Slope reliability analysis in spatially variable soils using sliced inverse regression-based multivariate adaptive regression spline

Reliability analysis of slope considering the spatial variability of soil properties may be subjected to the curse of high dimensionality, which leads to the traditional slope reliability analysis method cannot effectively carry out. This paper aims to propose a sliced inverse regression (SIR)-based multivariate adaptive regression spline (MARS) method for slope reliability analysis in spatially variable soils, which combines the advantages of both SIR and MARS. First, the Karhunen-Loeve (K-L) expansion is adopted to simulate the spatial variability of soil properties. Second, the slope reliability analysis based on the SIR-MARS method is proposed. Thereafter, the implementation procedure for slope reliability evaluation in spatially variable soils using the proposed method is summarized. The validity of the proposed method is illustrated with a single-layered c-phi slope and a two-layered c-phi slope. The results indicate that, in the case of higher dimensions of random variables, the MARS model with the aid of SIR can effectively establish the relationship between soil shear strength parameters of slopes in spatially variable soils and safety factor (FS). Moreover, the proposed method can obtain sufficiently accurate reliability results for both single-layer and two-layer slopes in spatially variable soils with a low computational cost. The proposed method provides an effective and practical way to solve the reliability problem of high dimensional spatial variation slope.

Automated summary

This paper proposes a reliable analysis method for slopes based on the SIR-MARS method, which effectively solves the high dimensionality problem under spatially variable soils. By simulating the spatial variability of soil properties and establishing the relationship between soil shear strength parameters and safety factor, the method obtains accurate reliability results at a low cost.