AI/Data Informatics: Computational Model Development, Validation, and Uncertainty Quantification: MGI/Uncertainty Quantification
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Computational Materials Science and Engineering Committee
Program Organizers: Saurabh Puri, Microstructure Engineering; Dennis Dimiduk, BlueQuartz Software LLC; Darren Pagan, Pennsylvania State University; Anthony Rollett, Carnegie Mellon University; Francesca Tavazza, National Institute of Standards and Technology; Christopher Woodward, Air Force Research Laboratory

Monday 8:30 AM
February 28, 2022
Room: 256A
Location: Anaheim Convention Center

Session Chair: Somnath Ghosh, Johns Hopkins University; Gary Whelan, Questek Innovations Llc


8:30 AM  Invited
10 Years of the Materials Genome Initiative: James Warren1; 1National Institute of Standards and Technology
    AI and data informatics, the focus of this symposium, are central to the future success of the Materials Genome Initiative (MGI). In this talk we will look back at the progress made by the MGI in the last decade, discuss the connections to AI and data informatics, the challenges that remain for the MGI to achieve its goals, and the US government’s strategy to address these challenges.

9:00 AM  Invited
Uncertainty Quantified Parametrically Homogenized Constitutive Models for Multi-scale Predictions of Fatigue Crack Nucleation in Ti Alloys: Somnath Ghosh1; Shravan Kotha1; Deniz Ozturk1; 1Johns Hopkins University
    This paper develops an uncertainty quantified, parametrically homogenized constitutive model (UQ-PHCM) for microstructure-sensitive simulation of deformation and fatigue crack nucleation in components of Ti alloys, viz. Ti-7Al and Ti64. The PHCMs are thermodynamically consistent, macroscopic constitutive models, whose parameters are explicit functions of Representative Aggregated Microstructural Parameters or (RAMPs) of microstructural morphology and crystallography. Machine learning tools operate on datasets generated by CPFEM to obtain these functional forms. A significantly reduced number of solution variables in the PHCM simulations make them several orders of magnitude more efficient with good accuracy. The UQ-PHCM framework is based on Bayesian inference to derive probabilistic microstructure-dependent constitutive laws of the macroscopic material response. The framework addresses three sources of uncertainty that accrue at the model development and response prediction stages, viz: (i) model reduction error, (ii) data sparsity, and (iii) microstructural variability. The validated UQ-PHCM is implemented to test its viability in real applications.

9:30 AM  
Inference, Uncertainty Quantification, and Uncertainty Propagation for Grain Boundary Structure-property Models: Oliver Johnson1; Eric Homer1; David Fullwood1; David Page1; Kathryn Varela1; Sterling Baird1; 1Brigham Young University
    We present a non-parametric Bayesian approach for developing structure-property models for grain boundaries (GBs) with built-in uncertainty quantification (UQ). Using this method we infer a structure-property model for H diffusivity in [100] tilt GBs in Ni at 700K based on molecular dynamics (MD) data. We then leverage these results to perform uncertainty propagation (UP) for mesoscale simulations of the effective diffusivity of polycrystals to investigate the interaction between structure-property model uncertainties and GB network structure. We observe a fundamental interaction between crystallographic correlations and spatial correlations in GB networks that causes certain types of microstructures (those with large populations of J2- and J3-type triple junctions) to exhibit intrinsically larger uncertainty in their effective properties.

9:50 AM Break

10:10 AM  
Uncertainty Quantification Framework for Robust Design of Fatigue Critical Alloys: Gary Whelan1; David McDowell2; Sam Sorkin1; Jiadong Gong1; 1QuesTek Innovations LLC; 2Georgia Institute of Technology
    ICME facilitates efficient design and development of new materials, as well as optimization of existing materials. However, uncertainty is prevalent in computational modeling workflows, particularly for extreme phenomena such as fatigue. Therefore, uncertainty quantification is a critical step to achieve effective use of modeling results to support robust materials design. This work demonstrates a computational framework to address uncertainty quantification and propagation across process-structure-property (PSP) linkages for robust optimization of engineering alloys in fatigue critical applications. Process-structure linkages are modeled using CALPHAD and structure-property linkages are modeled using the crystal plasticity finite element method (CPFEM). Both epistemic and aleatory uncertainties are quantified and propagated across the entire PSP domain. Reduced-order surrogate models are trained using high-fidelity CPFEM models to facilitate rapid propagation of uncertainty and inductive design exploration across structure-property linkages. This framework is demonstrated by carrying out a robust optimization of the Ti64 material system for biaxial fatigue conditions.

10:30 AM  
Using Polycrystals for Bayesian Inference and Uncertainty Quantification of Grain Boundary Structure-property Models: Brandon Snow1; Sterling Baird1; Christian Kurniawan1; David Fullwood1; Eric Homer1; Oliver Johnson1; 1Brigham Young University
    Bicrystals permit interrogation of the properties of individual grain boundaries (GBs) directly, but can be difficult, time-intensive, or expensive to synthesize experimentally. In contrast, polycrystals are ubiquitous, contain a broad diversity of GB types, and can be more straightforward and inexpensive to produce. We present a Bayesian inference strategy designed to enable the use of homogenized effective property measurements from polycrystals for the construction of grain boundary structure-property models. This strategy involves the solution of an inverse problem (determining the properties of individual GBs from the homogenized properties of polycrystals) and naturally provides uncertainty quantification (UQ) for the resulting GB structure-property models. We compare structure-property models inferred from polycrystals and bicrystals and find that under certain circumstances bicrystal data is preferred, while under other circumstances the use of polycrystal data may be more advantageous.