6th World Congress on Integrated Computational Materials Engineering (ICME 2022): Artificial Intelligence & Machine Learning I
Program Organizers: William Joost; Kester Clarke, Los Alamos National Laboratory; Danielle Cote, Worcester Polytechnic Institute; Javier Llorca, IMDEA Materials Institute & Technical University of Madrid; Heather Murdoch, U.S. Army Research Laboratory; Satyam Sahay, John Deere; Michael Sangid, Purdue University
Monday 10:30 AM
April 25, 2022
Room: Regency Ballroom DE
Location: Hyatt Regency Lake Tahoe
10:30 AM Invited
Materials Discovery via Machine Learning: Modeling Across Properties and Uncertainty Predictions: Francesca Tavazza1; 1National Institute of Standards and Technology
Next generation materials discovery is heavily dependent on the use of Machine Learning (ML). Key ingredients in ML models are well curated data, descriptors and appropriate algorithms. In this talk, we discuss the use of ML modeling in materials discovery for crystalline materials. Using density functional theory-generated crystalline materials data as training, we will compare descriptors and ML algorithms effectiveness across physical properties. In addition to predicting property, it is essential to assess the reliability of the ML model through uncertainty evaluation. While uncertainty on an estimated population variable are commonly reported for ML models, through quantities like mean absolute error or root mean square error, the uncertainty of each prediction is rarely evaluated. In this talk we address this issue by comparing different ways of determining the error on single predictions for a variety of material properties.
11:00 AM Cancelled
Physics Informed Neural Networks for Modeling the Thermomechanical Properties of Additively Manufactured Metals: Doyl Dickel1; Sungkwang Mun1; Sara Adibi1; Matthew Priddy1; Linkan Bian1; 1Mississippi State University
Simulation of the additive manufacturing process can allow for the optimization of processing parameters and real-time correction of additive builds. However, accurate computational techniques to model the thermal history and mechanical properties of a build are prohibitively expensive, often requiring hours or days on modern computer architectures. Innovations in automatic differentiation have greatly improved the ability of artificial neural networks to solve complicated boundary value PDEs, including those used in materials modeling. Physics informed neural networks (PINNs), which take advantage of these capabilities to generate solutions which enforce the physical equations governing the system, can be a useful tool to accelerate computational tools in this space. We present here the theory underlying PINNs to model the thermomechanical behavior in additive builds and show its ability to reproduce solutions using much more costly methods for the thermal history and mechanical behavior of a few sample additive builds.
11:20 AM
AI-based High-Throughput Screening Framework for Battery Materials Design: Alina Negoita1; Nasim Souly1; Alina Negoita1; Prateek Agrawal1; Christian Tae1; Vedran Glavas2; Julian Wegener2; Kai Gerstner2; Alex Alekseyenko3; 1VW GoA; 2VW AG; 3Audi oA
Identification of the optimal combination of parameters in materials design requires large effort since many possible candidates have to be evaluated by experiments or simulations. Creating synthetic materials already decreases the effort significantly, but one still needs to focus on a reduced selection of material combinations. We propose a data-driven high-throughput screening framework for materials design that allows materials screening of millions of candidates in only milliseconds per prediction. We demonstrate the feasibility of our approach on the design of battery cathode materials combining synthetic microstructure generation and electrochemical modeling considering also battery cell properties. We apply simple Machine Learning models on averaged microstructure properties for materials screening and complex Deep Learning models on the 3D microstructures to enable generative materials design. We use model uncertainty to efficiently create new simulation data samples for incremental model improvement. The most promising candidates selected by materials screening are then validated with simulations.
11:40 AM Invited
Advancements in EBSD Using Machine Learning: Kevin Kaufmann1; Chaoyi Zhu1; Hobson Lane1; Kenneth Vecchio1; 1University of California, San Diego
Electron backscatter diffraction (EBSD) is a powerful tool with the ability to collect diffraction patterns over large areas with relatively small step sizes, thus supporting multi-scale analysis. After EBSD pattern collection, current indexing techniques (whether Hough-based or dictionary pattern matching based) are capable of reliably differentiating between a user selected set of phases, if those phases contain sufficiently different crystal structures. Despite considerable efforts, the challenges of phase differentiation and identification remain. Recent improvements in EBSD detectors allow for unprecedented pattern collection rate and resolution, opening the door for implementing techniques from the data science field. This work demonstrates the application of convolutional neural networks for extracting crystallographic and chemical information from the information rich diffraction patterns and compares the results of this approach with solutions offered in commercial systems. Investigations into the internal mathematical operations of the “black box” algorithm operating on EBSD patterns will also be discussed.
12:10 PM
Prediction of Corrosion Behaviour of Additively Manufactured Nickel Based Super Alloy Using Machine Learning: Mythreyi Venkataramana1; Rohith Srinivaas M1; R Jayaganthan1; 1IIT Madras
This research work focuses on machine learning assisted prediction of corrosion behavior of selective laser melted (SLM) and post processed INCONEL 718. Inconel 718, a nickel based super alloy, fabricated using SLM technique with optimized process parameters was subjected to post processing treatments such as heat treatment and shot peening. Corrosion testing was performed in these specimens in both as built and post processed conditions using electrochemical techniques. Potentiodynamic polarization and electrochemical impedance spectroscopy analysis were employed to test the corrosion behavior in a 3.5 wt% NaCl environment. Corrosion data from these experiments were fit into various machine learning algorithms and the prediction models were built. The prediction efficiency of the built models was assessed by comparing the experimental and predicted results.The models’ performance was evaluated by standard metrics. The feature importance analysis was executed in order to determine the post processing parameters that influenced the corrosion behavior the most.