AI/Data informatics: Tools for Accelerated Design of High-temperature Alloys: AI Design and Thermodynamics
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
Thursday 8:30 AM
March 18, 2021
Room: RM 33
Location: TMS2021 Virtual
Session Chair: Pinar Acar, Virginia Tech; Michael Titus, Purdue University
8:30 AM Invited
Knowledge-driven Platform for Federated Multimodal Big Data Storage & Analytics: Kareem Aggour1; Vipul Gupta1; Andy Detor1; Scott Oppenheimer1; Joe Vinciquerra1; 1GE Research
One barrier to the adoption of AI for accelerating the design of advanced materials is the lack of robust mechanisms to manage the required experimental and simulation data and expert knowledge. Materials data management is particularly challenging due to the multimodal nature of the data, which can include numeric data, images, notes, and more. Further compounding the challenge is the different scales of data, from small (e.g., KB to MB) to big (e.g., TB or more). To address these challenges, GE Research has developed a knowledge-driven materials informatics platform that enables non-computer scientists to explore a knowledge graph model of the domain, to query and analyze data captured in different repositories. This talk will provide an overview of the platform, its strengths and limitations, and discuss its application to two use cases: (i) additive manufacturing process parameter optimization for a nickel-base superalloy, and (ii) the development of high entropy alloys.
9:00 AM
Exploring the Compositional Space of High Entropy Alloys via Sequential Learning: Juan Verduzco1; Zachary McClure1; David Farache1; Saaketh Desai1; Alejandro Strachan1; 1Purdue University
Refractory complex concentrated alloys (RCCAs) have shown high temperature strength surpassing superalloys and are of interest for a range of applications. Despite this interest, melting temperature values are scarce as experimental measurements are challenging. To address this challenge, we combine molecular dynamics simulations with sequential learning to develop predictive models for the melting temperature RCCAs as a function of composition and find high-melting temperature alloys. Using MD simulations as a service using nanoHUB, we create fully autonomous research workflows and show efficient exploration of the high-dimensional compositional space towards the optimal alloy composition. We will discuss challenges that arise from the stochastic nature of MD simulations and the uncertainties associated in the data-drive and physics-driven models.
9:20 AM
Uncertainty Reduction for Calculated Phase Equilibria: Richard Otis1; Brandon Bocklund2; Zi-Kui Liu2; 1Jet Propulsion Laboratory; 2Pennsylvania State University
The development of a consistent framework for Calphad model sensitivity is necessary for the rational reduction of uncertainty via new models and experiments. In the present work, a sensitivity theory for Calphad was developed, and a closed-form expression for the log-likelihood gradient and Hessian of a multi-phase equilibrium measurement was derived. A case study of the Cr-Ni system was used to demonstrate visualizations and analyses enabled by the developed theory. Criteria based on the classical Cramér–Rao bound were shown to be a useful diagnostic in assessing the accuracy of Bayesian parameter covariance estimates from Markov Chain Monte Carlo. The developed sensitivity framework was applied to estimate the statistical value of phase equilibria measurements in comparison with thermochemical measurements, with implications for Calphad model uncertainty reduction, as well as the design of new experiments.
9:40 AM
Predicting Vibrational Entropy of FCC Solids Uniquely from Bond Chemistry Using Machine Learning: Anus Manzoor1; Dilpuneet Aidhy1; 1University of Wyoming
Despite a well-recognized contribution of vibrational entropy (Svib) in the phase stability of alloys, it remains a peripheral quantity due to its high computational cost. In this work, using a combination of density functional theory (DFT) calculations and machine learning (ML), we show that the expensive Svib computations can be completely circumvented. This is possible because there exists a unique force constant (FC) – bond length relationship for every A-A and A-B bond and the influence of the alloy composition on FCs can be captured with the change in bond lengths only. The DFT database coupled with ML model allows to predict FCs between any two elements which in turn enables predicting Svib of any complex alloy thereby significantly reducing the computational costs. This work opens a new avenue to predict Svib of complex HEAs thereby making Svib as readily available as the mixing enthalpy.