The use of modern natural and man-made composite materials offers new capabilities that go way beyond those of materials used one or two decades ago. While easy to capture advantages such as weight savings are the driving forces in these developments, the behavior of such materials is more complicated to characterize.

Image-based techniques have a lot of potential in material characterization. They can benefit massively from the most recent developments in artificial intelligence (AI). In this presentation several examples will be presented that combine theoretical modeling approaches with machine learning in order to make image based predictions.

First, the automatic analysis of weld beads generated by laser dispersion will be discussed. The use of simple relations like conservation laws and models will demonstrate how machine learned models can be boosted at little effort. The focus will then switch from computer assisted image analysis towards 3D in silico property predictions. Making use of a dimensionality reduction approach a database is generated from random microstructure samples which is then allowing the use of artificial neural networks for the effective property forecasting. The 3D simulations are implemented in a high-efficiency solver making use of the Fast Fourier Transform that accepts 3D microstructure images as inputs.