Computational Thermodynamics and Kinetics: AI and ML
Sponsored by: TMS Functional Materials Division, TMS Materials Processing and Manufacturing Division, TMS: Chemistry and Physics of Materials Committee, TMS: Computational Materials Science and Engineering Committee, TMS: Integrated Computational Materials Engineering Committee, TMS: Solidification Committee
Program Organizers: Hesam Askari, University Of Rochester; Damien Tourret, IMDEA Materials Institute; Eva Zarkadoula, Oak Ridge National Laboratory; Enrique Martinez Saez, Clemson University; Frederic Soisson, Cea Saclay; Fadi Abdeljawad, Lehigh University; Ziyong Hou, Chongqing University
Wednesday 2:00 PM
March 22, 2023
Room: 26A
Location: SDCC
Session Chair: Mahmood Mamivand, Boise State University; Edwin Garcia, Purdue University
2:00 PM
Physics-Informed Machine Learning of the Thermodynamics and Kinetics of Point Defects in Alloys: Anjana Talapatra1; Anup Pandey1; Danny Perez1; Blas Uberuaga1; Ghanshyam Pilania1; 1Los Alamos National Laboratory
Desirable properties of multi-component alloys, such as corrosion, high-temperature oxidation and irradiation resistance, are highly sensitive to the formation and migration of point defects such as vacancies and interstitials. Point defect formation, diffusion and the associated energy barriers are governed by the interactions between individual and/or groups of atoms. In this work, we use Machine learning (ML) algorithms in tandem with molecular dynamics based Nudged Elastic Band (NEB) calculations to learn the composition- and configuration-dependent formation energies and migration barriers for vacancies. Specifically, we train deep neural network models using numerical representations of the local configurational environment and also implement a physics-informed approach ensuring that the detailed balance criteria are obeyed. We discuss various scenarios including i) comparison of results using relaxed as well as unrelaxed geometries, ii) multiple methods to implement the detailed balance criteria and iii) compositional complexity by comparing defect energetics in Cu-Ni and Cu-Au alloys.
2:20 PM
Prediction of High-temperature Elasticity of Tungsten Using Machine Learning and Data-driven Approach: Anruo Zhong1; Clovis Lapointe1; Alexandra Goryaeva1; Jacopo Baima1; Manuel Athènes1; Mihai-Cosmin Marinica1; 1Universite Paris-Saclay, CEA
By the means of machine learning and data-driven approaches, we investigate the elastic properties of tungsten, a ubiquitous material for future energetic systems [1], up to the melting temperature. We are able to explore the atomistic energy landscape of metals with ab initio accuracy up to the melting temperature. The present workflow, which combines the machine learning force fields [2] and the robust free energy sampling, is emphasized in the body-centered cubic tungsten and validated by the available experimental findings at low temperatures. Moreover, we are able to predict elastic properties of tungsten in the range of temperatures that cannot be addressed experimentally due to its high melting point, from 2100 K up to the melting point. [1] K. Arakawa, M.-C. Marinica et al. Nature Mater. 19, 508 (2020) [2] A. M. Goryaeva et al. Phys. Rev. Materials 5, 103803 (2021).
2:40 PM
Diffusivity in a Multicomponent Alloy Using Machine Learning and Variational Approaches: Dallas Trinkle1; Soham Chattopadhyay1; 1University of Illinois at Urbana-Champaign
The diffusivity of multicomponent or high-entropy alloys helps determine the ultimate range of stability of their solid solution phase. Computing the diffusivity of multicomponent alloys presents a challenge to sample the state space accurately and adequately while determining the transport coefficients. The large number of components suggests a density-functional theory-based approach, but the large state space makes kinetic Monte Carlo unattractive. Here, we use a variational method for the transport coefficients to consider two approaches that require orders-of-magnitude fewer calculations: a mean-field Green function method, and a novel hybrid machine-learning method. The methodologies are tested using classical potential simulations of high-entropy alloys, to quantify the computational savings as well as expected accuracy. Moreover, as both approaches are variational, they serve as upper limits to the true diffusivity; this can allow for more rapid screening in alloy design.
3:00 PM Invited
Chemistry and Processing History Prediction from Microstructure Morphologies: Mahmood Mamivand1; Amir Abbas Kazemzadeh Farizhandi1; 1Boise State University
Designing targeted microstructures has been a long-lasting problem in the materials science community. Historically experimental approaches and recently high-fidelity models have been used to address this challenge in a trial-and-error way. However, the inverse design has been out of reach. The primary motivation of this work is to develop a model that can look at a microstructure morphology and predict its chemistry and processing history. As a case study, we have focused on spinodal decomposition in FeCrCo alloy. We have developed a fused-data deep learning network that reads the Fe-based morphology and predicts how much initial Cr and Co along with what heat treatment temperature and time are needed to get that microstructure. The model is successfully validated against an experimental transmission electron microscopy (TEM) micrograph, even though the model is trained on synthetic data and the TEM image did not have the training data quality.
3:30 PM Break
3:50 PM
Rapid Machine Learning Estimation of Grain Boundary Segregation Vibrational Entropy Spectra in Dilute Polycrystals: Nutth Tuchinda1; Christopher Schuh1; 1Massachusetts Institute of Technology
The thermodynamics of grain boundary segregation must be quantitatively understood to design stabilized nanocrystalline alloys. While grain boundary segregation is a common phenomenon in polycrystalline materials, there is limited understanding of vibrational entropy effects for polycrystals with a multitude of solute segregation sites across the periodic table. We have therefore applied a machine learning accelerated method to efficiently sample and estimate vibrational effects within the harmonic approximation for dilute binary polycrystals with available interatomic potentials. The resulting substitutional segregation energy and vibrational entropy spectra are used with a spectral isotherm to demonstrate the average vibrational effects and generate a spectral grain boundary segregation database. This database can be applied to understand materials phenomena and design alloys involving grain boundary segregation at finite temperatures.
4:10 PM Invited
Machine Learning of Phase Diagrams: Applications to Energy Materials: Jarrod Lund1; Haoyue Wang1; Richard Braatz2; Edwin Garcia1; 1Purdue University; 2MIT
By starting from experimental- and ab initio-determined phase diagrams (PDs) of materials, a machine learning (ML) method is developed to infer the free energy function for each phase. The method samples the multidimensional space of material parameters and user-defined physical constraints into a database of millions of PDs in order to identify the target material properties. The method presented herein is 1000x to 100,000x faster than currently available approaches, and defines a new paradigm on the quantification of properties of materials and devices. As an example application, the developed methodology is combined with the most widely used thermodynamic models – regular solution, Redlich-Kister, and sublattice formalisms– to infer the properties of materials for lithium-ion battery applications. Applications to battery materials, such as LCO and LFP, and liquid electrolytes such as the EC-DMC-PC systems are presented.
4:30 PM Invited
Exploring New Frontiers of Thermal Transport: A Combined First-principles and Machine Learning Approach: Rinkle Juneja1; 1Oak Ridge National Laboratory
The behavior of collective atomic vibrations, i.e., phonons, is crucial for understanding stability and thermal transport properties of materials. Given the computational challenges in thermal transport estimation of complex materials, in this talk I will introduce how data-assisted machine learning (ML) can be used for accelerated prediction of transport properties. I will demonstrate the power of ML in discovering new material behaviors and in revealing unusual connections among transport properties using physics-aware descriptors and symmetry-based arguments. I will further showcase deeper fundamental symmetry insights dictating selection rules for thermal transport and unique scattering observables, validated by inelastic neutron scattering measurements, thereby providing new avenues in predicting thermal transport behaviors of materials. R.J. acknowledges support from the U. S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division.