First World Congress on Artificial Intelligence in Materials and Manufacturing (AIM 2022): AI-Assisted Development of New Materials/Alloys II
Program Organizers: Taylor Sparks, University of Utah; Michael Dawson-Haggerty, Kerfed, Inc.; Elizabeth Holm, University of Michigan; Jin Kocsis, Purdue University; Adam Kopper, Mercury Marine; Benji Maruyama; James Warren, National Institute of Standards and Technology
Tuesday 1:30 PM
April 5, 2022
Room: William Penn Ballroom
Location: Omni William Penn Hotel
Session Chair: Andrew Detor, DARPA/DSO
1:30 PM Invited
Optimizing Fractional Composition to Achieve Extraordinary Properties: Andrew Falkowski1; Steven Kauwe1; Taylor Sparks1; 1University of Utah
Data-driven materials discovery involves screening chemical systems with machine learning algorithms and selecting candidates based on a target property. The number of screening candidates grows infinitely large as the fractional resolution of compositions and the number of included elements increases. The computational infeasibility and probability of overlooking a successful candidate grow likewise. Our approach takes inspiration from neural style transfer and shifts the optimization focus from model parameters to the fractions of each element in a composition. By leveraging a pre-trained network with exceptional prediction accuracy (CrabNet) and writing a custom loss function to govern a vector consisting of element fractions, material compositions can be optimized such that a predicted property is maximized or minimized. The simplicity of the approach allows sophisticated multiobjective optimization algorithms, such as the hypervolume indicator, to be easily translated to inverse design problems for dopant tuning and other applications.
2:00 PM
Predictive Modeling of Creep Elongation and Reduction in Area in High Temperature Alloys Using Machine Learning: Madison Wenzlick1; Osman Mamun2; Ram Devanathan2; Kelly Rose3; Jeffrey Hawk3; 1National Energy Technology Laboratory; NETL Support Contractor; 2Pacific Northwest National Laboratory; 3National Energy Technology Laboratory
New materials are critical to enabling technological advancements in the energy industry. To address this need, the eXtremeMAT (XMAT) consortium of U.S. Department of Energy national laboratories aims to improve the alloy design process for energy-related materials through the development of novel models and tools to reduce the time from material design to deployment. This work leverages high quality data on high-temperature 9-12% Cr ferritic-martensitic steels and austenitic stainless steels, including alloy composition, processing, heat treatment, microstructure and creep test information. Machine learning regression algorithms were used to model the creep elongation and reduction in area of the alloys to understand the complex relationships between the alloy properties and their mechanical behavior, and to supplement existing creep rupture models. The feature importance and feature interactions were assessed. Domain knowledge was integrated into the model using alloy labeling and expert review to ensure interpretability and alignment with materials science properties.
2:20 PM
Machine Learning Enabled Model to Predict Mechanical Properties of Refractory Alloys: Trupti Mohanty1; K. S. Ravi Chandran1; Taylor D Sparks1; 1University of Utah
Refractory metal alloys with targeted mechanical properties find numerous high-temperature applications. However, experimental determination of mechanical properties particularly at high temperatures is quite challenging. In view of this, the present study is focused on harnessing the power of machine learning with an objective to predict yield strength (YS) and ultimate tensile strength (UTS) of novel refractory alloys at varying temperatures. Since only a limited number of experimental YS and UTS data is available therefore it is attempted to improve the predictive accuracy of the machine learning model by augmenting the number of training datasets by incorporating YS and UTS of multi-principal element alloys. In this work, the composition-based features are utilized for the development of machine learning models. The intended use of the developed model is to derive novel alloy composition with desired mechanical properties for high-temperature applications.
2:40 PM Break
3:10 PM
Machine Learning Guided Prediction of Thermal Properties of Rare-Earth Disilicates and Monosilicates: Mukil Ayyasamy1; 1University Of Virginia
Rare-earth silicates are promising candidate materials for the application of environmental barrier coatings (EBC) for SiC composites in gas-turbine engines. One of the most important design criteria is the good coefficient of thermal expansion (CTE) match with that of the SiC composites. Although theoretical rules and computational methods are of practical use for CTE prediction, unsatisfactory predictability and universality of models across different materials of the class impede rational design of EBC materials. In this work, we build a machine learning model for CTE prediction that is universal to both rare-earth disilicates and monosilicates. We consider descriptors based on unit cell polyhedral features obtained from our density functional theory calculations. We employ support vector regression ensembles via bootstrapping to determine prediction uncertainty along with the CTE predictions. Additionally, our model interpretability methods reveal quantitative structure-property relationships.
3:30 PM
Machine Learning Enabled Directed Energy Deposition of Functionally Graded Materials: Alex Kitt1; Lee Kerwin1; Anindya Bhaduri2; Luke Mohr1; Chen Shen2; Siyeong Ju2; Hyeyun Song1; Shenyan Huang2; Arushi Dhakad1; Sathyanarayanan. Raghavan2; Marissa Brennan2; Lang Yuan3; Changjie Sun2; 1EWI; 2GE Global Research; 3University of South Carolina
When developing functionally graded materials, residual stress, formation of detrimental phases, differences in liquidus temperatures, and crack formation modes must be considered. To overcome these complexities, this work synergies a diverse range of inputs within a Bayesian Hybrid Model. The focus of this work is a low-gamma' to high-gamma' high strength nickel alloy functionally graded materials built using laser blown powder directed energy deposition for gas turbine or jet engine applications. This presentation will describe how thermodynamic modeling, process modeling, and microstructure modeling are calibrated against process-monitoring, photo-micrographs, neutron scattering, and synchrotron measurements. Further, it will describe how the experimental results and calibrated models are leveraged to enhance the process development. Special focus will be given to data registration and prediction of cooling rates.