Eikonal-Based Models of Random Tessellations

Abstract

In this article, we propose a novel, efficient method for computing a random tessellation from its vectorial representation at each voxel of a discretized domain. This method is based upon the resolution of the Eikonal equation and has a complexity in O(N log N), N being the number of voxels used to discretize the domain. By contrast, evaluating the implicit functions of the vectorial representation at each voxel location has a complexity of O(N²) in the general case. The method also enables us to consider the generation of tessellations with rough interfaces between cells by simulating the growth of the germs on a domain where the velocity varies locally. This aspect constitutes the main contribution of the article. A final contribution is the development of an algorithm for estimating the multi-scale tortuosity of the boundaries of the tessellation cells. The algorithm computes the tortuosity of the boundary at several scales by iteratively deforming the boundary until it becomes a straight line. Using this algorithm, we demonstrate that depending on the local velocity model, it is possible to control the roughness amplitude of the cells boundaries.

Electrical Conductivity of Metal–Polymer Cold Spray Composite Coatings onto Carbon Fiber-Reinforced Polymer

“…For each nucleus, a cell propagates in 3D through the aggregate. The cell growth is simulated using the Eikonal equation [26]. When two cells meet, they stop growing and their contact surface is transformed into a volume filled with PEEK with a randomly generated thickness sampled from an exponential law.…”

A supervised algorithm entirely trained on a synthetic dataset to segment granular suspension images

“…In addition, the annotation itself is error prone due to the inevitable loss of attention of the operator in charge of annotating the images. The main originality of our approach is that we entirely trained the network on a series of synthetic images generated with morphological models (Figliuzzi, 2019;Figliuzzi et al, 2021;Jeulin, 2021;Stoyan et al, 2013), commonly used to simulate micro-structures in materials engineering (Bortolussi et al, 2018;Figliuzzi et al, 2016), rather than on images of real experiments. With our approach of artificial synthesis of the ground truth, the positions and sizes of the particles are known by construction, which allows us to get rid of the difficulty of obtaining a reliable learning database 1 .…”

A Bayesian Approach to Morphological Models Characterization

“…Materials used in industry often present a complex internal microstructure, which largely determines most of their physical properties at the macroscopic level (Jeulin, 1991;Ohser, 2009;Torquato, 2013;Moussaoui, 2018;2019;Figliuzzi, 2019). A common way to carry out a quantitative study of the microstructure influence on the macroscopic properties of materials is to generate random microstructures that reproduce their geometrical characteristics (Moreaud, 2012;Wang, 2015).…”