Article ID: ISIJINT-2019-089
As traditional experimental science appears to be inefficient for designing novel materials with desired properties because of the complex combination of processing conditions and chemical compositions, data-driven materials science is becoming increasingly important for materials design. A properties-to-microstructure-to-processing inverse analysis for steels is attempted via a machine learning approach in this work, where a potential best balanced property of tensile strength (TS) and total elongation (tEL) TS × tEL and its corresponding microstructure and processing conditions are explored using a genetic algorithm, which is implemented by an independently developed machine learning tool called the Materials Genome Integration System Phase and Property Analysis (MIPHA). The results demonstrate that a property-to-microstructure/processing method is sufficient to identify a best model performance, potential TS × tEL, and a reasonable relationship description among the processing, microstructure and property. A microstructure with Widmanstatten ferrite, banite, and martensite is found to be beneficial to a good balanced property.