Data-driven reduced order models for effective yield strength and partitioning of strain in multiphase materials

JOURNAL OF COMPUTATIONAL PHYSICS

卷:346

Article

Latypov, Marat I.¹  Kalidindi, Surya R.^2,3 

[1] Georgia Tech Lorraine, GT CNRS UMI 2958, F-57070 Metz, France

[2] Georgia Inst Technol, Woodruff Sch Mech Engn, Atlanta, GA 30332 USA

[3] Georgia Inst Technol, Sch Computat Sci & Engn, Atlanta, GA 30332 USA

关键词: Homogenization theories;    Strain partitioning;    Data science;    Multiphase materials;    2-point statistics;    Reduced-order models;   

DOI  :  10.1016/j.jcp.2017.06.013

来源: Elsevier

【 摘 要 】

There is a critical need for the development and verification of practically useful multiscale modeling strategies for simulating the mechanical response of multiphase metallic materials with heterogeneous microstructures. In this contribution, we present datadriven reduced order models for effective yield strength and strain partitioning in such microstructures. These models are built employing the recently developed framework of Materials Knowledge Systems that employ 2-point spatial correlations (or 2-point statistics) for the quantification of the heterostructures and principal component analyses for their low-dimensional representation. The models are calibrated to a large collection of finite element (FE) results obtained for a diverse range of microstructures with various sizes, shapes, and volume fractions of the phases. The performance of the models is evaluated by comparing the predictions of yield strength and strain partitioning in two-phase materials with the corresponding predictions from a classical self-consistent model as well as results of full-field FE simulations. The reduced-order models developed in this work show an excellent combination of accuracy and computational efficiency, and therefore present an important advance towards computationally efficient microstructure-sensitive multiscale modeling frameworks. (C) 2017 Elsevier Inc. All rights reserved.

【 授权许可】

Free