Forthcoming, 3:00 ~ 4:00 pm (ET)
November 21, 2024
- Christina Krawec (Nonproliferation and National Security Department, Brookhaven National Laboratory), “Open Source Analysis and Collection for International Atomic Energy Agency Safeguards”
Abstract: The International Atomic Energy Agency (IAEA), the United Nations partner organization dedicated to safeguarding nuclear technology, advocates for the peaceful use of nuclear energy while working to prevent its utilization for military purposes, including nuclear weapons. Christina Krawec, former IAEA employee and analyst, will provide an overview of the role open source information plays in the analysis of state nuclear programs — a pivotal element of the IAEA’s work. She will discuss open source tools and techniques used at the IAEA, including satellite imagery and emerging technology.
October 24, 2024
- Dr. Jen Willis (Data Scientist, AI and Decision Analytics Division, Pacific Northwest National Laboratory), “Arms Control Treaty Monitoring Systems Efficacy Evaluation”
Abstract: Nuclear arms control treaties like New START have included inspections where treaty partners are permitted to employ agreed-upon monitoring processes and technologies to agreed-upon items, to confirm treaty goals and objectives are being met the other treaty partner. Historically, such monitoring technologies, processes, and metrics are selected during treaty negotiations with little or no quantitative information about the efficacy that various selections could have on subsequent treaty monitoring activities, although such assessments may have been implicit in treaty negotiators’ decisions. In this research, we aim to explicitly calculate efficacy of potential monitoring technologies, processes, and metrics (TPM) using mathematical models.
We have developed a Bayesian model for this purpose that assumes different data streams will result from individual TPMs, each of which provide different fidelity and relevance to conclusions about treaty compliance. The Bayesian model accounts for such differences in its parameterization which includes performance characteristics such as true and false confirmation and rejection rates of each TPM. The performance characteristics can be derived from experimental results and/or subject matter expertise under different conditions (e.g., distance from an object, measurement area characteristics). The model also assumes some level of trust an inspecting partner could have in the host partner prior to the inspection. By combining this prior trust with the likelihood functions corresponding to each TPM, the model estimates a posterior likelihood or “efficacy” of that combination. Efficacy from several different TPM combinations can be compared to determine which is predicted to provide the best relative efficacy. The model can also be used to run sensitivity studies to provide additional information about which TPM are expected to have the biggest (or smallest) impacts on efficacy.
This model can eventually be integrated into a tool to support treaty negotiator decision-making in real time. It may also be used in the future to optimize efficacy simultaneously with other variables such as disruptiveness, invasiveness, and time associated with various TPM selections.
September 26, 2024
- Dr. Tom Danielson (Applied Sensing & Data Science Team Lead, Savannah River National Laboratory), “Understanding Event Trajectories Across Massive Temporal Datasets: Use Cases for Nuclear Non-Proliferation Research and Development”
Abstract: The massive open source is filled with implicit and explicit indicators of events of interest across time. However, for broad topical domains at a global scale, manual exploration and extraction requires significant effort and is likely to overlook key information. To overcome this challenge, researchers from the Savannah River National Laboratory have developed a pipeline that enhances a user’s ability to identify and extract domain-specific events across time. Built on a foundation of natural language processing, the pipeline identifies contextual shifts in keywords and phrases, which act as indicators of events of interest. In this approach, words, documents, local topics, and global topics are analyzed within the same embedding space, providing a unified approach to understanding event trajectories across time. Furthermore, the team has developed unique visual analytics techniques that facilitate exploration of the entire domain in one pass. Here, the technical approach of the pipeline will be presented and several case studies will be explored within the domain of “worldwide state sponsored civil nuclear energy”.
August 22, 2024
- Dr. Jordan Stomps (Nonproliferation Data Scientist, Oak Ridge National Laboratory), “Using gamma spectroscopy data augmentations and contrastive machine learning to characterize shielded radiological material transfers”
Abstract: The timely detection of shielded radiological material transfers is an important monitoring objective in nuclear nonproliferation. Labeling sufficient volumes of radiation data for successful supervised machine learning can be too costly when manual analysis is employed. Therefore, this work developed a machine learning model built on contrastive machine learning to utilize both labeled and unlabeled data and therefore alleviate the cost of labeling, using gamma spectra measured at the Multi-Informatics for Nuclear Operating Scenarios testbed at Oak Ridge National Laboratory. An encoder is contrastively trained using label-invariant augmentations to produce high-dimensional representations of spectra. A supervised classifier then uses these encoded representations to assign a label estimating whether a given transfer spectrum was of tracked nuclear material or not. Even a simple linear model built on these representations and trained on limited labeled data can achieve a balanced accuracy score of 80.30%. Principal Component Analysis demonstrates that representations provide a richer feature space for detecting nuclear material transfers by embedding distributional information from unlabeled data. Integrated Gradients connect a classifier’s decision boundary to spectral features, suggesting the framework learns relevant patterns in spectra that can be used for detecting transfers. When labeled data are scarce, this work suggests that training a supervised classifier should be prioritized over semi-supervised (compared to self-supervised) contrastive learning an encoder to maximize detection accuracy. Hyperparameter optimization was conducted, finding a locally optimum cross-validated balanced accuracy score. Overall, a methodology has been established for accurately classifying material transfers without the prohibitive cost of labeling.
July 25, 2024
- Dr. Danielle Mannion (Senior Scientist, Savannah River National Laboratory), “Mass Spectrometry Research and Development for Nuclear Nonproliferation and Safeguards”
Abstract: The nuclear nonproliferation and safeguards communities continues to seek rapid, accurate, and precise characterization capabilities for the measurement of uranium isotopic compositions. Current in-facility methods utilize various destructive and nondestructive techniques but may require long duration measurement periods, handling of reactive materials, and/or struggle to adequately discriminate near-natural isotope ratios. Mass spectrometry (MS) is considered the gold standard for analysis of relatively long-lived actinides such as uranium and plutonium; however, conventional MS analysis often requires time consuming sample preparation and complex analytical methodologies that are difficult to perform in-field or in-facility. In recent years, researchers at Savannah River National Laboratory (SRNL) have worked to develop two mass spectrometry technologies to overcome these gaps: matrix assisted ionization mass spectrometry (MAI-MS) and virtual-slit cycloidal mass spectrometry (VS-CMS). MAI-MS research has focused on the development of uranium isotopic measurement methods using a non-conventional ion source for rapid analysis in-facility. VS-CMS is a transformational mass spectrometry method that could provide high precision single particle isotopic measurement in a compact instrument. SRNL’s work with VS-CMS is in collaboration with both University and National Lab partners. Both research endeavors will be discussed as well as future plans.
May 23, 2024
- Dr. Ken Dayman (Nonproliferation Data Scientist, Oak Ridge National Laboratory), “Data Science for Nonproliferation: Concepts, Requirements, and Constraints”
Abstract: Modern data science has the potential to enhance and enable capabilities across the nuclear nonproliferation mission landscape. However, nonproliferation applications present a unique combination of requirements and constraints on successful data analysis and interpretation development. These challenges largely stem from the high-consequence nature of nonproliferation decisions and the relative merits and limitations of different dataset generation methods. This talk will outline these concepts and illustrate a selection of their ramifications on research developing data-driven methods to analyze environmental samples and characterize reactor operations
April 25, 2024
- Dr. Vincent DiNova (Senior Nuclear Engineer, Savannah River National Laboratory), “Re-Imagining Qualification for Additive Manufacturing”
Abstract: Additive manufacturing offers exciting potential for the future of materials science and component design; however, qualification remains a major hurdle to widespread adaptation and usage. To fully realize the potential additive manufacturing poses, new analytical tools must be developed. The aim of our research at Savannah River National Laboratory is to develop software and technical solutions to better understand AM materials and components and ultimately develop a pathway to qualify AM parts for insertion. From software development to machine learning solutions: past/present/future NDE and analytical techniques developed at SRNL will be presented.
January 25, 2024
- Dr. Evangelina Brayfindley (Senior Data Scientist, Pacific Northwest National Laboratory), “Data Science for Safeguards”
Abstract: This will be an introduction to the IAEA, Safeguards, and the challenges of building data science approaches to safeguards questions. This will include an overview of several data science projects–including NLP and synthetic data generation projects–happening both at the US national labs as well as at the IAEA.
November 30, 2023
- Dr. Robert Lascola (Senior Fellow Scientist and Group Lead, Online Monitoring, Savannah River National Laboratory), “Raman Spectroscopy – A Valuable Tool for Monitoring Nuclear Materials Processing”
Abstract: Raman spectroscopy is an optical analysis technique that is complementary to infrared (IR) absorbance spectroscopy for the analysis of molecular species. Although the Raman scattering signal is far weaker per molecule than IR absorbance, the technique has certain properties that make it well suited for applications related to nuclear materials processing. For example, Raman spectroscopy is usually practiced using visible light, which makes it suitable for use with optical fibers. In this way, instruments and personnel can be kept out of radiological environments. Of the optical-based techniques, Raman is uniquely sensitive to molecules like N2, O2, and the various isotopologues of H2. And, in the last 20-30 years Raman has been the subject of intense commercial development, such that Raman spectrometers are being used for process monitoring in many industries. In this talk, I will introduce the method and provide three examples of how it is being used or developed to support nuclear materials processing at SRNL. These examples include: monitoring nuclear fuel dissolution, isotopic hydrogen analysis for tritium processing, and tracking of key constituents in actinide separations.
October 26, 2023
- Zach Condon (Prof. Richard Vasques’s group, The Ohio State University), “Unfolding Neutron Energy Spectra using Neural Networks”
Abstract: Acquiring accurate neutron energy spectrum information is of vital importance to national security as well as personal safety. Unfolding neutron energy spectra from detector responses is a heavily researched area due to the importance of neutron energy for determining radiation dose received. A novel detection system, the passive neutron spectrometer (PNS), is being investigated for use in energy spectrum unfolding techniques. The benefit of this detector is the passive detection of neutrons through the use of 55 thermoluminescent dosimeters or gold foils contained within a single polyethylene sphere. Multiple experimental PNS detector responses were unfolded using the well-established MAXED and GRAVEL algorithm as well as through neural network technique. The neural networks used for this research will be trained and optimized using training data from the IAEA. To increase the accuracy of the unfolded network, techniques for developing artificial training data will also be explored.