Reliability Analysis of Slope Stability Using Censored Samples and Genetic Algorithm

In geotechnical slope stability analysis, soil properties have a great extent of randomness and thus are extremely amenable to comprehensive probabilistic treatment. The randomness of soil parameters is involved in slope analysis using probability distributions. Due to technical and economic reasons, it is sometimes difficult to collect complete information of soil properties in full or non-censored domain. In such special cases, the analysis of slope stability needs to be conducted with truncated random variables, which are characterized by truncated probability distributions. The contributions of the extended abstract include twofold: Firstly, censored samples are used to determine truncated probability distributions based on the principle of maximum entropy and Akaike information criteria, which is the least unbiased as it is derived from a systematic maximization. Secondly, the truncated probability distributions obtained are implemented in a first order reliability method to calculate the probability of failure or reliability index of geotechnical slopes. Genetic algorithm is used to tackle the problem of convergence that arises in the usual iterative algorithm of the first-order reliability method when one or more of the random variables are described by truncated probability functions. The application of the proposed method is demonstrated by performing reliability analysis of a slope with censored samples to determine the probability of failure.