Abstract: Application of Computer-Vision Techniques in Materials Science and Engineering
The quantitative characterization of microstructure is a central problem in materials science and engineering, because the microstructure formed during the processing of technical alloys strongly governs the macroscopic properties of a material. Extracting such information from microscopy data, however, is fundamentally an image-analysis problem: relevant features must be detected and classified. Geometric properties of these features must be determined and their significance and effect on materials properties must be evaluated. In the past, specialists performed this task by manually evaluating micrographs. This lecture introduces how computer-vision methods can be used to automate the analysis of micrographs from both conventional optical microscopy and crystallographic orientation imaging. After outlining the role of microstructure in process-structure-property-performance (PSPP) relations, the lecture will discuss core concepts from image processing and computer-vision, including segmentation, feature extraction, and learning-based approaches. Real examples will then be presented from ongoing materials science projects, including convolutional neural networks for the analysis of optical micrographs and orientation-based methods for the quantitative evaluation of electron backscatter diffraction (EBSD) and related orientation imaging datasets. The lecture aims to show that automated microstructure analysis is not merely a practical engineering tool, but also a rich applied-mathematics problem at the interface of imaging, data science, and physical interpretation.