Constant improvements in imaging technology, huge datasets and growing processing power are driving the possibilities in quantitative microstructure analysis (QMA) forward and allow a more efficient quantification of microstructural features (e.g. phase fractions, particle size distributions, defects and inhomogeneities) of microstructures and fine geometries. Complex relationships can be understood better and microstructural features can be tracked over a longer period. This means that manufacturing parameters can be correlated with the microstructure and irregularities can be detected for quality assurance tasks.
Intelligent image processing approaches help to examine complex or inhomogeneous microstructures. The used methods in classic QMA often fail to differentiate optically similar phases especially if non-optimal imaging conditions, as for example shading or defocus in microscopic images occur. For those issues, machine learning (ML) for segmentation and classification of different microstructure types as well as prediction of material properties from a microstructure´s image data offers a huge potential for materials microscopy.
With the examples of massive and sintered steel, Li-ion batteries and magnets the application and powerful advantages of ML in materials microscopy will be demonstrated. Specifically trained models enable the prediction of the hardness of a martensitic and bainitic 100Cr6 steel from microstructure data. ML is also applied to assess the quality of Li-ion batteries by detecting electrode layer thickness variations and abnormalities in the electrodes automatically. By the example of pores in sintered steels, non-metallic inclusions in rolled steel and a multiphase magnetic material it will be demonstrated how machine learning allows to achieve a more robust and reproducible image segmentation, and hence quantification of various phases in the materials.