AI/Data Informatics: Computational Model Development, Validation, and Uncertainty Quantification: Session IV
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Computational Materials Science and Engineering Committee
Program Organizers: Saurabh Puri, VulcanForms Inc; Francesca Tavazza, National Institute of Standards and Technology; Dennis Dimiduk, BlueQuartz Software LLC; Darren Pagan, Pennsylvania State University; Kamal Choudhary, National Institute of Standards and Technology; Saaketh Desai, Sandia National Laboratories; Shreyas Honrao, Aionics; Ashley Spear, University of Utah; Houlong Zhuang, Arizona State University
Tuesday 2:30 PM
March 21, 2023
Room: Cobalt 520
Location: Hilton
Session Chair: Praveen Kumar, Indian Institute of Science
2:30 PM
Closed-loop Discovery of Materials with Simultaneous Electronic and Mechanical Property Targets: Christopher Stiles1; Elizabeth Pogue1; Alexander New1; Brandon Wilfong2; Gregory Bassen2; Izze Hedrick2; Edwin Gienger1; Christine Piatko1; Janna Domenico1; Kyle McElroy1; Timothy Montalbano1; Michael Pekala1; Nam Le1; Christopher Ratto1; Andrew Lennon1; Tyrel McQueen2; 1Johns Hopkins University Applied Physics Laboratory; 2Johns Hopkins University
Machine learning (ML) techniques present tremendous opportunities to accelerate materials design and discovery. However, significant developments are required to adapt approaches from other domains. For example, ML models for materials must generally contend with sparser and more inhomogeneous data. These challenges compound for practical tasks that require simultaneous optimization of multiple properties. We present results from a “closed-loop” approach integrating ML model predictions with experimental synthesis and characterization that provide new data to update our models. We first demonstrated success in discovery of superconducting compounds using information from the Materials Project and Supercon databases, together with data from in-house synthesis and characterization of crystal structure and critical temperature, to train our ML models. Building on this framework, we have expanded our approach to include prediction of mechanical properties, characterized experimentally using high-throughput nanoindentation. Our work demonstrates the promise of ML techniques applied toward materials discovery while optimizing multiple properties simultaneously.
2:50 PM
An Information Theory Based Approach for Training Machine Learned Potentials: Jason Gibson1; Jan Janssen1; Laura Lopes1; Richard Hennig2; Danny Perez1; 1Los Alamos National Laboratory; 2University Of Florida
The promise of obtaining accuracy on par with first-principle calculations at the computational cost of empirical potentials has made machine-learned interatomic potentials attractive alternatives for studying and characterizing materials. However, while many machine-learned potentials have reported accuracy within several meV/atom of first principle reference data, these errors represent only configurations that do not significantly differ from the training. In comparison, performance on truly novel configurations can incur errors orders of magnitude larger.This work leverages more than 7M atomic environments in tungsten that have been optimized to maximize its informational entropy in feature space, ensuring broad coverage compared to hand-crafted datasets. First, we investigate the dependence of the test errors on the training set size for various popular machine-learned potentials and delineate data-bound and model-bound regimes. We then compare different strategies to optimally sub-select training data from this large dataset to maximize the transferability/cost tradeoff of the resulting potentials.
3:10 PM
Extraction of Creep Parameters from Indentation Creep Experiment: An Artificial Neural Network-Based Approach: Raj Mahat1; Vikram Jayaram1; Praveen Kumar1; 1Indian Institute of Science
Indentation creep is a faster alternative to conventional uniaxial creep testing with an added advantage of small probing volume; however, interpretation of indentation creep data in terms of uniaxial creep response is challenging due to the complex stress field below the indenter. Here, a fully-connected sequential multi-layered artificial neural network (ANN), trained using finite element (FE) indentation creep simulations, was used to map the displacement-time response obtained from indentation creep experiments to the corresponding uniaxial creep parameters. ANN was trained using a back-propagation algorithm based on a batch-gradient descent to map the indentation displacement-time inputs and the uniaxial creep parameters used in the FE simulations. The trained ANN was tested on the nanoindentation creep data of commercial purity Pb at room temperature, and its prediction was compared with the uniaxial creep parameters. A match between the experimental data and the ANN prediction for stress exponent and time exponent was noted.
3:30 PM
Interlaced Characterization and Calibration: Online Bayesian Optimal Experimental Design for Constitutive Model Calibration: Denielle Ricciardi1; Tom Seidl1; Brian Lester1; Amanda Jones1; Elizabeth Jones1; 1Sandia National Laboratories
There has been a growing reliance on computational simulations to make important engineering decisions. The calibration of these complex material models is an essential step of material certification; however, existing calibration approaches can be data and time intensive, contributing to an elongated material discovery process. This work reconsiders the calibration process with the proposed Interlaced Characterization and Calibration (ICC) framework for the calibration of phenomenological constitutive models for solid mechanics. Bayesian optimal experimental design is used to perform inference on model parameters and to determine the next best experiment to conduct to reduce parameter uncertainty. Dimension-reduced, full-field digital image correlation data is generated under heterogeneous loading conditions and is used to calibrate the model. The data collection process and inference are performed iteratively in a feedback loop to drive the calibration to an optimal state.SNL is managed and operated by NTESS under DOE NNSA contract DE-NA000352.
3:50 PM
Machine Learning-based Multi-objective Optimization for Efficient Identification of Crystal Plasticity Model Parameters: Marko Knezevic1; 1University of New Hampshire
This paper presents a Pareto-based multi-objective machine learning methodology for efficient identification of crystal plasticity model parameters. Specifically, the methodology relays on a Gaussian processes-based surrogate model to limit the number of calls to a given model, and, consequently, to increase the computational efficiency. The constitutive parameters pertaining to an Elasto-Plastic Self-Consistent model including a dislocation density-based hardening law, a backstress law, and a phase transformations law are identified for two alloys, a dual-phase (DP) steel, DP780, subjected to load reversals and a stainless steel (SS), 316L, subjected to strain-rate and temperature sensitive deformation. The optimization objectives were the quasi-static flow stress data for the DP steel case study, while a set of strain-rate and temperature sensitive flow stress and phase volume fraction data for the SS case study. The procedure and results for the two case studies are presented and discussed illustrating advantages and versatility of the methodology.
4:10 PM Break
4:30 PM
Robust and Efficient Method for Calibration of Thermal Models for Additive Manufacturing: Michael Groeber1; Joy Forsmark1; Yang Huo1; 1The Ohio State University
Fasting acting models for additive manufacturing are in great demand, in part to be used in computational approaches for designing/optimizing processing strategies and parameters. However, there are questions regarding there accuracy, domain of utility, and how best to calibrate and validate them. In this work, we will present a systematic approach to calibrating and validating one such model, with detailed discussion on the challenges encountered. Specifically, we will focus on methods that are practical for engineering implementation in the vision of ICME.
4:50 PM
A Deep Neural Network Formulation for Anisotropic Yield Prediction: Anderson Nascimento1; Sharan Roongta2; Martin Diehl3; Irene Beyerlein1; 1University of California, Santa Barbara; 2Max-Planck-Institut für Eisenforschung; 3Katholieke Universiteit, Leuven
At the continuum level, the plastic anisotropy of a wide range of metals and alloys is well described by advanced phenomenological yield surfaces. Relevant difficulties in their usage, however, are associated with the non-trivial parameter identification process and the non-uniqueness of the anisotropy coefficients, commonly noticed in practice and reported in the literature. Alternative avenues for plastic flow prediction have been studied, and machine learning based approaches have gained notoriety due to their high fitting capabilities. A deep neural network based surrogate model, trained with virtual stress data points and with performance comparable to advanced phenomenological yield functions has been developed. Important features such as well defined flow vector, convexity and yield prediction are studied and compared against benchmark yield criteria.
5:10 PM
Training Material Models Using Gradient Descent Algorithms: Tianju Chen1; Mark Messner1; 1Argonne National Laboratory
Oftentimes, calibrating accurate models devolves into the problem of fitting the model parameters against experimental test data. Here, we present the Pyopmat package, an open source framework for calibrating constitutive models against experiment data subjected to various loading conditions using machine learning techniques. The package calculates the exact gradient of the model response with respect to the parameters using a combination of automatic differentiation and the adjoint method. Given this exact gradient, we compare the performance of several gradient-based optimization techniques in fitting realistic constitutive models against data. We demonstrate the efficiency and accuracy of our package through example problems using both synthetic data, generated using known parameter sets, under uniaxial and cyclic loading conditions and also actual high temperature creep-fatigue test data. Besides, Bayesian statistical framework is also embedded in Pyoptmat. The uncertainty of the experiment data are well quantified through utilizing stochastic variational inference.