Towards accurate processing-structure-property links using deep learning

Establishing links between the processing, the microstructure and the properties of steel is of utmost importance for rational material design. Including information from the microstructure proves difficult as it requires quantifying the visual information included in microscopy images. The increasing performance of deep learning models in computer vision offers great potential in this regard. Herein, we investigate how features from deep learning models can help us to predict information about the hardness and the composition based on SEM images for a set of complex martensitic steels.