关键词: 粗粒土;颗粒形状;密实度;孔隙比;形状指标;离散元;神经网络;

Abstract: Compactness is one of the most important criteria to measure the mechanical behaviors of coarse-grained soils. Abundant experiments show that the grain morphology has a significant influence on the compactness of granular materials, but few researchers focus on the effects of specific shape descriptors. In this paper, the pebble grains and block grains of Dashixia rockfill dam were selected as the materials, and the morphology of the grains was obtained by 3D scanning. Then we calculated the shape descriptors of grains. Based on the morphology, a discrete element cluster model was established to simulate the coarse-grained soil grains. The grain aggregate was prepared by using the compression boundary method, and the void ratios of all samples were measured. Parameter sensitivity analysis combined with BP neural network and Olden method showed that the shape descriptors with strong sensitivity to compactness were three-dimensional sphericity, convex, three-dimensional roundness and two-dimensional roundness in turn. According to normal distribution of the three-dimensional sphericity of coarse-grained soil grains, a numerical simulation was carried out considering different normal distributions of the three-dimensional sphericity. It showed that the void ratios of the samples basically remained unchanged with the increasing of distribution range of three-dimensional sphericity, and the change of the distribution range had little effect on the void ratio.

Key words: coarse-grained soil; grain morphology; compactness; void ratio; shape descriptor; DEM; neural network