Western Washington
Data-driven Discovery Seminar Series (W2D2S2)

Machine Learning in Materials Science: Progress and Challenges. Dr. Ram Devanathan | PNNL

Bio: Ram Devanathan is the Director of the Energy Processes & Materials Division in the Energy & Environment Directorate at Pacific Northwest National Laboratory (PNNL) in Richland, Washington. He is passionate about advancing the clean energy revolution through decarbonization, electrification, and energy efficient advanced manufacturing. Devanathan’s technical interests include the design of materials for extreme environments, multiscale modeling, and machine learning for materials discovery. He serves as an active volunteer in the American Chemical Society, Materials Research Society, and The Minerals, Metals & Materials Society. He is a recipient of the US Department of Energy’s Outstanding Mentor Award for his efforts to involve high school, undergraduate, and graduate students in computational materials science, and the American Ceramic Society’s Richard M. Fulrath Award. Devanathan is a Fellow of the American Ceramic Society, the American Chemical Society, and the Oppenheimer Science and Energy Leadership Program.

Abstract: : This talk will start with an overview of the application of machine learning to solve materials science problems. Existing materials data infrastructure, communities of practices, and results from the literature will be introduced. Next, the challenges of data quality and small data and the need to incorporate physics into the models will be presented along with recent progress at PNNL in these areas. The talk will conclude by presenting a strategy to advance energy and environment research at PNNL through data management and machine learning with emphasis on energy storage as a use case.

Microscopy is All You Need. Prof. Sergei Kalinin | University of Tennessee, Knoxville

Abstract: : Machine learning and artificial intelligence (ML/AI) are rapidly becoming an indispensable part of physics research, with domain applications ranging from theory and materials prediction to high- throughput data analysis. Over the last several years, increasing attention is attracted to the use of AI interacting with physical system as a part of active learning – including materials discovery and optimization, chemical synthesis, and physical measurements. For these active learning problems, microscopy arguably represents an ideal model application combining aspects of materials discovery via observation and spectroscopy, physical learning with relatively shallow priors and small number of exogenous variables, and synthesis via controlled interventions. In this presentation, I will discuss recent progress in automated experiment in scanning probe and electron microscopy, ranging from feature to physics discovery via active learning. The applications of classical deep learning methods in streaming image analysis are strongly affected by the out of distribution drift effects, and the approaches to minimize though are discussed. I will further illustrate transition from post-experiment data analysis to active learning process, including learning structure-property relationships and materials discovery in composition spread libraries. Here, the strategies based on simple Gaussian Processes often tend to produce sub-optimal results due to the lack of prior knowledge and very simplified (via learned kernel function) representation of spatial complexity of the system. Comparatively, deep kernel learning (DKL) and structured Gaussian Processes methods allow to realize both the exploration of complex systems towards the discovery of structure-property relationship and enable automated experiment targeting physics (rather than simple spatial feature) discovery. The latter is illustrated via experimental discovery of ferroelectric domain dynamics in piezoresponse force microscopy. For probing physical mechanisms of tip-induced modifications, I will demonstrate the combination of the structured Gaussian process and reinforcement learning, the approach we refer to as hypothesis learning. Here, this approach is used to learn the domain growth laws on a fully autonomous microscope. The future potential of Bayesian active learning for autonomous microscopes is discussed. These concepts and methods can be extended from microscopy to other areas of automated experiment.

Bio: Prof. Sergei Kalinin is a professor at the University of Tennessee, Knoxville (currently on sabbatical at Amazon), following 20 years at Oak Ridge National Laboratory. He received his MS degree from Moscow State University in 1998 and Ph.D. from the University of Pennsylvania (with Dawn Bonnell) in 2002. His research focuses on the applications of big data and artificial intelligence methods in materials discovery and atomically resolved imaging by scanning transmission electron microscopy and scanning probes for applications including physics discovery and atomic fabrication, as well as mesoscopic studies of electrochemical, ferroelectric, and transport phenomena via scanning probe microscopy. Prof. Kalinin has co-authored >650 publications, with a total citation of ~45,000 and an h-index of >100. He is a fellow of MRS, APS, IoP, IEEE, Foresight Institute, and AVS; a recipient of the Blavatnik Award for Physical Sciences (2018), RMS medal for Scanning Probe Microscopy (2015), Presidential Early Career Award for Scientists and Engineers (PECASE) (2009); Burton medal of Microscopy Society of America (2010); 4 R&D100 Awards (2008, 2010, 2016, and 2018); and a number of other distinctions.

High-throughput Materials Synthesis in Atomic Layer Processing. Professor David Bergsman, University of Washington

Abstract: Recent years have seen a surge of interest in the development of scalable tools for nanomaterials synthesis. Many emerging technologies, like solar cells, batteries, catalysts, and membranes, rely on atomically-precise materials design to operate effectively. Tools for creating these materials with increased scalability and decreased costs are thus required to enable the widespread adoption of these technologies. One suite of tools, known collectively as atomic layer processing (ALP), is particularly interesting for nanomaterials synthesis, due to its ability to create ultrathin films with sub-nanometer thickness and compositional control. ALP has also been used in the semiconductor industry over several decades, making it easy to deploy in other manufacturing processes. Commercial solar panels and battery electrodes have already started to incorporate ALP-deposited films as passivation layers. However, as demand for nanotechnology increases, there is a continued need to expand the library of materials that can be made with these tools and to accelerate the pace with which these materials are deployed. This presentation will highlight how the Bergsman Research Group at the University of Washington is expanding the applicability of ALP by creating new processes with higher-throughput screening tools. We will describe some of the use cases of ALP in semiconductor processing, catalysis, and membrane separations, along with the challenges associated with making new materials processes with these tools. Then, we will discuss existing methods for more rapid deployment of new processes. Finally, we will discuss our work to speed up materials development using a high-throughput reactor, along with our efforts to implement a connected data pipeline for easier analysis of process parameters.

Bio: David Bergsman (he/him) is an Assistant Professor in the Department of Chemical Engineering at the University of Washington. He received his B.S. in Chemical Engineering from the University of Washington in 2012 and his PhD in Chemical Engineering from Stanford University in 2018 under the mentorship of Prof. Stacey Bent. He later completed postdoctoral work at the Massachusetts Institute of Technology with Prof. Jeff Grossman. Now, at UW, his research is focused on using ultrathin films and coatings to tackle challenges in energy, water, sustainability, and semiconductor processing.

Welcoming our AI Overlords: Operationalizing Machine Learning for Materials Discovery and Design. Dr. Steven R. Spurgeon | Pacific Northwest National Laboratory and University of Washington

Abstract: Artificial intelligence (AI) promises to reshape scientific inquiry and enable breakthrough discoveries in areas such as energy storage, quantum computing, and biomedicine. While it is now possible to produce nanomaterials in almost limitless configurations, engineering of desirable functionality depends on precise control of atomistic structure and defects. Complex synthesis pathways can lead to significant deviations from idealized structures, which occur at length scales that are challenging to probe experimentally and theoretically. Mastery of materials is therefore predicated on the ability to acquire and interpret complex, heterogeneous, and fast-evolving data streams, a task uniquely suited to emerging AI and machine learning methods. Here I will discuss our efforts to develop a new framework for materials discovery in the electron microscope, leveraging low-level system automation, domain-grounded data pre-processing, and emerging sparse data analytics to extract truly statistical insights. I will discuss the current and future potential of this platform to both unlock experimentation at scale and derive richer, more meaningful physical models for important material systems.

Bio: Dr. Steven R. Spurgeon is a research scientist in the Energy and Environment Directorate at Pacific Northwest National Laboratory, with an affiliate appointment as an Associate Professor of Physics at the University of Washington. He serves as thrust lead for PNNL’s Chemical Dynamics Initiative and an editor of the international journal Microscopy and Microanalysis. His work focuses on developing artificial intelligence and machine learning approaches to accelerate the synthesis, characterization, and modeling of nanomaterials for next-generation electronics, quantum computing, and energy storage. He has published over 65 journal articles and book chapters and has received awards from the U.S. Department of Energy, the National Science Foundation, the Materials Research Society, the Microscopy Society of America, and the U.S. Department of Defense. Prior to joining PNNL, he received his Ph.D. in Materials Science from Drexel University and his B.S. in Materials Science from Carnegie Mellon University.

Shining light on perovskites. Prof. Rob Berger | Western Washington University

Abstract: Compounds crystallizing in the ABX3 perovskite structure are studied for many applications, including solar energy conversion. This class of materials includes both lead-halide perovskites that absorb light in photovoltaic solar cells (e.g., CH3NH3PbI3) and transition metal-oxide perovskite photocatalysts (e.g., SrTiO3). One reason for the technological versatility of perovskites is that their composition and structure, and consequently their electronic properties, are highly tunable. These compounds can be engineered through elemental substitution, strain, layering, and defects, allowing for the optimization of their properties. I will summarize our group’s work over the past several years, in which we use density functional theory (DFT) calculations to explore how changes in composition and atomic structure affect the electronic structure and properties of perovskites for solar energy conversion. This work has led us to both fundamental understanding and concrete predictions of new materials.

Bio: Rob Berger is an associate professor in the Western Washington University Department of Chemistry. Since joining WWU in 2013, Dr. Berger’s research group has used computation to understand relationships among the atomic and electronic structure of solids for energy applications. In the classroom, Dr. Berger teaches primarily physical chemistry and general chemistry.

Machine-learning-guided descriptor selection for predicting corrosion resistance in multi-principal element alloys. Dr. Ankit Roy | Pacific Northwest National Laboratory

Abstract: More than $270 billion is spent on combatting corrosion annually in the USA alone. As such, we present a machine-learning (ML) approach to down select corrosion-resistant alloys. Our focus is on a non-traditional class of alloys called multi-principal element alloys (MPEAs). Given the vast search space due to the variety of compositions and descriptors to be considered, and based upon existing corrosion data for MPEAs, we demonstrate descriptor optimization to predict the corrosion resistance of any given MPEA. Our ML model with descriptor optimization predicts the corrosion resistance of a given MPEA in the presence of an aqueous environment by down selecting two environmental descriptors (pH of the medium and halide concentration), one chemical composition descriptor (atomic % of element with minimum reduction potential), and two atomic descriptors (difference in lattice constant (Δa) and average reduction potential). Our findings show that, while it is possible to down select corrosion-resistant MPEAs by using ML from a large search space, a larger dataset and higher quality data are needed to accurately predict the corrosion rate of MPEAs. This study shows both the promise and the perils of ML when applied to a complex chemical phenomenon like the corrosion of alloys.

Bio: Dr. Ankit Roy graduated from his PhD program in Mechanical Engineering from Lehigh University (PA) in August, 2021 and has been working for PNNL since Sept 2021. He works on Li-Al-O ceramics to model the radiation damage in them using molecular dynamics (MD). He is also working on modeling radiation damage in some Ti alloys using molecular dynamics. In his PhD program, he used DFT, MD and machine learning to explore the properties of high entropy alloys (HEAs).

Accelerating Atomistic Simulations of Solid-Phase Processing with Neural Network Potentials. Dr. Jenna Pope | Pacific Northwest National Laboratory

Abstract: Neural network potentials (NNPs) can greatly accelerate atomistic simulations, allowing one to sample a broader range of structural outcomes than possible with ab initio methods. This talk presents an active sampling algorithm to train an NNP that shows DFT-level accuracy in dynamic shear simulations of a model Cu-Ni multilayer system. We then use the NNP in conjunction with a perturbation scheme to stochastically sample structural and energetic changes caused by shear-induced deformation, demonstrating the range of possible intermixing and vacancy migration pathways that can be obtained as a result of the speedups provided by the NNP.

Bio: Jenna Pope is a Data Scientist in the National Security Directorate at Pacific Northwest National Laboratory. Her research focuses on the application of data science and deep learning to chemistry and materials science. Her projects are highly interdisciplinary and involve close collaboration with both experimentalists and modelers/theoreticians. She is a member of the American Chemical Society and serves on review panels for NSF. She received a BS in chemistry from the University of West Florida and a PhD in computational chemistry from the University of Georgia. She publishes under the name Jenna A. Bilbrey.

Differential Property Prediction: A Machine Learning Approach to Experimental Design in Advanced Manufacturing. Loc Truong | Pacific Northwest National Laboratory

Abstract: Advanced manufacturing techniques have enabled the production of materials with state-of-the-art properties. In many cases however, the development of physics-based models of these techniques lags behind their development in the lab. This means that material and process development proceeds largely via trial and error. This is sub-optimal since experiments are cost-, time-, and labor-intensive. In this work we propose a machine learning framework, differential property classification (DPC), which enables an experimenter to leverage machine learning's unparalleled pattern matching capability to pursue data-driven experimental design. DPC takes two possible experiment parameter sets and outputs a prediction of which will produce a material with a more desirable property specified by the operator. We demonstrate the success of DPC on AA7075 tube manufacturing process and mechanical property data using shear assisted processing and extrusion (ShAPE), an emerging solid phase processing technology. We show that by focusing on the experimenter's need to choose between multiple candidate experimental parameters, we can reframe the challenging regression task of predicting material properties from processing parameters, into a classification task on which machine learning models can achieve good performance.

Bio: Loc Truong joined PNNL in 2021 as a data scientist in the Data Sciences and Analytics Group. Loc's research background is primarily focused on computer vision, and adversarial machine learning. He’s constantly looking for new ways to apply data science to solve real world problems.

Making Machine Learning and Explainable Artificial Intelligence Work for Chemistry. Dr. Johannes Hachmann | University at Buffalo

Abstract: The use of modern machine learning, informatics, and data mining approaches is a relatively new development in the chemical and materials domain. These techniques have been exceedingly successful in other application fields, and since there is no fundamental reason why they should not have a similarly transformative impact on chemical and materials research, there is now a concerted effort by the community to introduce data science in this new context. They hold tremendous promise for the practical realization of accelerated discovery and inverse design. However, adapting techniques from other application domains for the study of chemical and materials systems requires a substantial rethinking and redevelopment of the existing methods.

Bio: Prof. Johannes Hachmann is Associate Professor in the Department of Chemical and Biological Engineering at the University at Buffalo. The Hachmann group’s research fuses (first-principles) molecular and materials modeling with virtual high-throughput screening and modern big data science (i.e., the use of database technology, machine learning, and informatics) to advance a data-driven discovery and rational design paradigm in the chemical and materials disciplines. The primary application focus is on the development of novel molecular materials and catalysts, e.g., for renewable energy technology and advanced electronics. Prof. Hachmann pursued undergraduate studies at the University of Jena and University of Cambridge (UK) before earning MS and PhD degrees in Theoretical Chemistry at Cornell University. After a postdoctoral appointment spearheading the Clean Energy Project at Harvard University with the Aspuru-Guzik group at Harvard University, Prof. Hachmann joined the faculty of the University at Buffalo in 2014.

Well-calibrated and domain-informed deep learning methods for quantifying the change in microstructural features post-radiation on LiAlO2 pellets. Dr. Karl Pazdernik | Pacific Northwest National Laboratory

Abstract: LiAlO2 is an important material that is used as a tritium producer for the Tritium Modernization Program. To better understand the tritium release from the material during irradiation, a comprehensive microstructural analysis of unirradiated and irradiated LiAlO2 is required. Recently, deep learning has been employed as a fast approach to classifying various microstructural features in LiAlO2 pellets that are visualized by scanning electron microscopy (SEM), including grains, grain boundaries, voids, precipitates, and zirconia impurities. While these methods can produce high overall accuracy, it can also decrease significantly when the SEM image is collected under different conditions and while the overall accuracy may remain high, challenging regions such as the boundaries between microstructural features may be predicted with much higher error. To be more robust to image variability and properly quantify regions of lower confidence, we present a well-calibrated and domain-informed deep learning approach to semantic segmentation of SEM images which includes uncertainty quantification and allows for comparison across multiple SEM images. We highlight areas where prediction is less certain and summarize the effects of radiation on LiAlO2.

Bio: Dr. Karl Pazdernik is a Senior Data Scientist within the National Security Directorate at Pacific Northwest National Laboratory (PNNL), a team lead within the Foundational Data Science group at PNNL, and a Research Assistant Professor at North Carolina State University. He is the program lead for the Open-Source Data Analytics program and a principal investigator on projects that involve disease modeling and image segmentation for materials science. His research has focused on the uncertainty quantification and dynamic modeling of multi-modal data with a particular interest in text analytics, spatial statistics, pattern recognition, anomaly detection, Bayesian statistics, and computer vision applied to financial data, networks, combined open-source data, disease prediction, and nuclear materials.

Deep Learning Tools for Advanced Manufacturing Applications. Dr. Keerti Kappagantula | Pacific Northwest National Laboratory

Abstract: Several advanced manufacturing techniques, such as solid phase processing or additive manufacturing, are of great interest to industry given their ability to achieve components with never-seen-before properties, forms and/or energy efficiency. However, most of these approaches are also associated with emerging computational capabilities owing to their nascent stage of development. As such, often design of experiments or property predictions corresponding to a specific process regime/form factor/feedstock is Edisonian in nature. In such situations, machine learning and artificial intelligence tools provide a viable pathway for data driven analysis for predicting process conditions or optimal performance metrics of interest. Additionally, development of such surrogate models may also accelerate technology deployment while accelerating the R&D cycle. This walk will provide some recent examples of the different machine learning tools deployed in advanced manufacturing regimes. The advantages and limitations of such approaches and their adoption in manufacturing industry will also be discussed.

Bio: Dr. Kappagantula is a Senior Scientist and the Team Leader in the Advanced Materials & Manufacturing Group at PNNL. Her research focuses on developing high performance materials through the use of nanotechnology, advanced manufacturing processes, atomistic-to-mesoscale modeling and machine learning/artificial intelligence/data-driven methods. Prior to joining PNNL in 2019, she was an Assistant Professor (Tenure Track) of Mechanical Engineering at Ohio University. She also helped manage the Center for Advanced Materials Processing as the Assistant Director. Dr. Kappagantula manages projects that manufacture metal, polymer and ceramic composites that are used applications ranging from energy-efficient electric machines to sustainable carbon-negative building materials. She is funded by various DOE Offices such as Vehicles Technology Office, Advanced Manufacturing Office, and Fossil Energy in identifying materials of interest, developing manufacturing approaches, and transitioning the technologies into commercial use.

Modeling Solid Phase Processes with Physics Informed Machine Learning for Prediction and Control. Ethan King | Pacific Northwest National Laboratory

Abstract: Solid phase processes plunge a rotating tool into source material to heat and remodel the material to achieve desired properties and conformations. Temperature during processing is tightly linked to microstructure formation and thus final material properties but is difficult to model due to the complex thermo-mechanical feedbacks that determine generation of heat. This talk presents the use of both neural ordinary differential equations and partial differential equation models that capture known physics to leverage limited data for learning the relationship between process inputs and temperature. We show that these approaches can closely fit measurement data and be used to design process controls.

Bio: Ethan is a data-scientist at the Pacific Northwest National Laboratory. Since joining PNNL in 2020 his research has focused on integrating domain knowledge with machine learning and differentiable programming for dynamic modeling, control, and optimization across applications within biology, power systems, and materials science.

Mathematical frameworks for comprehensively exploring process parameter/property relationships in advanced manufacturing. Dr. Henry Kvinge | Pacific Northwest National Laboratory

Abstract: Advanced manufacturing holds the promise of producing materials with critically needed properties. To take advantage of this opportunity however, researchers need to be able to understand how the process parameters which define the manufacturing process relate to the end properties of the material. Since many different process parameters could theoretically yield the same property, the standard machine learning approach would be to train a model that, given a set of process parameters as input, predicts material properties as output (e.g., regression or classification). This however is only of modest help when the real goal is to understand the distribution of all process parameters that lead to the desired property, allowing the materials scientist to choose the one that satisfies a range of other criteria that they may have. This latter problem requires the training of a generative “inverse model”, that can learn multiple process parameter distributions. In this talk we discuss current progress on this problem using the topological construction of a fiber bundle.

Bio: Henry Kvinge is a mathematician/data scientist at Pacific Northwest National Lab. He works at the intersection of deep learning and mathematics with a special interest in understanding how machine learning can be adapted to handle the uncertainties and ambiguities of the real world. He has a special interest in model robustness and applications to materials science.

Characterizing crystallization kinetics and properties of thermoplastic composites. Prof. Mark Peyron | Western Washington University

Abstract: Thermoplastic composites are increasing important in aerospace, automotive, and other industrial applications. But the necessary detailed processing understanding lags that of thermosetting composites (such as carbon-epoxy systems). In thermosets, a rich understanding of cure kinetics is key to understanding and modeling the composite properties. For thermoplastic composites, mechanical, heat transfer, flow, and deformation properties are important to characterize, but the critical kinetic process for most materials of interest is crystallization. Crystallization kinetics are most-commonly studied using isothermal differential scanning calorimetry, but there are numerous challenges and problems with this approach. Alternatively, dynamic DSC studies offer opportunities for more complete and more consistent kinetic characterization. We have been investigating experimental and modeling approaches that reduce uncertainty, suppress user subjectivity, and allow estimation of prediction uncertainty using novel experimental and analysis techniques. A case study based on unidirectional carbon fiber tape prepreg with a high-performance poly(ether ketone ketone) resin (CF-PEKK) will be presented and contrasted with unreinforced PEKK resin. Uncertainty in data, modeling, and predictions will be addressed.

Bio: Prof. Peyron earned a BS Chemical Engineering from the University of Idaho and a PhD in Chemical Engineering from the University of Washington with a focus on polymer chemistry and physics. Peyron conducted Postdoctoral Research at the University of Cambridge with a focus on developing magnetic resonance hardware and relaxometry for porous media. He has industry experience prior to WWU in the areas of environmental engineering, fuel cell polymers, magnetic resonance in oil and gas, pulp and paper. Since 2014, Prof. Peyron has been at WWU in the Engineering and Design department, with a research focus on sustainable materials, as well as characterization and modeling of thermosetting and thermoplastic materials and composites.