Machine learning-supported visual analytics for high resolution X-ray inspection of metal matrix composites

Abstract

Metal matrix composites are utilized in a multitude of applications due to their mechanical and thermodynamical properties, which are highly dependent on the microstructure. A detailed characterization is thus vital for a sound understanding of the material’s properties. X-ray computed tomography, in particular high resolution synchrotron imaging, presents a promising inspection method for this purpose. However, a high-resolution inspection of medium-sized samples produces very large volumetric datasets, which prevents a proper data analysis with commonly available tools and software. We propose a workflow for analyzing large volumetric datasets of particle-reinforced metal matrix composites, from 3D renderings of the datasets to qualitative and quantitative characterizations of the material regarding shape and spatial distribution of the contained particles. Each step in this workflow is designed to be applicable to arbitrarily large volumetric datasets. Application-dependent visualizations facilitate derived secondary information to become accessible, generating in-depth insights despite the large number of particles. The workflow is demonstrated on a large high-resolution dataset in qualitative and quantitative evaluations, whose visual representations confirm that the distribution of particles within the sample is quite homogeneous albeit the presence of minor agglomerations.