Computational Method for Microstructure Design to Achieve Target Properties of Multiphase Metallic Materials

DoctorThree-dimensional (3D) microstructure-based models offer unique opportunities in understanding the mechanical behavior of materials on a microscopic level that collectively defines the final properties of the material, making these models a powerful computational tool for microstructure design. Nevertheless, the execution of computational microstructure design confronts several challenges including reliability of microstructure-based simulations and an efficient method to overcome the computational cost of these simulations. The reliability of microstructure-based models relies on their physical and theoretical foundation on how a material with specific microstructure responds under a thermomechanical load. In a continuum framework, microstructure-based full-field numerical approach with physical constitutive models offers one of the most reliable results. Unfortunately, incorporation of physical details, including microstructure, in full-field models is reflected in the computationalcost of the models. Due to the high dimensional design space created by complex microstructural variables, which control both inter- and intra-phase mechanical behaviors, practicality dictates that adirect application of microstructure-based models to tasks such as materials design that require numerous evaluations is not readily feasible. Therefore, the aim of this thesis is twofold: 1) to construct microstructure-based models that can accurately simulate micro-/macro-scale mechanical behavior of materials, and 2) to devise a method that can utilize expensive models for microstructure design. Microstructure-based models require accurate representative volume element (RVE) and robust constitutive formulation for mechanical behavior of relevant microstructural variables. The former is accomplished through statistically equivalent 3D microstructure and the latter by continuum elasto-plasticity with relevant hardening equations. The effectiveness of 3D microstructure-based models is established by modelling microscopic phenomena and predicting macroscopic properties of Al-SiC composite and twinning-induced plasticity (TWIP) steel followed by experimental verifications. Finally, a probabilistic framework based on Bayesian statistics, or Gaussian process (GP), with an application-specific kernel and low dimensional representation of microstructure using correlation statistics is proposed to represent and manipulate uncertainty about models and predictions. As a demonstrative purpose, structure-property (SP) linkages that are accurate over a wide range of two-phase microstructures were constructed by evaluating only a small number of microstructure-based models. Furthermore,the implementation of expected improvement algorithm successfully identified microstructures with target properties using even smaller number evaluations. For verification, the framework with was applied to dual phase (DP) steels with different microstructures and carbon content. The microstructure-based simulations were based on physical model and well-calibrated model parameters to reproduce experimental data. Results confirmed that an impressive level of accuracy can be achieved even with a small amount of evaluations. Furthermore, the quantified linkages were transferable to experimental data because microstructure-based simulations were based on experimental data