A data-driven and machine-learning study on microstructure-property relations in steel

Abstract

Multi-phase metallic materials such as Advanced High-Strength Steels (AHSS) are of great importance in a wide variety of high-tech industries due to their higher strength compared to conventional (mild) forming steels. The higher strength leads to various advantages in weight, safety and environmental friendliness. In order to develop new AHHS steels, the steel industries make use of multi-scale microstructure modelling to predict mechanical properties from the microstructure features.
This thesis aims at the development of relations between the features of multi-phase metallic microstructures of steels and the mechanical properties of the material. The quantitative characterization of the microstructure will be more involved than is now in use for estimations of the mechanical properties, which is a necessity because of the complexity of multi-phase microstructures. Moreover, the prediction of mechanical properties on the basis of microstructural features will be extended beyond the usual limitation of the yield stress to properties like hole expansion capacity and impact energy. Statistical approaches combined with machine learning algorithms are used to find relations between microstructure features and mechanical properties. Interpretations of the machine learning algorithms are also discussed and the possible deeply embedded relations among mechanical properties are systematically studied.
The research in this thesis deepens the insight into the mechanical behaviour of the microstructure in multi-phase steels and strongly improves property predictions, not only based on microstructure features, but also on deformation properties. Results of this thesis can be directly implemented in microstructure modelling and are directly available for researchers within the steel industry for developing new materials.