Generative reconstruction of 3D volume elements for Ti-6Al-4V basketweave microstructure by optimization of CNN-based microstructural descriptors

Abstract

We present a methodology for the generative reconstruction of 3D Volume Elements (VE) for numerical multiscale analysis of Ti-6Al-4V processed by Additive Manufacturing (AM). The basketweave morphology, which is typically dominant in AM-processed Ti-6Al-4V, is analyzed in conventional Electron Backscatter Diffusion (EBSD) micrographs. Prior \b{eta}-grain reconstruction is performed to obtain the out-of-plane orientation of the observed grains leveraging Burgers orientation relationship. Convolutional Neural Network (CNN) - based microstructure descriptors are extracted from the 2D data, and used for cross-section-based optimization of pixel values on orthogonal planes in 3D, using the Microstructure Characterization and Reconstruction (MCR) implementation MCRpy [16]. In order to utilize MCRpy, which performs best for binary systems, the basketweave microstructure, which consists of up to twelve distinct grain orientations, is decomposed into several separate two-phase systems. Our reconstructions capture key characteristics of the titanium basketweave morphology and show qualitative resemblance to experimentally obtained 3D data. The preservation of volume fraction during assembly of the reconstruction remains an unadressed challenge at this stage.