First World Congress on Artificial Intelligence in Materials and Manufacturing (AIM 2022): Machine Learning/Deep Learning in Materials and Manufacturing IV
Program Organizers: Taylor Sparks, University of Utah; Michael Dawson-Haggerty, Kerfed, Inc.; Elizabeth Holm, University of Michigan; Jin Kocsis, Purdue University; Adam Kopper, Mercury Marine; Benji Maruyama; James Warren, National Institute of Standards and Technology
Tuesday 1:30 PM
April 5, 2022
Room: Riverboat
Location: Omni William Penn Hotel
1:30 PM Invited
Now On-Demand Only - Atomistic Line Graph Neural Network for Improved Materials Property Predictions: Kamal Choudhary1; Brian DeCost1; 1National Institute of Standards and Technology
Graph neural networks (GNN) have been shown to provide much improved performance for representing and modeling atomistic materials compared with descriptor-based machine-learning models. While most existing GNN models for atomistic predictions are based on atomic distance information, they do not explicitly incorporate bond angles, which are critical for distinguishing many atomic structures. Furthermore, many material properties are known to be sensitive to slight changes in bond angles. We develop Atomistic Line Graph Neural Network (ALIGNN) using a GNN architecture that performs message passing on both the interatomic bond graph and its line graph corresponding to bond angles. We train 55 models for predicting several solid-state and molecular properties available in the JARVIS-DFT, Materials project and QM9 databases. ALIGNN can outperform some of the previously known GNN models by up to 43.8 % in accuracy with model training speed better or comparable to them.
2:00 PM
Hybrid Approach Combining Machine Learning and Finite Element Simulation for Process and Material Optimization: Pierre-Yves Lavertu; Emilie Storms1; 1e-Xstream Engineering, part of Hexagon
In today’s industrial context, the pace of innovation is constantly accelerating while development cycles are shortening. In this framework, additive manufacturing and Machine Learning (ML) are combined in order to provide a cutting-edge solution.Dimensional stability and performance of 3D printed parts are highly dependent on the material selection and process parameters. Machine learning algorithms can be built to assess correlation between the process and material parameters to accelerate material selection and product development. The built ML algorithms can be used to determine optimal parameters which minimize warpage and maximize structural performance (stiffness, strength, durability, …). ML algorithms can also be used for material data enrichment minimizing testing cost while optimizing coverage of the design space.
2:20 PM
Efficient, Interpretable Atomistic Graph Neural Network Representation for Angle-Dependent Properties and its Application to Optical Spectroscopy Prediction: Tim Hsu1; Nathan Keilbart1; Stephen Weitzner1; James Chapman1; Tuan Anh Pham1; Roger Qiu1; Xiao Chen1; Brandon Wood1; 1Lawrence Livermore National Laboratory
Graph neural networks have gained popularity for learning properties of molecular and atomic structures thanks to their physically informed representation of atoms and bonds. However, standard GNNs lack explicit information about bond angles, which impact electronic structure and chemical hybridization. A recent formulation (ALIGNN) uses auxiliary line graphs to explicitly represent bond angles. In this work, we extend ALIGNN to also capture dihedral angles and chirality (ALIGNN-d). Such encoding is particularly important for amorphous/disordered systems and functional material design in phase spaces where angular information is critical. Using optical absorption spectroscopy of copper aqua complexes as a demonstration, we show that ALIGNN-d can accurately and efficiently predict key spectral features. Further, the expressive power of ALIGNN-d is shown to match the far less scalable maximally-connected graph, in which all pairwise bonds are explicitly encoded. Lastly, we show how the physical motivations underlying ALIGNN-d can be leveraged for model interpretability.
2:40 PM Break
3:10 PM
Multi-scale Structure-Property Relationships in Low Carbon Steels: Johan Westraadt1; Lindsay Westraadt1; 1Nelson Mandela University
Small-punch creep (SPC) testing is currently used to evaluate the creep-rupture properties of steels used in the petrochemical and power generating industries. This study explores microstructure-property relationships in service-exposed low carbon steels using ML. These models can be used to rank the microstructural features in terms of their importance on the SPC-test and prioritize/reduce SPC testing requirements. A dataset consisting of 120x3 steel microstructures and their associated SPC-rupture time was collected. Optical micrographs of the etched surfaces were quantified using various feature extraction methods including 1- and 2-point statistics, and convolutional neural networks. The extracted microstructural features were then reduced using PCA and used as inputs for training regression models using different ML techniques. A selection of 10 samples with the largest testing errors were then investigated using secondary electron imaging to incorporate the finer pearlite sub-structures. This multi-scale model had a lower training error for the outlier samples.
3:30 PM
Qualitative Assessment of Degradation and Ageing Behaviour of Epoxy-Al Nanocomposites Through Machine Learning Assisted With LIBS.: Sneha Jayaganthan1; Naresh Chillu2; Sarathi Ramanujam2; Jayaganthan Rengaswamy2; 1IIT Goa; 2IIT Madras
Epoxy resin loaded with conductive nanofillers using Al below percolation threshold have improved its charge trap properties, to use as insulation structure. It is essential to monitor multi stress ageing of polymer nanocomposites for predicting it’s failure. In the present work, LIBS investigations assisted by machine learning (ML) algorithms are used to predict its ageing behaviour. Various ageing conditions such as water ageing, corona and γ irradiation were performed on the epoxy polymer along with optimized 5 wt% Al filled epoxy nanocomposite. The LIBS spectral intensities were used for detecting the levels of decompositions during ageing. The ML algorithms such as support vector machine and neural networks are adopted to LIBS spectral data for classifying the degradation level. The comparative analysis of machine learning algorithms is made with respect to their classification accuracies using confusion matrices, which helps in assessing the level of degradation for condition monitoring of power systems.
3:50 PM
Research Acceleration via Machine Learning for Characterization of Growing Dendritic Crystals from In Situ X-Ray Videos of Alloy Solidification: Jonathan Mullen1; Mert Celikin1; Pádraig Cunningham1; David Browne1; 1University College Dublin
The observation and measurement of dendrites over time, through the use of in-situ X-Ray videos, can offer key insights into alloy solidification behaviour. However, depending on the constraints in place during data acquisition, the resulting videos can be difficult to assess due to imaging related issues, such as high noise or low contrast. Conventional image analysis and enhancement techniques alone can lead to a considerable reduction in measurement accuracy or to the need to manually assess video frames. Our approach demonstrates that, by using machine learning as a constituent part of an integrated analysis system, it is possible to obtain useful results from videos which are otherwise difficult or time-consuming to assess. This is shown through the automated assessment of individual dendrites within a thin Al-20wt%Cu alloy sample for two X-Ray videos, acquired as part of the MASER-13 microgravity sounding rocket project, and a comparison against solely conventionally obtained results.