First World Congress on Artificial Intelligence in Materials and Manufacturing (AIM 2022): Machine Learning/Deep Learning in Materials and Manufacturing VI
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
Wednesday 9:00 AM
April 6, 2022
Room: 3' Rivers
Location: Omni William Penn Hotel
Session Chair: Katelyn Jones, Carnegie Mellon University
9:00 AM Invited
Combining Multimodal Data of Fatigue Fracture Surfaces for Analysis in a CNN: Katelyn Jones1; Anthony Rollett1; Elizabeth Holm1; 1Carnegie Mellon University
Computer vision and machine learning techniques can be applied to microstructural images and data to make predictions and connections between the microstructure and material responses. These methods can be applied to large amounts of data from a diverse set of images, when there is enough training data. Convolution neural networks (CNNs) are a crucial tool in analyzing fatigue fracture surface images and connecting their microstructure and fatigue characteristics to loading values, crack length, and crack growth rate. This project collects data from Ti-6Al-4V fracture surfaces in the form of BSE and SEM images, and height data from Scanning White Light Interferometry, and combines them to train a CNN and identify high stress points, crack initiation sites, and predict values such as stress intensity factor and crack growth rate. The images used to develop this model, creation of the CNNs, identified fatigue properties, and fracture characteristics will be presented.
9:30 AM
Machine Learning Based Prediction of Corrosion Behavior in Additively Manufactured Titanium Alloy: Nithin Konda1; Mythreyi OV1; Jayaganthan R1; 1Indian Institute of Technology Madras
In the present work, machine learning-based corrosion behavior modeling has been carried out in Laser powder bed fused titanium alloy. Using optimized process parameters, titanium alloy samples were produced using the LPBF technique and subjected to heat treatment and shot peening. Potentiodynamic polarization and electrochemical impedance tests were conducted at room temperature in an alkaline environment. The collected experimental data was then utilized for predictive model development using 8 different machine learning algorithms. The model performances were compared using the standard metrics such as RMSE, MSE, and R2. The feature importance analysis was carried out to assess the magnitude of effect each of the post-processing parameters had on corrosion performance. This work enhances the understanding between post-processing parameters and the degradation behavior oftitanium alloy using a machine learning approach.
9:50 AM
Machine Learning Based Prediction of Fatigue Crack Growth Rate in Additively Manufactured Ti6Al4V Alloy
: Jayaganthan Rengaswamy1; Nithin Konda1; 1IIT Madras
In the present work, machine learning based fatigue crack growth rate (FCGR) modelling has been carried out for Ti-6Al4V alloy fabricated through additive manufacturing (AM) technology. The FCGR data of Ti6Al4V alloy fabricated through laser powder bed fusion (LPBF) and other AM routes were utilized for fatigue life prediction using machine learning (ML) algorithms such as support vector regression, random forest, and deep neural networks. The model performances estimated using these algorithms were compared using the standard metrics such RMSE, MSE & R2. The feature importance analysis was carried out to assess the magnitude of effect of each of the post processing techniques and built orientation on FCGR behaviour. The prediction of fatigue crack growth rate in Ti-6Al-4V alloy made by ML is compared with Paris law. The microstructural characteristics of as built and post-processed LPBFed Ti-6Al-4V are used to substantiate its fatigue life predicted through ML techniques.
10:10 AM Break
10:40 AM
Latent Variable Rietveld Model for High Throughput Quantitative X-ray Diffraction Analysis: Brian DeCost1; Austin McDannald1; Howie Joress1; Jason Hattrick-Simpers2; 1National Institute of Standards and Technology; 2University of Toronto
Recent advances in applied machine learning have enabled automated qualitative analysis of high throughput x-ray diffraction data. However, expert inspection and interpretation of model outputs remains an integral component current unsupervised diffraction models, and even supervised methods are not yet capable of quantitative diffraction analysis. We seek to bridge this gap in high throughput quantitative analysis by combining a flexible probabilistic machine learning approach with established physics-based Bayesian Rietveld modeling approaches. The joint analysis of related diffraction data enabled by this approach should allow for higher sensitivity to minority or trace phases. The principal weakness of this approach is that the prior distribution of constituent phases must be explicitly specified; discovery of novel, unanticipated phases is not feasible in this framework. However, the physical underpinnings of the model enable quantitative inference of phase fractions and coarse microstructural information on high throughput x-ray diffraction data.
11:00 AM
Neural Message Passing for Prediction of Abnormal Grain Growth in Monte Carlo Simulations of Materials Processing: Ryan Cohn1; Elizabeth Holm1; 1Carnegie Mellon University
Neural message passing allows deep learning to operate on irregular graph structures. This technique shows promise for modeling materials represented as graphs of connected grains. We apply neural message passing to predict the occurrence of abnormal grain growth (AGG) in materials processing simulations. AGG occurs when the growth rate of a small subset of grains far exceeds that of typical matrix grains during processing, significantly altering the properties and performance of various material systems. Despite this, AGG is not well understood and is therefore difficult to control during processing. Thus, there is significant interest in applying deep learning to predict the occurrence of AGG during processing. After generating a dataset of Monte Carlo simulations of grain growth, we train deep learning models to predict the occurrence of AGG. Preliminary results from this study indicate that a neural message passing model outperforms a comparable computer vision approach for this task.
11:20 AM
Automated and Intelligent Analysis of Extended X-Ray Absorption Fine Structure (EXAFS) and X-Ray Photoelectron Spectroscopy (XPS): Miu Lun Lau1; Jeff Terry2; Min Long1; 1Boise State University; 2Illinois Institute of Technology
We extend our Genetic Algorithms (GA) based software NEO<\I> to both EXAFS and XPS materials characterization data analysis for spectra related problems. We employ modular design of various crossover and mutation options, allowing the code to be extensible to different material characterization methods. In the case of EXAFS and XPS, the basic framework remains identical except for the unit of object function, where each fitting model is implemented differently. Our software has the capabilities to explore different parameters and structures of samples due to irradiation or annealing and measure their effects. Our results demonstrate optimal fitness scores with minimal human intervention for both crystalline and amorphous structures. We also apply our software to in-situ data of SnS2 batteries and observe the expected cycling behavior of bond distance in both Sn-Li and Sn-Sn scattering paths.