AI/Data informatics: Design of Structural Materials: AI/ML for Integrating Experiments and Simulations; Steels
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS Structural Materials Division, TMS: Mechanical Behavior of Materials Committee, TMS: Computational Materials Science and Engineering Committee, TMS: Integrated Computational Materials Engineering Committee
Program Organizers: Jennifer Carter, Case Western Reserve University; Amit Verma, Lawrence Livermore National Laboratory; Natasha Vermaak, Lehigh University; Jonathan Zimmerman, Sandia National Laboratories; Darren Pagan, Pennsylvania State University; Chris Haines, Ccdc Army Research Laboratory; Judith Brown, Sandia National Laboratories
Wednesday 8:30 AM
March 17, 2021
Room: RM 32
Location: TMS2021 Virtual
8:30 AM Invited
A Physics-informed Bayesian Experimental Autonomous Researcher for Structural Design: Keith Brown1; 1Boston University
The combination of automated experimentation and active learning has resulted in autonomous systems that accelerate the pace of research in fields such as materials science, chemistry, biology, and mechanics. For instance, we have recently reported a Bayesian experimental autonomous researcher (BEAR) that combines 3D printing and automated testing to realize structural materials that have highly tuned non-linear mechanical performance. While this approach was initially a black-box process, mechanics insight in the form of finite element analysis (FEA) contains critical insight that can in principle be useful, despite not completely capturing experimental performance. Here, we describe recent efforts to make the BEAR physics-informed through the incorporation of FEA. The interplay of high fidelity experimentation and comparatively low fidelity but high throughput simulation makes this an interesting case study for how best to efficiently use these disparate data streams for structural optimization.
8:50 AM
Solving Inverse Problems for Process-structure Linkages Using Asynchronous Parallel Bayesian Optimization: Anh Tran1; Tim Wildey1; 1Sandia National Laboratories
In this talk, given a target microstructure, we present an active learning high-throughput microstructure calibration framework to derive a set of processing parameters, which can produce an optimal microstructure that is statistically equivalent to the target microstructure. The problem is formulated as a multi-objective optimization problem. Each objective function measures the difference using the same microstructure descriptor between a candidate and a target microstructure. To significantly reduce the physical waiting wall-time, we enable the high-throughput feature by an asynchronous parallel Bayesian optimization to exploit high-performance computing resources. Case studies in additive manufacturing and grain growth are demonstrated with kinetic Monte Carlo (kMC) simulation as a forward predictive model. For a given target microstructure, the target processing parameters are successfully recovered. Tran, A., Mitchell, J. A., Swiler, L., & Wildey, T. (2020). An active learning high-throughput microstructure calibration framework for solving inverse structure-process problems in materials informatics (194) 80-92, Acta Materialia.
9:10 AM
Model Reification with Batch Bayesian Optimization: Richard Couperthwaite1; Danial Khatamsaz1; Abhilash Molkeri1; Douglas Allaire1; Ankit Srivastava1; Raymundo Arroyave1; 1Texas A&M University
The integration of computational materials simulations with experiments is a key component of Integrated Computational Materials Engineering. Recent advances in high-throughput experimental work have opened the possibility of testing more material properties in a parallel manner. Much of the experimental work in this regard has relied on the combinatorial exploration of the design space which is not an efficient approach. Recent advances in Batch Bayesian Optimization provide a means for generating batch predictions while also removing the challenging step of determining the surrogate model hyper-parameters. The current work proposes a method to combine a multi-scale model fusion approach with a Batch Bayesian Optimization method to provide the opportunity to predict design space parameter batches in a guided manner. This approach has shown promise for reducing the cost and time required to optimize material properties or design novel materials.
9:30 AM Invited
Structural Response Statistics of Deformed Polycrystals Leading to Rare Events: Curt Bronkhorst1; Peter Marcy2; Hansohl Cho3; Scott Vander Wiel2; Satyapriya Gupta1; Veronica Anghel2; George Gray2; 1University of Wisconsin, Madison; 2Los Alamos National Laboratory; 3Korea Advanced Institute of Science and Technology
We know qualitatively that the structure of metallic polycrystals has a strong impact on its response to mechanical loading. The heterogeneous nature of aggregate composite materials leads to highly statistical structural response at length scales related to grain size and dislocation interaction. A two-times difference between minimum and maximum von Mises stress for deformed cubic materials is typical. The von Mises stress variability is also very high in the vicinity of grain boundaries and triple lines and remains so for several microns away. Peach-Koehler force anomalies such as split screw dislocation core found in some materials leading to non-Schmid effects alters the statistical response of materials. Grain size effects can also alter the statistical response of materials in ways which we a just now beginning to explore. Quantitative results using advanced theories for granular behavior will be presented. We will also discuss implications for engineering models and material design.
10:00 AM Invited
Data-driven Approaches for Automated Analysis of Non-metallic Inclusions that Form during Steel Processing: Mohammad Abdulsalam1; Nan Gao1; Elizabeth Holm1; Bryan Webler1; 1Carnegie Mellon University
Non-metallic inclusions are small oxide, sulfide, or nitride particles that have numerous effects on the processing and properties of steels. Inclusions arise during liquid steel processing and their control is an important objective of steel refining. Improved control strategies have been enabled by improved characterization methods. Automated characterization methods provide multidimensional data on shape, size, and chemical composition for thousands of inclusions on time scales of hours. Measurement and analysis times are, however, still too long for direct process feedback. This talk will review efforts using machine learning and computer vision methods to automate data analysis and increase the speed of data acquisition. We utilize clustering algorithms to identify large agglomerations of inclusions that are particularly detrimental to processing and properties. We have also investigated regression and computer vision methods to extract composition information from images of inclusions to reduce the need for direct composition measurement.
10:30 AM
Incorporating Historical Data & Past Analyses for Improved Tensile Property Prediction of 9% Cr Steel: Madison Wenzlick1; Ram Devanathan2; Osman Mamun2; Kelly Rose3; Jeffrey Hawk3; 1Leidos Research Support Team for the National Energy Technology Laboratory; 2Pacific Northwest National Laboratory; 3National Energy Technology Laboratory
Integrating published data-driven results & tools into future analyses strengthens our understanding of materials systems and reduces redundancy. However, the results of machine learning analyses can be difficult to generalize from and interpret. Key methods, assumptions and data limitations are often not included. Additionally, the quality of data published in literature is highly varied. This work explores the challenges in reproducing clustering and regression analyses from literature published using the same dataset, the 9Cr Database generated through DOE’s eXtremeMAT project. This work investigates attributes of data science methods including data manipulation and processing steps, programming tools & functions used, and key assumptions made during analysis that are necessary for reproducing results. These analyses focus on correlation, clustering, and regression methods to drive insights into processing, composition, microstructure impact on tensile strength and yield stress and improve the efficiency of alloy design.