AI/Data informatics: Tools for Accelerated Design of High-temperature Alloys: Uncertainty Quantification, AI Tools, and Environmental Degradation
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Integrated Computational Materials Engineering Committee, TMS: Computational Materials Science and Engineering Committee
Program Organizers: Michael Titus, Purdue University; Pinar Acar, Virginia Tech; Andrew Detor, DARPA/DSO; James Saal, Citrine Informatics; Dongwon Shin, Oak Ridge National Laboratory
Wednesday 8:30 AM
March 17, 2021
Room: RM 33
Location: TMS2021 Virtual
Session Chair: Michael Titus, Purdue University; James Saal, Citrine Informatics
8:30 AM Invited
Domain and Uncertainty Quantification in Machine Learning Models of Alloy Properties: Dane Morgan1; Ryan Jacobs1; Benjamin Blaiszik2; 1University of Wisconsin-Madison; 2University of Chicago
Machine learning methods have the potential to dramatically expand alloy property databases through training on existing data and predicting properties for new compositions. However, it is essential to quantify the domains and uncertainties of machine learning models to develop reliable databases. In this talk, we explore a common Bayesian (Gaussian process regression) and ensemble (random forest decision trees) method for assessing domains and uncertainties using alloy diffusion data as an example. We show that Gaussian process regression is better at determining model domain than random forest, but that the random forest error bars are more accurate. We also describe the MAterials Simulation Toolkit – Machine Learning (MAST-ML) and Foundry environments that support efficient data management, machine learning model development, and accessible model dissemination.
9:00 AM
Domain Knowledge-informed, Process-mapping AI Graph for Designing Fe-based Alloys: Vyacheslav Romanov1; 1National Energy Technology Laboratory
Continuous improvement in efficiency of a power plant relies on designing materials for use at increasingly higher temperature and/or pressure, for 100,000s hours of operation. Due to complexity, non-linearity and high-dimensionality of the problem, traditional Machine Learning (ML) approaches require unreasonably large datasets for the data-driven model development. Science-based material and process engineering complements hard data with, sometimes soft and intuitive, empirical domain knowledge. Artificial Intelligence (AI) was used in this study to incorporate such knowledge into computational graph architecture (process-mimicking artificial neuron design, causal layer and graph structures, ensemble modeling of latent states) and learning procedures (variable transformation, fuzzy physics pre-training and freezing of deep layers, virtual microstructure representation, and adversarial multi-objective optimization). The first alloys design pathways suggested by the AI tool (pyroMind) passed a preliminary engineering review on soundness and transparency.
9:20 AM Invited
Toward High Throughput Design and Development of Multi-principal Element Alloys for Corrosion and Oxidation Resistance (MPEAs): Mitra Taheri1; Todd Hufnagel1; Chris Wolverton2; James Rondinelli2; Jason Hattrick-Simpers3; Brian DeCost3; Elizabeth Opila4; John Scully4; Jean-Philippe Couzinie5; Nick Birbilis6; 1Johns Hopkins University; 2Northwestern University; 3NIST; 4University of Virginia; 5University Paris-Est Créteil (UPEC); 6Australian National University
Multi-Principal Element Alloys (MPEAs) are the subject of emerging interest due to their compositional profile, which holds the promise of superior mechanical properties and thermal stability. It is critical to understand the atomic to mesoscale tuning parameters for MPEAs to harness critical properties, such as corrosion/oxidation resistance, for coatings and extreme applications. With millions of permutations of MPEAs in existence, however, it’s virtually impossible to nail down the “right” combination without innovation. Recent advances in high throughput approaches present an opportunity for alloy design and testing that enabling tracking, curation, and dissemination of thousands of MPEAs. This talk reviews results from a combination of materials design, machine learning, and high throughput characterization in a team effort to (1) explore currently untapped compositional space, (2) predict and control passivation/complex oxide evolution, and (3) define alloy/corrosive environment operating parameters based on bulk and surface phenomena.
9:50 AM
Advanced Data SCiENce Toolkit for Non-data Scientists (ASCENDS)
- A Case Study of the Oxidation Kinetics of NiCr-based Alloys: Jian Peng1; Rishi Pillai1; Marie Romedenne1; Sangkeun Lee1; Govindarajan Muralidharan1; Bruce Pint1; J. Allen Haynes1; Dongwon Shin1; 1Oak Ridge National Laboratory
We introduce an easy-to-use, versatile, and open-source data analytics frontend, ASCENDS (Advanced data SCiENce toolkit for Non-Data Scientists), developed to enable data-driven materials research. The toolkit can analyze the correlation between input features and target properties, train machine learning (ML) models, and make predictions with the trained surrogate models. We introduce the use of ASCENDS to predict the oxidation kinetics of NiCr-based alloys as a function of alloy chemistry and temperature. We compare two different oxidation models (a simple parabolic law and a statistical cyclic-oxidation model) to represent the high-temperature oxidation kinetics of NiCr-based alloys in dry- and wet-air within the context of data analytics. Understanding the oxidation characteristics correlated with the features will support and promote new alloy development with further improved performance. This research was sponsored by the U. S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Vehicle Technologies Office, Propulsion Materials Program.
10:10 AM
Expanding Materials Selection via Transfer Learning for High-temperature Oxide Selection: Zachary Mcclure1; Alejandro Strachan1; 1Purdue University
Complex concentrated alloys (CCAs) with higher operating temperatures than today's current alloys can improve system performance in several applications. While the strength properties of many CCAs outperform Ni-based superalloys, the oxidation properties are not ideal. Selecting an appropriate oxide scale with high melting temperatures, thermodynamic stability, and low ionic diffusivity is critical for alloy development.While some properties exist for many oxides, available melting temperature data is limited. The determination of melting temperatures is time consuming and costly, both experimentally and computationally. Instead we use data science tools to develop predictive models from existing data. The relatively small number of available melting temperature values precludes the use of standard tools; therefore, we use a multi-step approach via sequential learning where first principles data is leveraged to develop more appropriate models. The models are used to predict the desired properties for nearly 11,000 oxides and quantify uncertainties in the space.
10:30 AM Invited
Optimal Design of High-temperature, Oxidation-resistant Complex Concentrated Alloys: Alejandro Strachan1; 1Purdue University
Metallic alloys capable of maintaining high strength and oxidation resistance at high temperatures are key in aerospace and energy applications. State-of-art Ni-based superalloys are limited by their melting temperature of ~1300 C. In recent years, refractory complex concentrated alloys (RCCAs) emerged as a possible alternative, with high-temperature strength surpassing Ni superalloys. Unfortunately, their oxidation resistance is not ideal. The optimization of RCCAs is hindered by the high-dimensionality of the design space and the fact that full-scale experiments of the quantities of interest are costly and time consuming. I will discuss the combination of multi-fidelity experiments and physics-based modeling with machine learning tools with the ultimate goal of designing RCCAs with unprecedented combination of high-temperature strength and oxidation resistance. Specifically, I will discuss the integration of information from disparate sources and with different uncertainties into predictive models and the use of surrogate models to reduce the number of full-scale experiments required.