AI for Big Data Problems in Advanced Imaging, Materials Modeling and Automated Synthesis: AI for Materials Design
Sponsored by: TMS Advanced Characterization, Testing, and Simulation Committee
Program Organizers: Mathew Cherukara, Argonne National Laboratory; Subramanian Sankaranarayanan, University of Illinois-Chicago; Badri Narayanan, University of Louisville

Wednesday 2:00 PM
October 12, 2022
Room: 311
Location: David L. Lawrence Convention Center

Session Chair: Mathew Cherukara, Argonne National Laboratory; Badri Narayanan, University of Louisville; Subramanian Sankaranarayanan, University of Illinois Chicago


2:00 PM  Cancelled
A Novel Training Methodology for Phase Segmentation of Steel Microstructures Using a Deep Learning Algorithm: Nikhil Chaurasia1; Shikhar Jha1; Sandeep Sangal1; 1Indian Institute of Technology Kanpur
    In this work, an efficient training methodology for the segmentation of ferrite - pearlite microstructures using UNET machine learning architecture (a semantic classifier) is presented. However, this requires a very large number of training microstructures, which are generally not available. A novel method is proposed for circumventing the above problem. First polycrystalline templates are created by simulating a 3D nucleation and growth model following Avrami kinetics. Subsequently, cropped images of pearlite and ferrite (from real microstructures) are randomly positioned on the individual grains of the polycrystalline templates, producing synthetic microstructures of varying fractions of the two constituents. A few thousand synthetic microstructures were created using a small number of cropped images. The UNET trained on the synthetic training set was tested on real ferrite-pearlite microstructures and an accuracy of about 99% is obtained, which substantiates its robustness compared to current state-of-the-art methods.

2:20 PM  
Data-driven Search for Promising Intercalating Ions and Layered Materials for Metal-ion Batteries: Shayani Parida1; C. Barry Carter1; Avanish Mishra1; Arthur Dobley2; Avinash Dongare1; 1University of Connecticut; 2EaglePicher Technologies
    An innovative combination of atomic-scale modeling and machine learning methods is used to find layered materials beyond graphite for anodes and intercalating ions beyond Li in metal-ion batteries with higher power efficiencies. A dataset is created using density functional theory (DFT) calculations to estimate the theoretical capacities and voltages for various metal ions intercalated in layered materials. A gradient boosting decision trees (GBDT) classifier is developed to screen for 2D material and ion combinations based on voltages and structural deformations/transformations. Further, a regression model is developed to predict the binding energies of the feasible layered anodes and intercalating species. The study reveals the importance of elemental features to predict the binding properties of intercalating species for a given layered material. The framework of the approach, the ML algorithm, and the discovery of layered materials as anodes for the next generation metal ion batteries will be discussed.

2:40 PM  Cancelled
Hybrid GNN Approach to Industrial Time Series and IoT Applications: Atish Bagchi1; 1SPSA DIGITAL
    This research proposed a hybrid forecasting & classification model which combines the expressive and extrapolation capability of GNN enhanced with the estimates of entropy and spectral changes in the sampled data and additional temporal contexts to reconstruct the likely temporal trajectory of machine behavioural changes. The proposed real-time model belongs to the Deep Learning category of machine learning and interfaces with the sensors or through 'Process Data Historian', SCADA etc., to perform forecasting and classification tasks. The research demonstrates that a hybrid GNN based approach enhanced with entropy calculation and spectral information can effectively detect and predict a machine's behavioural changes. The model interfaces with a plant's 'process control system' in real-time to perform forecasting and classification tasks to aid the asset management engineers to operate their machines more efficiently and reduce unplanned downtimes. A series of trials are planned for this model in the future in other manufacturing industries.

3:00 PM  
Multi-property Graph Networks for Novel Materials Discovery: Alexander New1; Nam Le1; Michael Pekala1; Kyle McElroy1; Janna Domenico1; Christine Piatko1; Elizabeth Pogue1; Tyrel McQueen2; Christopher Stiles1; 1Johns Hopkins University Applied Physics Laboratory; 2Johns Hopkins University
    Machine learning (ML) approaches have the potential to accelerate material property prediction. Conventional approaches to materials discovery are expensive, but supervised ML models can rapidly screen large materials databases and identify candidates to test. When searching for a candidate, multiple properties will determine its relevance. Ideally, a single machine learning model could predict all desired properties for a given material. This is analogous to the ML concept of multi-task learning – by leveraging similarity between different prediction tasks, this single model will be better at prediction than a suite of property-specific models. We compare some state-of-the-art multi-task ML models to single-property models for predicting elastic material properties. These models often do not surpass single-property models, in contrast to existing findings in supervised learning. We present loss-surface curvature metrics that have the potential to explain this performance disparity, further suggesting research directions for multi-task ML to perform well for property prediction.

3:20 PM Break

3:40 PM  
Rapid Metallic Alloy Development Leveraging Machine Learning: Nhon Vo1; Ha Bui2; 1NanoAL LLC; 2Amatrium Inc.
    Alloy development and optimization have advanced thanks to multi-scale computer simulations and high-throughput experiments. The recent rise of machine learning has given the materials designers another powerful tool for developing new materials, but much has been undiscovered or utilized beyond research. This work demonstrates how a particular machine learning program is embedded in the traditional alloy development workflow to shorten the timeline and save costs for industry users significantly. Multiple material properties, typically time-consuming to measure or calculate, can be predicted with high accuracy. Unlike other traditional methods, the new alloy development workflow nearly acts as a ‘black box’ process, requiring minimal materials science expertise from users. We will demonstrate alloy development programs in the metal industry that successfully leverage this new toolset. Lastly, we will outline current limitations, including data consolidation, to be addressed in the future to capture the power of machine learning for alloy development fully.

4:00 PM  
Machine Learning Guided Prediction of Rupture Time of 347H Stainless Steel: Mohammad Fuad Nur Taufique1; Madison Wenzlick2; Arun Sathanur1; William Frazier1; Ram Devanathan1; Keerti Kappagantula1; Shoieb Ahmed Chowdhury1; 1Pacific Northwest National Laboratory; 2National Energy Technology Laboratory
    The creep resistance of 347H stainless steel depends on alloying elements, microstructures, and loading conditions. Therefore, it is important to predict the creep resistance i.e. rupture time of a 347H stainless steel before implementing an engineering design. Usually, semi-empirical time-temperature relations such as the Larson–Miller parameter (LMP) and the Manson–Haferd parameter (MHP) are used to predict the rupture time. However, these methods depend on empirical constants and often provide an unsatisfactory estimation for rupture lifetime for short to medium-term creep testing. In this study, we propose machine learning (ML) based calculations to accurately predict the rupture time of 347H stainless steels and highlight the importance of a high-quality dataset on the model performance. We achieved a coefficient of determination (R2) value of 0.88 from the gradient boosting regressor, which indicates a reliable model to predict the rupture time even though the data size is limited.

4:20 PM Concluding Comments