quadratic polynomial, which is a common feature of the RSM. The coefficient values of the closed-form solutions are found using a least squares method to give good agreement between the RSM equations and numerical results. The RSM equations for each performance function are used to carry out Monte Carlo simulations and the results presented as cumulative distribution functions. The resulting cumulative distribution functions can be used for quantitative reliability-based assessment of wall facing deformations and reinforcement strains under operational conditions for walls matching the physical test walls used in the initial numerical model verification.