Anisotropic Materials in 3D by Geometrically Informed Learning

The world is in 3D and almost all materials are 3D with their properties determined by their structure. Therefore, efficient 3D structural characterization has emerged as a crucial need in several scientific areas ranging from materials science to biology. X-ray 3D imaging provides a suitable approach and has developed immensely over the past few years. A major bottleneck, that very strongly limits the applicability of such methods, is the analysis of 3D imaging data. This is especially so for materials with anisotropic structural elements such as fiber composites or channels for liquid flow. To understand and model such materials, accurate identification of the fiber/channel network is essential but current methods are challenged by noise/limited resolution. Therefore, we propose to develop and implement deep learning methods incorporating topological/morphological considerations to enable accurate and efficient analysis of advanced materials. We will also generate benchmark data, both synthetic and experimental ones, with which to test such methods. Additionally, we will apply the methods to important example applications: liquid flow in bone and fiber composite materials relevant to e.g. sustainable energy production.