Segmentation and Classification of Lower Bainite and Tempered Martensite High Strength Steels Using Deep Learning

Because of their superb mechanical properties, high-strength steel (HSS) is a crucial structural metal for many industries, and has seen great use in applications under extreme environments. However, they are very susceptible to hydrogen embrittlement (HE), a phenomenon where the ingress of H leads to the material’s ductility loss and abrupt failures. The susceptibility of HSS to HE has been demonstrated to be strongly dependent on the microstructure therein. Two common microstructure types within HSS are lower bainite (LB) and tempered martensite (TM). It has been reported that LB shows more resistance to HE than TM, attributed to difference in the distribution and morphology of carbide precipitates. However, the characterization and comparison of microstructures in LB and TM have been mainly qualitative, and often ad-hocly relying on highly localized features, thus prone to uncertainties particularly in situations when there is no prior knowledge of the heat treatment. Employing deep Learning, we have established a workflow for efficient segmentation and classification of microstructure images of HSS. The workflow enables streamlining of quantitative analysis of carbides in terms of the aspect ratio, size, alignment and density. Our results showed that, despite difference in local regions, LB and TM exhibit little or marginal difference in global characteristics of carbides. Meanwhile, our deep learning approach has been shown to accurately differentiate the LB and TM microstructures, with an impressive 96.81% accuracy. Our findings demonstrate the potential of deep Learning as a powerful tool in quantitatively analyzing and distinguishing complex microstructures in HSS.