Response surface guided adaptive slope reliability analysis in spatially varying soils

Response surface methods are widely used in slope reliability analysis due to the high efficiency. However, they are often criticized of the curse of high dimensionality, problem-dependent accuracy, and mechanism-free interpretation. To address these issues, this study proposes a response surface-guided adaptive slope reliability analysis method for slopes in spatially varying soils. A sparse quadratic polynomial is initialized using a two-fold dimensionality reduction technique. The response surface-based preliminary slope reliability analysis is successfully corrected to an unbiased, finite element-based target slope reliability analysis, during which the response surface updates iteratively with an increasing high accuracy from special emphasis nearby the failure domain. Two spatially varying soil slopes are investigated. Results indicate that the proposed method can provide an efficient and unbiased estimation of finite element-based target slope failure probability with a low variability, develop a more accurate and problem-oriented response surface model for slope reliability analysis, and explore failure mechanisms of slope stability in spatially varying soils. Its performance is insensitive to the accuracy of initial response surface. Additionally, it prominently outperforms both response surface and finite element-based subset simulations, regardless of high-dimensional or low-failure-probability problems. These advantages significantly enhance the application of response surface on practical slope reliability analysis.