AI/Data Informatics: Computational Model Development, Validation, and Uncertainty Quantification: Poster Session
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
Tuesday 5:30 PM
March 1, 2022
Room: Exhibit Hall C
Location: Anaheim Convention Center
NOW ON-DEMAND ONLY – M-8: A Data-driven Surrogate Model for Fast Predicting the U-10Mo Fuel Grain Structures during the Hot Rolling and Annealing Processes: Yucheng Fu1; William Frazier1; Kyoo Choi1; Lei Li1; Zhijie Xu1; Vineet Joshi1; Ayoub Soulami1; 1Pacific Northwest National Laboratory
The U-10Mo fuel foil fabrication process steps of hot rolling and annealing play an important role in determining the foil grain structure. The grain structure then impacts the foil’s performance during irradiation. To enable fast prediction and optimization of U-10Mo grain structure during these processing steps, a data-driven surrogate model has been developed using simulation data, produced from a coupled Kinetic Monte Carlo (KMC) Potts model and finite element method (FEM) model. The microstructures simulated using these techniques cover a wide range of initial grain size, uranium carbide volume fraction and rolling reductions. With the acquired high fidelity data, the deep learning, stack ensemble-based surrogate model has been trained on the acquired dataset. This data-driven surrogate model demonstrates good accuracy in predicting the U-10Mo fuel microstructures and guides the selection of proper processing parameters for U-10Mo foil production.
M-9: Determination of Aluminum Mechanical Properties Using Small Punch Test and Advanced Numerical Methods: Matan Tubul1; Ziv Ungarish1; Shay Amar1; Eytan Kochavi2; 1Negev Nuclear Research Center; 2Ben Gurion University of the Negev (on Sabbatical leave from NRCN)
The Small Punch Test (SPT) is a mechanical testing method in which a foil is bended to failure using a ball punch, while the force-displacement curve is recorded, and is used to obtain the mechanical properties of the foil. This work describes a method to determine the yield stress and ultimate tensile strength of aluminum alloys, using SPT and an "Inverse procedure" which uses advanced numerical methods (finite element model (FE), artificial Neural network (NN)). The SPT method was examined using experiments, a verified and validated FE, and comparison to tensile test results. Several aluminum alloys were tested. The developed FE and NN model can adequately reproduce the experimental results and is suitable for understanding the stresses/strain distribution during SPT of aluminum miniature samples. The applicability of SPT and its ability to obtain the mechanical properties of different aluminum alloys, using the"Inverse procedure" will be discussed.
M-10: Investigating the Suitability of Tableau Dashboards and Decision Trees for Particulate Materials Science and Engineering Data Analysis: Bryer Sousa1; Richard Valente1; Aaron Krueger1; Eric Schmid1; Danielle Cote1; Rodica Neamtu1; 1Worcester Polytechnic Institute
Informed integration of data-driven models for materials processing has yet to be fully realized due to data science knowledge gaps, incomplete materials and processing datasets, and a lack of data-driven tools designed explicitly for classically trained engineers. On the other hand, modern particle size distribution analyzers enable hundreds of thousands of particle-to-particle size, shape, and morphological properties to be easily gathered. Accordingly, we present suitable data analysis, sharing, and visualization approaches for developing a powder particle classification based upon powder morphology and size metrics for Flowability on Demand (FoD). We demonstrate the utility of Tableau dashboards connected to a live powder database for making data-driven integration convenient to assess, visualize, and analyze particulate data; thus, making comparisons between the features of individual powders and micro-particulate constituents accessible for traditional materials scientists and engineers. The FoD framework reduced the time taken for common workflows by 81.13% for FoD-based tasks.
M-11: Machine Learning Based Prediction of Melting Temperature and Coefficient of Thermal Expansion Coefficient Using SciGlass Datasets: Jong Ho Kim1; 1Rist
Ceramic materials vary in properties depending on composition and can be designed with various materials depending on composition. Existing composition design was a way to explore desired composition based on domain knowledge, but a new method has recently been proposed due to rapid advances in machine learning and artificial intelligence.An open database SciGlass exists for the composition of glass materials and provides material properties in various compositions. It provides more than 30,000 data sets of materials, enabling machine learning using them. In this work, we test a model to measure melting temperature and thermal expansion coefficients of glass or ceramic materials using various machine learning techniques. As a result, both traditional methods such as random forest and recent deep learning models have shown satisfactory results. The results of the Factsage calculation were compared with that of the two-component ceramic material and showed some consistent results.