Addressing materials' microstructure diversity using transfer learning

Materials' microstructures are signatures of their alloying composition and processing history. Automated, quantitative analyses of microstructural constituents were lately accomplished through deep learning approaches. However, their shortcomings are poor data efficiency and domain generalizability across data sets, inherently conflicting the expenses associated with annotating data through experts, and extensive materials diversity. To tackle both, we propose to apply a sub-class of transfer learning methods called unsupervised domain adaptation (UDA). UDA addresses the task of finding domain-invariant features when supplied with annotated source data and unannotated target data, such that performance on the latter is optimized. Exemplarily, this study is conducted on a lath-shaped bainite segmentation task in complex phase steel micrographs. Domains to bridge are selected to be different metallographic specimen preparations and distinct imaging modalities. We show that a state-of-the-art UDA approach substantially fosters the transfer between the investigated domains, underlining this technique's potential to cope with materials variance.

Automated summary

Automated, quantitative analyses of materials' microstructures have been achieved through deep learning methods, but face challenges of low data efficiency and lack of domain generalizability. A solution proposed is the application of unsupervised domain adaptation (UDA) methods, which have shown promising results in complex phase steel micrographs.