A data-driven machine learning approach to predicting stacking faulting energy in austenitic steels

Stacking fault energy (SFE) is an intrinsic material property whose value is crucial in determining different secondary deformation mechanisms in austenitic (face-centered cubic, fcc) steels. Considerable experimental and computational work suggests that the SFE itself is highly dependent—in a complex manner—on chemical composition and temperature. Over the past decades, there have been a large number of efforts focused on determining the composition dependence of SFE in austenitic steel alloys by means of experimental, theoretical or computational methods. Unfortunately, experimental methods suffer from the indirect nature of the methodologies used to estimate the value of SFE, while computational and/or theoretical approaches are either limited by the physics that they can incorporate into the predictions or have more practical limitations associated, for example, to the size of the systems that can be modeled or the assumptions that must be made. In this paper, we review the major experimental and computational approaches to determine SFE in austenitic steel alloys, and we discuss their limitations. We then demonstrate a data-driven machine learning technique to mine the literature of experimental SFE data in steels, while algorithms at the fore-front of machine learning have been used to visualize the SFE data and then construct a three-class classifier. The classifier is used then to predict likely secondary deformation mechanisms of untested compositions, while the classifier itself is presented as a valuable tool for the further development of austenitic steel alloys in which the specific secondary plastic deformation mechanisms are a feature to design for. The data as well as the entire analysis workflow are made available to the wider community through a public github repository.