Slope reliability analysis considering effect of autocorrelation functions

The autocorrelation function (ACF) is a prerequisite for properly characterizing the spatial variability of soil properties. The effect of different types of ACFs on the slope reliability has not been qualitatively evaluated. A procedure for simulating correlated non-Gaussian random fields based on the Cholesky decomposition technique with midpoint discretization is proposed. The typical ranges of autocorrelation length of shear strength parameters of soils are summarized. An example of reliability analysis of frictional/cohesive soil slope is then presented to investigate the effect of five common types of ACFs for the geostatistical analysis on the slope reliability. The influence of cross-correlation, variability of soil properties and different scales of fluctuation of the cohesion and friction angle are taken into account, respectively. The results indicate that the proposed method is computationally simple and easily implementable for simulating the correlated non-Gaussian random fields with any geometry, and it can effectively evaluate the slope reliability with a sufficient accuracy. The differences in the slope reliability underlying five types of ACFs are more obvious when the negative cross-correlation and vertical scales of fluctuation become stronger, and the variability of soil properties becomes smaller, respectively. These differences become very significant when the cohesion and friction angle take different vertical scales of fluctuation. Additionally, the target random fields are very smooth and the slope reliability is underestimated underlying the square exponential, second-order autoregressive or cosine exponential ACFs, and they may account for the spatially correlated soil properties more realistically. In contrast, the target random field is a roughly varying field and the slope reliability is overestimated underlying the exponential ACF.