Algorithm Development in Materials Science and Engineering: Algorithms and Machine Learning Approaches for Microscale
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Integrated Computational Materials Engineering Committee, TMS: Phase Transformations Committee, TMS: Computational Materials Science and Engineering Committee
Program Organizers: Mohsen Asle Zaeem, Colorado School of Mines; Garritt Tucker, Colorado School of Mines; Charudatta Phatak, Argonne National Laboratory; Bryan Wong, University of California, Riverside; Mikhail Mendelev, NASA ARC; Bryce Meredig, Travertine Labs LLC; Ebrahim Asadi, University of Memphis; Francesca Tavazza, National Institute of Standards and Technology
Wednesday 8:30 AM
February 26, 2020
Room: 31C
Location: San Diego Convention Ctr
Session Chair: David Rivera, Lawrence Livermore National Laboratory; Grace Gu, University of California, Berkeley
8:30 AM
Generative Deep Neural Networks for Inverse Materials Design using Backpropagation and Adaptive Learning: Grace Gu1; Chun-Teh Chen1; 1University of California, Berkeley
Machine learning (ML) has shown great potential to accelerate materials discovery and design processes in a manifold way. Most ML-based investigations were dedicated to training models to predict properties of interests from material descriptors. After training, ML models were used as filters to explore the material design space in a brute-force manner. However, this brute-force screening methodology can only be applied to problems with a small design space. Here, we present Generative Inverse Design Networks (GIDNs), a general-purpose inverse design approach using backpropagation and adaptive learning. The novelty of the proposed approach is that it is integrated with random search in the design process to overcome the local minimum problem paired with adaptive learning to improve the performance of superior designs and to reduce the amount of training data needed to do so. Results show that GIDNs outperform other common optimization methods including gradient-based topology optimization and genetic algorithms.
8:50 AM
Development of an Evolutionary Deep Neural Net for Materials Research: Nirupam Chakraborti1; Swagata Roy1; 1Indian Institute Of Technology
Modeling and optimization in many materials related problems routinely involve noisy, nonlinear data from diverse sources. This novel algorithm, now tested on several problems, eliminates noise and extracts the meaningful trends from such data, using some multi-objective evolutionary algorithms, instead of the existing training methods. Some small Neural Nets with flexible topology and architecture are fed with random subsets of the problem variables, ensuring that each variable is used at least once. They evolve through a tradeoff between two conflicting requirements that they should be of maximum accuracy and at the same time of minimum complexity, defined through the number of parameters used. Mathematically this leads to a Pareto-optimal problem, and the evolutionary algorithms that are used to train them are geared to handle that. These subnets are then assembled using a number of hidden layers; a linear least square algorithm is used for the optimization of the associated weights.
9:10 AM
Persistent Homology: Unveiling the Topological Features in Materials Data: Chaitali Patil1; Lucas Magee1; Supriyo Chakraborty1; Yusu Wang1; Stephen Niezgoda1; 1The Ohio State University
Persistent homology and persistence diagrams (PD) are important tools in a topological data analysis of large data sets to help effectively represent and analyze the underlying structures and features behind them. Such analysis is also useful to find the good descriptors for the machine learning. PDs are now being applied in many different research areas ranging from finance to biology. Specifically for materials science, new approaches had been proposed to classify and analyze defects, porosities, atomic structures and microstructures by using the PDs. Here, we present a case study for analyzing deformation behavior of the highly anisotropic material such as zirconium using the PDs. Zirconium was deformed under uniaxial compression at the room temperature using a 3D fast Fourier transform-based elasto-viscoplastic (FFT-EVP) crystal plasticity model. Effect of different textures on the evolution of the micromechanical fields was studied by means of PDs.
9:30 AM
Inverse Solutions Based on Reduced-order Process-structure-property Linkages Using Markov Chain Monte Carlo Sampling Algorithms: Yuksel Yabansu1; Almambet Iskakov1; Anna Kapustina2; Sudhir Rajagopalan3; Surya Kalidindi1; 1Georgia Institute of Technology; 2Siemens AG; 3Siemens Corporate Technology
Deployment of advanced engineering materials in a commercial product can take multiple decades from the initial discovery. Process-structure-property (P-S-P) linkages play a critical role in designing the advanced engineering materials and they have been fairly well established since microstructure informatics tools became an integral part of building the linkages. P-S-P linkages where the flow of information occurs from process to property through material structural measures are deductive methodologies and are called forward P-S-P linkages. However, material design requires a goal/target oriented approach which aims to find the suitable processing/manufacturing conditions that correspond to tailored properties. Inverse solutions to P-S-P linkages depend on the accuracy and efficacy of the deducted information from forward P-S-P linkages. This study presents a novel framework that utilizes sampling algorithms to establish an inverse approach to forward P-S-P linkages for materials design.
9:50 AM
Calibrating Strength Model Parameters using Multiple Types of Data: Jeffrey Florando1; Jason Bernstein1; Amanda Muyskens1; Matthew Nelms1; David Rivera1; Kathleen Schmidt1; Nathan Barton1; Ana Kupresanin1; 1Lawrence Livermore National Laboratory
Bayesian calibration is a common method for estimating model parameters and quantifying their associated uncertainties; however, calibration becomes more complicated when the data arise from different types of experiments. In the case of material strength, additional types of data are often needed to access higher experimental strain rates. For strength models, it is desirable for parameter estimates to be valid across this range of regimes, especially if there is not expected to be a transition in the underlying physical mechanisms involved in the strength response. Here, we employ different types of data: stress-strain curves from low-strain rate experiments and deformed profiles from higher-strain rate experiments. In particular, we highlight data fusion techniques for incorporating different measurement types into Bayesian calibration and present our results for tantalum. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.
10:10 AM Break
10:30 AM
Advances in a Phase Field Dislocation Dynamics Model to Account for Various Gamma-surfaces of Hexagonal Close Packed Crystallography: Claire Weaver1; Abigail Hunter2; Anil Kumar2; Irene Beyerlein1; 1University of California Santa Barbara; 2Los Alamos National Laboratory
Until recently, the use of a phase field model to simulate dislocation mediated plasticity was applied only to cubic crystallographic materials. We present enhancements to the phase field dislocation dynamics formulation in order to consider hexagonal close packed metals. The enhancements are directly informed with density functional theory and allow the model to consider slip-mode dependent energetics and their influence on dislocation behavior. We present results for multiple slip modes active in hcp metals.
10:50 AM
Hierarchical Integration of Atomistically-derived Dislocation Mobility Laws into Discrete Dislocation Dynamics Simulations: Darshan Bamney1; Khanh Dang2; Laurent Capolungo2; Douglas Spearot1; 1University of Florida; 2Los Alamos National Laboratory
Empirical piecewise dislocation mobility functions derived from atomistic simulations of straight dislocations are incorporated into a discrete dislocation dynamics (DDD) framework to improve the representation of the physics of dislocation motion. Mobility functions for different dislocation character angles, which account for linear phonon damping and non-linear radiative damping regimes, are integrated via a linearization strategy fully compliant with the thermodynamic equation of motion used in DDD. In addition, the role of non-Schmid stresses on dislocation mobility is incorporated via the same framework. Results show excellent agreement between dislocation loop kinetics simulated by atomistics and DDD. For instance, modified mobility due to the local stress state and faceting due to modified core structure, as observed in atomistics, are captured by DDD simulations. Large-scale DDD simulations of plastic deformation in Al will be presented to highlight the importance of the new mobility laws in characterizing the elasto-plastic transition in mesoscale plasticity models.
11:10 AM
A Multi-GPU Implementation of a Full-field Crystal Plasticity Solver for Efficient Modeling of High-resolution Microstructures: Adnan Eghtesad1; Kai Germaschewski1; Ricardo A Lebensohn2; Marko Knezevic1; 1University of New Hampshire; 2Los Alamos national laboratory
In this research, we present a high-performance implementation of a full-field elasto-visco-plastic fast Fourier transform (EVPFFT) crystal plasticity solver to take the advantage of graphics processing units (GPUs) across nodes of a supercomputer. To this end, the implementation combines the OpenACC programming model for GPU acceleration with MPI for distributed computing. Moreover, the FFT calculations are performed using the efficient Compute Unified Device Architecture (CUDA) FFT library, called CUFFT. Finally, to maintain performance portability, OpenACC-CUDA interoperability for data transfers between CPU and GPUs is used. The overall implementations are termed ACC-EVPCUFFT for single GPU and MPI-ACC-EVPCUFFT for multiple GPUs. To facilitate performance evaluation studies of the developed computational framework, deformation of a single phase copper is simulated, while to further demonstrate utility of the implementation for resolving fine microstructures, deformation of a dual-phase steel DP590 is simulated. The implementations and results are presented and discussed here in.
11:30 AM
A Self-consistent Parametric Homogenization Framework for Fatigue in Ni-based Superalloys: George Weber1; Somnath Ghosh1; Maxwell Pinz1; 1Johns Hopkins University
A framework to determine higher-scale constitutive laws as functions of microstructural properties is developed for Ni-based superalloy Rene88-DT γ-γ’ microstructures and is generalizable to most multiscale problems. A dislocation density-based crystal-plasticity finite-element model, at the scale of subgrain microstructures containing γ’-precipitates in a γ-matrix within a single crystal, is implemented to incorporate micromechanical effects of precipitate-precipitate interactions. This model is complemented with a stochastic, statistically-equivalent microstructure generator that instantiates meshed microstructures from distributions of subgrain morphological descriptors. These instantiations are embedded in a homogenized medium and simulated with the crystal plasticity model, employing a handshake region to accurately capture evolving boundary conditions near the homogeneous-heterogeneous boundary. Self-consistency is achieved through a directed search of the parameter space of the homogenized model, informed by a reduced-order machine learning methodology. This framework effectively incorporates the lower-scale mechanics into the higher-scale simulation, providing insights into multiscale mechanisms and design implications for materials processing.