Algorithm Development in Materials Science and Engineering: Solution Algorithms for Solidification Microstructure
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS Functional Materials Division, TMS Structural Materials Division, TMS: Computational Materials Science and Engineering Committee, TMS: Integrated Computational Materials Engineering Committee, TMS: Phase Transformations Committee, TMS: Solidification Committee, TMS: Chemistry and Physics of Materials Committee
Program Organizers: Adrian Sabau, Oak Ridge National Laboratory; Ebrahim Asadi, University of Memphis; Enrique Martinez Saez, Clemson University; Garritt Tucker, Colorado School of Mines; Hojun Lim, Sandia National Laboratories; Vimal Ramanuj, Oak Ridge National Laboratory
Wednesday 2:00 PM
March 22, 2023
Room: Cobalt 502B
Location: Hilton
Session Chair: Adrian Sabau, Oak Ridge National Laboratory; Ebrahim Asadi, University of Memphis
2:00 PM Invited
A Recursive Grain Remapping Scheme for Irregular Morphologies in Phase-Field Models: Alexander Chadwick1; Peter Voorhees1; 1Northwestern University
The multi-order parameter phase-field model (PFM) is a popular approach for simulating microstructural evolution in polycrystalline materials. However, computational resources limit the number of order parameters that can be employed, which reduces the number of crystallographic orientations and causes grains in the same order parameter to merge into one feature. Grain remapping eliminates these issues by dynamically assigning features to order parameters throughout the simulation. We present a remapping scheme that recursively decomposes grains into binary trees of axis-aligned bounding boxes (AABBs) while adding minimal overhead to existing codes, even with adaptive time stepping. As we increase the recursion depth, we obtain approximately conformal representations of grains that are efficiently tested for intersection. We demonstrate the scheme in a PFM of additive manufacturing, where grains have long columnar shapes with nonconvex features. We find that memory consumption is reduced by at least fourfold compared to bounding sphere schemes.
2:40 PM
An OpenMP GPU-Offload Implementation of a Cellular Automata Solidification Model for Laser Fusion Additive Manufacturing: Adrian Sabau1; Lang Yuan2; Jean-Luc Fattebert1; 1Oak Ridge National Laboratory; 2University of South Carolina
A cellular automata (CA) solidification code, which takes into account alloy segregation, diffusion and external temperature distributions, was used to simulate the formation of solidification microstructures for laser powder bed fusion additive manufacturing (LPBFAM). OpenMP offloading was used to accelerate simulations on GPU-based HPC platforms. Data on code speed-up and scalability is presented. The performance results on Summit at the Oak Ridge Leadership Computing Facility indicate that using a precomputed list of interface cells significantly decreased the wall-clock time on GPUs. A rapid directional solidification problem, representative of LPBFAM, was considered to demonstrate the CA code capability on Summit. A mesh size of 0.05µm was found to yield accurate elongated grain microstructure and elongated subgrain features, in qualitative good agreement with experimental data. The results presented in this study indicate that the implementation strategies on GPU-based HPC platforms for the CA code are appropriate for novel HPC exascale platforms.
3:00 PM
Characterization of the Evolution of the Grain Boundary Network Using Spectral Graph Theory: Jose Nino1; Oliver Johnson1; 1Brigham Young University
Changes to the Grain Boundary Network (GBN) caused by grain growth influence the final properties of the microstructure. If we could characterize the structure of GBNs, then it would be possible to perform design/optimization for improved material performance. However, the structure of GBNs is highly complicated. Traditional microstructural descriptors like orientation distribution function or even grain boundary character distribution fail to encode the main features of the GBN, including, e.g., its topological structure and the spatial distribution of GB types. For this reason, we apply a computational technique called Spectral Graph Theory which allows us to encode the GB character information as well as the topological structure of the GBN. We calculate and analyze the spectrum of several microstructures from grain growth simulations. Finally, we develop a reconstruction method to obtain the microstructure from its spectrum and evaluate whether the spectrum encodes the main features of the GBN.
3:20 PM Break
3:40 PM
Characterizing Microstructure Evolution in Latent Space for Machine Learning Applications: Saaketh Desai1; Ankit Shrivastava1; Marta D'Elia1; Habib Najm1; Remi Dingreville1; 1Sandia National Laboratories
Characterizing and quantifying microstructure evolution is critical to forming quantitative relationships between process conditions, resulting microstructure, and observed properties. Machine-learning methods such as recurrent networks can accelerate the development of these relationships by accelerating materials simulations, while techniques such as reinforcement/active learning can help improve representations and target specific microstructures/properties. However, these methods rely on the non-trivial task of identifying low-dimensional microstructural fingerprints that effectively relate process conditions to properties. In this work, we survey and discuss the ability of various linear/non-linear dimensionality reduction methods such as Principal Component Analysis, Karhunen Loeve Expansion, autoencoders/variational autoencoders, and diffusion maps to quantify and characterize the learned latent space microstructural representations and their time evolution. We target microstructure evolution problems such as spinodal decomposition, thin film deposition, and grain growth. This work paves the way to identify representation schemes that handle a variety of microstructural features across length scales for various machine-learning applications.
4:00 PM
Data Assimilation for Estimation of Microstructural Evolution during Solid-state Sintering: Integration of Phase-field Simulation and In-situ Experimental Observation: Akimitsu Ishii1; Akinori Yamanaka2; Akiyasu Yamamoto2; 1National Institute for Materials Science; 2Tokyo University of Agriculture and Technology
Sintering is a fundamental technology for manufacturing various materials. Improvement of the properties of sintered materials requires the prediction of their microstructure. Phase-field (PF) method is a powerful numerical simulation methodology for predicting microstructural evolutions during a solid-state sintering. However, material parameters required for the PF simulation are largely unknown. Recently, data assimilation (DA) has attracted attention as an effective method for estimating unobservable states and unknown material parameters. DA enables the estimations by integrating experimental data and numerical simulation results based on Bayesian inference. In this work, we estimate material parameters related to low-temperature sintering of copper by integrating morphological data observed using an in-situ scanning transmission electron microscopy (STEM) and PF simulation results of the solid-state sintering. To estimate materials parameters at a low computational cost, in this study, we used an efficient data assimilation method called DMC-BO. This work was supported by JST CREST (JPMJCR18J4).
4:20 PM
Diffuse Interface Technique to Simulate Fluid Flow and Characterize Complex Porous Media: Robert Termuhlen1; Genzhi Hu1; Jason Nicholas1; Hui-Chia Yu1; 1Michigan State University
Using molten silver to infiltrate through sintered porous interlayers is a promising new technique to create strong brazes between steel and ceramics in solid oxide fuel cells. Estimating infiltration speed is needed for optimizing the fabrication process. In this work, a diffuse interface embedded boundary method known as the Smoothed Boundary Method (SBM) is utilized to facilitate simulations of fluid dynamics involving complex geometries. Using this method, the geometry is described by a continuous domain parameter, thus allowing the straightforward reformulation of the time-dependent Navier-Stokes equations in terms of this domain parameter. In this case, a mesh conformal to the geometric boundaries of the microstructure is not required and the numerical simulation process is greatly simplified. The direct flow simulation allows the calculation of the permeability of porous media, a property commonly used to predict infiltration speed. We use these calculations to investigate the infiltration of silver in porous microstructure.
4:40 PM Cancelled
Numerical Modeling of Porosity Formation and Dendrite Growth with Lattice Boltzmann Method(LBM) – Cellular Automata(CA): Wonjoo Lee1; Howon Lee2; Seong-hoon Kang2; Jonghun Yoon1; 1Hanyang University; 2Korea Institute of Materials Science
Two-dimensional microstructural morphologies of dendritic growth and porosity evolution in the solidified Al-Cu alloy were investigated by conducting lattice Boltzmann (LB)-cellular automata (CA) method. The LB and CA adopted to simulate the diffusion of alloy solute and hydrogen in liquid melt and phase transition during solidification, respectively. The effects of applied cooling rates and initial hydrogen concentration on the morphologies were confirmed by comparing with experimental results. The lower cooling rate activated the large porosity size due to sufficient growth time. The porosity size also increased in the higher hydrogen concentration under the same cooling rate. The proposed LB-CA model does not only describe the different types of dendrite growth with respect to the solidification conditions such as cooling rate, but also simulate the multiphase morphology including gas porosity in solidification process of Al-Cu alloy system.