An efficient algorithm to create discrete element samples with predefined properties incorporating the random field theory is introduced in this paper. The algorithm considerably reduces the time needed to generate a large scale domain as only a small initial sample with dynamic packing is used. Three-dimensional anisotropic random fields are generated using the Local Average Subdivision (LAS) method accounting for the spatial variability. The random fields are then mapped on the discrete element domain and uncertain parameters of each particle are obtained from the corresponding random field cell. Triaxial tests are conducted on large soil samples with the dimensions of 1.5m × 3.0m × 1.5m comprising over 150,000 spherical particles. The normal and tangential stiffnesses of the particles are selected as random variables since they have a significant effect on the soil behavior under triaxial testing conditions. Monte Carlo simulation is implemented to analyze the probabilistic features of the output values. The results of parametric studies are also presented.
展开▼
