Generation of Artificial Images of Cross Sections of WC/Co Composite Alloys Using Diffusion Networks

The study of statistical properties of microstructures of composite materials is carried out by analysing microphotographs of material cuts. The obtained two-dimensional structure represents cuts of composite elements, the geometrical properties of which can be studied by computer vision methods. Hundreds of micro-images can be collected from one microstructure, which allows to study statistical properties of mutual arrangement of composite elements. To increase sampling, the possibility of creating artificial microstructures using diffusion neural networks is being investigated. The plausibility criterion is the comparison of corresponding distributions for real and artificial microstructures, in particular using the Wasserstein–Kantorovich–Rubinstein distance.