Ensemble modeling of non-linear relations for 9% Cr-family steel (tensile strength)

Materials data analytics (MDA) methodology was developed in this study, for dealing with non-linear relationships and sparsity in materials data. Motivation for this research comes from the desire to shorten the rigorous and time-consuming materials qualification process, for new fossil energy materials applications. The focus is on 9% Cr-family steels used as structural materials in boiler and turbine applications in power generation. The analyzed 9-12% Cr steel data set, for 82 iron base alloy compositions and processing parameters, displayed results of tensile strength in 34 columns by 915 rows. To address non-linearity of the tensile properties, data analyses were carried out in composition-based clusters. The cluster-based models were tested and further refined using additional data set, with 16 alloy compositions. The ensemble of competitive models proved to be a viable tool in identifying piecewise-linear features, including cooperative effects of multiple variables. Physics based models and empirical domain knowledge were used to guide the data-driven models selection process.