3D phase-field simulations to machine-learn 3D information from 2D micrographs

Abstract

A novel approach is developed to support retrieval of 3D information from 2D experimental micrographs. The approach utilizes 3D phase-field simulations to train an artificial intelligence (AI) machine. In a first step, the phase-field simulations have to be validated to reproduce microstructural features which characterize elementary processes which govern processing and high temperature service exposure. The qualified 3D simulation setup is then applied to produce a high number of 2D simulated micrographs by automated sectioning. These simulated micrographs are then used to train a gradient boosting regression model together with the 3D information from the simulations. In the final step, the model is applied to 2D experimental micrographs to retrieve the hidden 3D features. The approach is generally applicable to all kinds of metallic materials, minerals or ceramics which can be treated quantitatively by phase-field simulations. In this paper we concentrate on the process of directional coarsening, referred to as "rafting", in the field of creep of single crystal Ni-base superalloys. The experimental and modeling aspects of the evolution of the volume fraction of the $\gamma'$ phase during long term creep are discussed.