The ability of granular materials to retain fine particles transported within their void space by seepage flow depends strongly on the geometric characteristics of their pore network (pore sizes and constriction sizes). Hence, characterizing the pore network of a granular assembly obtained by means of a micro-tomography scanning or a numerical discrete simulation is of great importance in assessment of its filtration efficiency. Here, we determine characteristics of the pore network of virtual samples composed of spherical particles simulated by using the DEM. A new criterion is proposed to merge neighboring tetrahedra issued from the weighted Delaunay triangulation. To do so, we extend the concept of inscribed void sphere, initially defined for each tetrahedron, to each polyhedral sub-domain constituted of merged tetrahedra. This inscribed void sphere fits the best the void space within the sub-domain. Flat tetrahedra are first eliminated by a primary merging procedure taking into account two basic geometric conditions required for each pore. Adjacent sub-domains are then merged depending on the level of overlap between their inscribed void spheres. The pore size distributions and constriction size distributions (CSD) of granular samples with different grain size distributions obtained with the new merging criterion are compared to those given by two other criteria often used in the literature. The new criterion allows us to reduce greatly the inherent subjectivity in characterizing the granular pore network and to remediate the drawbacks of the two considered criteria in the literature. Moreover, CSDs given by these different criteria tend to converge for gap-graded and widely graded materials. The CSDs obtained with the new merging criterion are used to estimate the controlling constriction sizes \(D_c^*\) of the considered samples, and the estimated values of \(D_c^*\) are compared to Kenney and Lau’s empirical rule.

粒状材料保留通过渗流在其空隙空间中运输的细颗粒的能力在很大程度上取决于其孔隙网络的几何特征（孔尺寸和收缩尺寸）。因此，表征通过微断层扫描或数值离散模拟获得的粒状组件的孔网络对于评估其过滤效率非常重要。在这里，我们确定使用DEM模拟的由球形颗粒组成的虚拟样品的孔网络的特征。提出了一种新的准则来合并从加权Delaunay三角剖分中发出的相邻四面体。为此，我们将最初为每个四面体定义的内接空球的概念扩展到由合并的四面体组成的每个多面体子域。这个内接的空球最适合子域内的空空间。考虑到每个孔所需的两个基本几何条件，首先通过一次合并程序消除了平四面体。然后根据相邻的子域内接的空球之间的重叠程度进行合并。将通过新的合并标准获得的具有不同粒度分布的颗粒状样品的孔径分布和收缩尺寸分布（CSD）与文献中经常使用的其他两个标准进行比较。新标准允许我们大大降低表征颗粒孔网络的固有主观性，并弥补了文献中两个已考虑标准的缺陷。而且，由这些不同标准给出的CSD趋于收敛于间隙分级和广泛分级的材料。使用新的合并标准获得的CSD用于估计控制收缩量\（D_c ^ * \）所考虑的样品中，和的估计值\（D_c ^ * \）进行比较，以肯尼和Lau的经验法则。