ABSTRACT
Quantification of randomnesses in soil properties is the first step in probabilistic reliability analysis and design in civil engineering, of which the accuracy has a critical impact on the final probability of failure. Available procedures to determine probabilistic distributions are restricted in the family of presumed classical distributions, such as lognormal and normal distributions, and disputes exist over which distribution fits a sample data best. The estimation of distribution parameters can rarely account for information of higher order moments. This chapter presents a distribution-free approach to quantify the randomness of soil properties using a method that combines the maximum entropy principle (MEP) and the Akaike information criterion (AIC). The MEP is used to infer a series of maximum entropy probabilistic distributions from sample moments, and the optimal order of the distributions is determined by AIC. The implementation of the algorithm is described in detail and the procedure of the methodology is provided in a flowchart. The entropy distribution is obtained without any classical distributions presumed. A case study is presented for the undrained shear strength of soil in Nipigon River area, Ontario, Canada. An unbiased probabilistic distribution for the soil property was obtained based on optimal-order moments of soil samples from the vane shear test. Comparative studies show that the entropy distribution obtained can accurately quantify the inherent probabilistic uncertainties ubiquitous in soil properties.
