Practical Tools for Integration and Analysis in Materials Engineering: Session I
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS Structural Materials Division, TMS: Titanium Committee, TMS: Computational Materials Science and Engineering Committee, TMS: Integrated Computational Materials Engineering Committee
Program Organizers: Adam Pilchak, Pratt & Whitney; Michael Gram, Pratt & Whitney; William Joost; Raymundo Arroyave, Texas A&M University; Charles Ward, AFRL/RXM
Monday 8:30 AM
March 15, 2021
Room: RM 34
Location: TMS2021 Virtual
8:30 AM
Introductory Comments: Practical Tools for Integration and Analysis in Materials Engineering: Adam Pilchak1; 1US Air Force Research Laboratory
Introductory Comments
8:35 AM Invited
Accelerated Tools for Disordered-materials Discovery: Stefano Curtarolo1; 1Duke University
In this presentation we will discuss novel high-throughput methods to address synthesizability of high-entropy systems. Research sponsored by DOD.
9:05 AM
Calculation of First Principles Based Thermodynamic and Kinetic Materials Properties Using CASM: Brian Puchala1; John Thomas2; John Goiri2; Anton Van der Ven2; 1University of Michigan; 2University of California, Santa Barbara
CASM is an open source statistical mechanics software package that automates the construction and first-principles based parameterization of effective Hamiltonians that can be used to calculate finite temperature thermodynamic and kinetic properties of multi-component crystalline materials. These properties provide a critical link between electronic structure and continuum descriptions of materials. Recent developments in CASM allow for the construction of cluster expansion effective Hamiltonians of mixed discrete and continuous degrees of freedom, enabling treatment of strain, displacements, magnetic spin, and user defined degrees of freedom. CASM can be easily installed as a conda package and integrates with common density functional theory (DFT) software. Here we will give an overview of CASM's capabilities and recent applications. We will also introduce potential users to online training materials including videos and tutorials that make use of training data available on the Materials Commons.
9:25 AM
A Framework for Closed-loop Materials Design Using Density Functional Theory: Vinay Hegde1; Kevin Williams1; Travis Ludlum1; Maxwell Hutchinson1; Eric Lundberg1; Bryce Meredig1; 1Citrine Informatics
Over the past few decades, density functional theory (DFT) has become the de-facto computational workhorse for atomistic modeling of materials. However, owing to the inherent complexities associated with parameter choices, post-processing steps, and workflows, mastering the correct practical use of DFT takes significant time and effort. While several high-throughput databases of DFT-calculated materials properties exist, they do not address the need for performing DFT calculations of a new material/property on demand, especially by non-experts. Here, we discuss a comprehensive end-to-end framework for standardizing, running, managing, and storing DFT calculations, composed of modular and extensible open-source software, with interfaces that enable ease-of-execution for DFT novices while retaining flexibility for expert users. We present an active learning-driven search for novel water-splitting perovskites as a demonstration of how the framework can be readily integrated into fully-automated, closed-loop materials design efforts.
9:45 AM
Batch Reification Fusion Optimization (BAREFOOT) Framework: Richard Couperthwaite1; Danial Khatamsaz1; Abhilash Molkeri1; Douglas Allaire1; Ankit Srivastava1; Raymundo Arroyave1; 1Texas A&M University
Developments in high-throughput materials manufacturing and testing have necessitated the development of design frameworks capable of making batch recommendations at each iteration in the optimization process. A Reification-Fusion based framework has already been developed and has been shown to greatly reduce the time required for the optimization of the mechanical properties of dual-phase steel. Since this framework is based on Bayesian optimization principles and so constructs Gaussian Process models of all the data used, it is an ideal candidate for a novel batch optimization procedure. This Batch optimization procedure samples from the Gaussian Process Hyper-parameter space to generate many realizations of the Gaussian Processes. This process removes any assumptions about the shape of the underlying function from the analysis and is capable of providing batch predictions. The current work has integrated these approaches into a combined framework that is capable of reducing the time and cost for optimizing material properties.
10:05 AM Invited
Microstructural Modeling with FiPy: Jonathan Guyer1; Daniel Wheeler2; James Warren2; 1National Institute of Standards & Technology; 2National Institute of Standards and Technology
The FiPy partial differential equation solver was devised with particular emphasis on solving the microstructural evolution problems that arise in materials science and engineering. In our experience, computational materials scientists often resort to simple finite difference codes with explicit forward Euler time stepping, limiting accessible geometries and time scales. FiPy provides an easy-to-learn Python interface to more sophisticated algorithms, including cell-centered finite volume on arbitrary meshes, fully and semi-implicit solutions, and parallel computatioms using either of the highly respected PETSc or Trilinos sparse solver libraries. While designed for materials scientists, by materials scientists, FiPy has been employed in fields as diverse as astrophysics, geoscience, and optometry. This talk will demonstrate application of FiPy to microstructural modeling, including a recent series of phase field benchmarks, showing both conventional Python programs and interactive scripting in jupyter notebooks.
10:35 AM
A Private Ledger Architecture Tailored for Secure Workflow Management in Additive Manufacturing Facilities: Evan Diewald1; Jack Beuth1; 1Carnegie Mellon University
While additive manufacturing has revolutionized the way parts are designed, built, and brought to customers, this disruptive processing paradigm presents unique vulnerabilities that challenge existing infrastructures. Distributed ledgers, or “blockchains” are a promising solution, but in their raw form, they are inefficient frameworks not suited for agile manufacturing workplaces. The AM Blockchain is an application designed to facilitate secure, decentralized, and unobtrusive workflow management in AM facilities. Contrary to permissionless systems such as those popularized by Bitcoin and Ethereum, the AM Blockchain is built on a custom private ledger architecture driven by IPFS. The consensus mechanism is a modified version of the Proof of Reputation algorithm, which allows for high transaction throughput without need for costly mining procedures. The modular smart contract handles a diverse array of transaction types, including secure file transfer of part drawings, and it supports IoT integration for seamless tracking of powder and post processing usage.
10:55 AM Invited
LAMMPS as a Tool in Materials Modeling Workflows: Steve Plimpton1; Aidan Thompson2; Mitch Wood2; 1Sandia National Laboratories; 2Sandia National Labs
The LAMMPS molecular dynamics package is widely used as a stand-alone code. However we've also tried to make it easy to use in tandem with other codes for multiphysics or multiscale models, or as part of workflows for data generation or analysis. In this talk I'll highlight features LAMMPS has for those modes of use. They include calling it as a library, wrapping it with Python or invoking Python from LAMMPS, different ways to couple it to other codes or databases, and using it in a machine learning context, either for developing and running machine-learned interatomic potentials or training them. I'll also give some practical examples of these kinds of use cases.
11:25 AM
The Materials Commons 2.0: A Collaboration Platform and Information Repository for the Global Materials Community: Brian Puchala1; Glenn Tarcea1; Tracy Berman1; John Allison1; 1University of Michigan
The Materials Commons is an information repository and collaboration platform developed by the PRedictive Integrated Structural Materials Science (PRISMS) Center at the University of Michigan. The Materials Commons helps researchers document their workflow, capturing what was done and how, linking to data so that a project can be explored and understood. The Materials Commons makes it easy to construct and publish datasets that can be assigned a unique digital object identifier (DOI) and found within Google Datasets, and readily supports projects with large files with Globus file transfer. We recently completed a rewrite of Materials Commons on a modern web stack. In this talk we will give an overview of Materials Commons 2.0, and show how to upload data, show how project data can be explored and visualized, and demonstrate how to publish a dataset. We will also show how integration with PRISMS software tools enables multi-scale collaboration.