Correlative microscopy meets machine learning - An approach for a combined microstructure quantification from microscopic images and EBSD data utilizing ML-based analysis

The increasingly complex microstructures of modern materials cannot always be fully assessed by just one characterization method alone. For a fully comprehensive description of the microstructure across multiple scales of observation, several characterization methods can then be combined in a correlative approach. In this work, a correlative approach for combined microstructure quantification from optical microscopy (OM), scanning electron microscopy (SEM), and electron backscatter diffraction (EBSD) is presented.

Using the example of a dual-phase steel consisting of a ferritic matrix and a carbon-rich second phase (pearlite, various types of bainite or martensite), ways of evaluating such a correlative data set using machine learning (ML) methods are presented. The accuracies with which the carbon-rich second phase can be classified based on the different characterization methods are investigated. This will reveal which characterization method is sufficient or necessary for which complexity of classification, and where the limits of a pure OM-based classification lie. It will also be investigated to what extent an objective ground truth for ML can be specified in an automated manner purely on the basis of the EBSD data, which are not based on how the microstructure is perceived by the human expert's eye.