This report summarizes the findings of a four months FY24 Advanced Science & Technology (AS&T) LDRD Quick Targeted Investigation (QTI) project focused on the exploration of supervised dimension reduction approaches based on autoencoders. Autoencoders have been extensively employed in literature for unsupervised learning tasks, however, their use for supervised regression tasks, which are common within scientific applications, has been limited. Motivated by linear dimension reduction strategies like Active Subspaces and Adaptive Basis, we explored the possibility of employing autoencoders to discover a non-linear manifold able to represent the original function in fewer dimensions. In this report, we discuss a neural network architecture and we perform a numerical campaign on several problems ranging from simple two-dimensional functions to a model problem for magnetohydrodynamics in five dimensions. In our preliminary results, we show that the proposed approach is found to be superior to linear dimension reduction strategies in representing the target function even with a single latent variable.
In several mission contexts, it is desirable to estimate the performance of large language models (LLMs) on tasks that we cannot run directly. In light of published “scaling laws” our hypothesis is that some tasks should be consistently more challenging than others based on characteristics of the task. The goal of this project was to begin quantifying how much information about LLM performance can be gained from the features of a model and a task. Two of our statistical models struggled to converge. Pass/fail test results may provide limited information for inference beyond model quality and task difficulty, but we see no evidence at this time for significant feature interaction effect sizes, arguing for simple models. Future work extending the models to capitalize on perplexity of ground truth answers is suggested. This project also introduces “Depth of Knowledge Variant Testing” as a strategy for more finely assessing language models on open domain question and answer tasks. We developed sets of questions that ask a language model to produce similar information while demonstrating increasing depth of knowledge, and also relabeled existing Q&A test questions with their depth of knowledge. Our results suggest further consideration of Bloom’s taxonomy and further refinement of prompts to properly elicit information at varying depths. In the course of this work, we set up a basic infrastructure for standardizing tasks and testing many language models on these tasks. In addition to testing the predictive quality of model features and performance across test suites, with this project we have introduced two new task features to contextualize each test question: the Dewey Classification main category of information covered, and the Bloom’s taxonomy level that corresponds to the depth of knowledge probed by the question. Splits across these and other features produced over five hundred task subtypes with distinct feature vectors, which we tested on half a dozen models.
Synchronous condensers have been used in power systems for decades, but they are now attracting interest as a potential means for solving certain issues associated with inverter-based resources in islanded power systems. They have been extensively studied in applications to large wind farms and weak grids, but there has been very little study of synchronous condensers in intentionally-islanded systems, and especially self-healing ones. This report documents the results of an LDRD project intended to create modeling tools for use in the study of synchronous condensers in these smaller off-grid systems, and to explore synchronous condensers in this application. The results indicate that synchronous condensers have the potential to provide many benefits, but there are also several unanswered questions and technical challenges requiring further study.
Corrosion challenges persist throughout SNL’s mission areas. The primary difficulty lies in the fact that corrosion typically manifests as isolated, rare events, making preemptive identification exceedingly difficult. Our current strategy for addressing corrosion issues, such as anomalies and SFIs, is similarly isolated and reactive. This method is costly, time-consuming, heavily dependent on a limited number of experts, and offers minimal understanding of the overall damage distribution within the stockpile. This technical challenge is not unique to corrosion but is also prevalent in other material aging phenomena, such as tin-whisker growth in lead-free solder and fatigue failure of springs.
MELCOR has been used extensively to facilitate virtual investigations into severe nuclear accidents for light-water reactors (LWRs). Non-light water reactors (non-LWRs) render some LWR-centric approaches potentially unsuitable. MELCOR has been instrumental in analyzing source terms for LWRs and has recently expanded its applicability to non-LWRs. To simplify radionuclide (RN) tracking, MELCOR currently groups elements into 17 classes, each containing representative species. This grouping, optimized for LWRs, is not appropriate for non-LWRs due to the different chemistry. This necessitates a reevaluation of radionuclide transport modeling. This report introduces a new class scheme for MELCOR tailored to MSR modeling, expanding the current 17 classes to 32 and are explained in the context of a UF4 fueled FLiBe carrier MSR. It provides a discussion and justification for the new groupings and outlines a methodology for discovering and defining additional classes in MELCOR using a sample calculated RN inventory and a Gibbs energy minimizer (GEM).
We study the problem of multifidelity uncertainty propagation for computationally expensive models. In particular, we consider the general setting where the high-fidelity and low-fidelity models have a dissimilar parameterization both in terms of number of random inputs and their probability distributions, which can be either known in closed form or provided through samples. We derive novel multifidelity Monte Carlo estimators which rely on a shared subspace between the high-fidelity and low-fidelity models where the parameters follow the same probability distribution, i.e., a standard Gaussian. We build the shared space employing normalizing flows to map different probability distributions into a common one, together with linear and nonlinear dimensionality reduction techniques, active subspaces and autoencoders, respectively, which capture the subspaces where the models vary the most. We then compose the existing low-fidelity model with these transformations and construct modified models with an increased correlation with the high-fidelity model, which therefore yield multifidelity estimators with reduced variance. A series of numerical experiments illustrate the properties and advantages of our approaches.
The Sea Change Project, managed and financed by SWITCH Maritime, produced the first zero-emission 100% hydrogen fuel-cell powered commercial ferry in the world. The MV Sea Change is a 70-foot catamaran equipped with a 360-kW fuel cell powertrain and a 250-bar 246 kg capacity hydrogen storage system. The 75-passenger vessel began public passenger service in July 2024 as part of the San Francisco Bay Ferry system. Summarized here are aspects of the design of the vessel, its construction, and vessel bunkering with high-pressure hydrogen. Results from a fuel cell performance analysis are also presented, along with estimates of the greenhouse gas (GHG) and criterial pollutant emissions assuming different sourcings of hydrogen. Lessons learned from the project will be briefly reviewed.
Seismic imaging methods are critical for Global Security and Energy & Homeland Security missions and activities that rely on subsurface characterization, but traditional methods remain computationally expensive and require significant labor hours and expertise to execute. Within the past few years, machine learning (ML), namely deep learning (DL), has been used to develop data-driven end-to-end full waveform inversion (FWI) methods to estimate 2D P-wave velocity (Vp) models in a fraction of the time as conventional FWI. These methods, however, are trained on simplistic acoustic wave seismic data and Vp models that are not realistic nor representative of real-world observations, leaving a large gap between the state-of-the-art and deployable, feasible, and practical DL FWI methods. Here, we generate a synthetic active-source, 3D, elastic wave seismic data set and a variety of Vp models with realistic geologic structure for training DL FWI methods. We evaluate six different methods that have performed well for acoustic DL FWI or medical imaging tasks using our more realistic dataset. We find that these six trained models do not match the performance of published acoustic end-to-end DL FWI methods, indicating more training data may be needed, physics may need to be incorporated to achieve good accuracy at the sacrifice of the end-to-end advantage, and/or novel methods need to be developed to enable end-to-end DL FWI methods to perform well for real-world seismic data.
Bugs in digital logic have led to some significant security vulnerabilities. Hardware bugs are particularly troublesome since they cannot be easily patched. Additionally, if the bug is in the root of trust, all trust built upon it can be vulnerable. Traditional testing either require a deep knowledge of the system, creative attack vectors and lots of human interaction. This is not scalable as there are very few engineers that can wear the hat of a designer, a verification engineer, and a cybersecurity expert. Hardware fuzzing is a relatively new research area in dynamic hardware testing. It has proven to be an effective method for discovering bugs, unexpected behaviors, and security vulnerabilities in software. While hardware fuzzing is new to the hardware domain, it has a strong track record in software testing. Fuzzing is a testing technique that randomly mutates the input data to uncover bugs or vulnerabilities in the design. It is especially good at finding corner cases that test engineers can not envision. Another advantage over other dynamic testing techniques is that, if done well, deep knowledge of the design is not required. Additionally, fuzzing scales well. If the system is set up correctly, it can run unsupervised for weeks if necessary. In this work, we propose using hardware fuzzing to improve the input vector generation for an information flow tracking tool. To get reasonable throughput of test vectors, an emulator is targeted as the execution platform. Efficient emulator execution has some specific requirements.
Modern computing systems are capable of exascale calculations, which are revolutionizing the development and application of high-fidelity numerical models in computational science and engineering. While these systems continue to grow in processing power, the available system memory has not increased commensurately, and electrical power consumption continues to grow. A predominant approach to limit the memory usage in large-scale applications is to exploit the abundant processing power and continually recompute many low-level simulation quantities, rather than storing them. However, this approach can adversely impact the throughput of the simulation and diminish the benefits of modern computing architectures. We present three novel contributions to reduce the memory burden while maintaining, and sometimes improving, performance in simulations based on finite element discretizations. The first contribution develops dictionary-based data compression schemes that detect and exploit the structure of the discretization, due to redundancies across the finite element mesh. While these schemes are shown to reduce memory requirements by more than 99% on meshes with large numbers of identical mesh cells, there are applications where this structure does not exist. The second contribution leverages a recently developed augmented Lagrangian optimization algorithm to enable r-adaptivity for meshes with the goal of enhancing the redundancies in the mesh. The third contribution extends these methods to patch-based linear solvers and preconditioners by compressing local matrices. Numerical results demonstrate the effectiveness of the proposed methods to detect, enhance and exploit mesh structure on a suite of examples inspired by large-scale applications.
The effects of raw water leaching at the U.S. SPR were modeled using SANSMIC. In addition to 18 caverns identified in previous leaching reports, six caverns have been identified for further monitoring based on the results of this report. SANSMIC continues to serve as a useful tool for monitoring changes in cavern shape. However, some caverns where string cuts were made in the middle of drawdowns are showing overprediction in the SANSMIC model when compared with recent sonars.
This paper describes Secure MQTT, a scheme proposed by Sandia National Laboratories that employs Identity-Based Encryption (IBE) for securing MQTT communications. Secure MQTT assumes and is conformant with an MQTT version 5 environment.
We propose a novel approach to developing formally verified systems through Multi-rigor Agile Verification. Multi-rigor Agile Verification is rooted in the hypothesis of Rigor Independence, that a system’s specification and verification architecture depend primarily on the system requirements to be verified, and they depend very little on the rigor level of the methods used to verify those requirements. Due to its iterative nature, Multi-rigor Agile Verification promises to mitigate many of the high upfront design costs experienced by formally verified systems and to deliver a better-architected, and thus better-trusted, system in the end. We then discuss the tooling needed to perform Multi-rigor Agile Verification and go in depth to build one of those tools, which directly generates executable prototype code from declarative formal specifications using the Maude rewrite-logic framework.
For acceptable implementations of technologies like wireless communications, remote monitoring, etc., strong mitigations must be developed and evaluated to ensure that new attack pathways do not increase risk for Advanced Reactors. Secure Elements can be adopted and adapted for this purpose based on tamper resistance and cryptographic abilities, but research must be done to properly integrate into critical components such as FPGA-based Important to Safety systems in conjunction with current and future regulations on cyber security features in Advanced Reactors. Typically, the integration of a Secure Element happens during the POST and UEFI boot of a computing platform, performed by the Operating System, which is not possible with FPGAs because they do not include these firmware components. Work must be done to identify a reliable and secure method for integration in FPGA-based systems which lack Operating Systems and therefore complex boot procedures, system calls, etc.
U.S. nuclear power facilities face increasing challenges in meeting dynamic security requirements caused by evolving and expanding threats while keeping costs reasonable to make nuclear energy competitive. The past approach has often included implementing security features after a facility has been designed and without attention to optimization, which can lead to cost overruns. Incorporating security into the design process can provide robust, cost-effective, and sufficient physical protection systems. The purpose of this report is to capture lessons learned by the Advanced Reactor Safeguards and Security (ARSS) program that may be beneficial for other advanced and small modular reactor (SMR) vendors to use when developing security systems and postures. This report will capture relevant information that can be used in the security-by-design (SeBD) process for SMR and microreactor vendors.
Afterglow in x-ray imaging for high-speed radiography is a constraint that limits imaging systems to low-light/fast decay screens which create poor data. Current approaches focus purely on using low-light yield screens with fast decay to avoid multiple exposure pileup due to afterglow. The goal of this work is to develop a statistical estimation approach to leverage afterglow to improve image quality thus allowing for higher quality imaging components to be used. This will allow for bright screens will slow decay to be used, and then a post-processing step applies the statistical estimation to separate each frame with superior signal compared to low-light/fast decay screens.
Achieving maximum performance in a Z-pinch Magnetized Liner Inertial Fusion (MagLIF) implosion requires high levels of symmetry to reach high pressures, high areal densities, and high efficiency energy coupling between the compressed fuel and the imploding shell. In this work, we present the development of a nonlinear model to describe an asymmetric MagLIF liner undergoing magneto-Rayleigh-Taylor (MRT) instabilities in the z = 0 plane. We investigate and present results from our model that show the degradation of the pressure at stagnation under various kinds of asymmetries that can arise during MagLIF experiments.
This report summarizes the work performed under the author's two-year John von Neumann LDRD project, which involves the non-intrusive surrogate modeling of dynamical systems with remarkable structural properties. After a brief introduction to the topic, technical accomplishments and project metrics are reviewed including peer-reviewed publications, software releases, external presentations and colloquia, as well as organized conference sessions and minisymposia. The report concludes with a summary of ongoing projects and collaborations which utilize the results of this work.
Anomalous behavior poses serious risks to assured performance and reliability of complex, high-consequence systems. For spaceborne assets and their state-of-health (SOH) telemetry, the challenges of high-dimensional data of varying data types are compounded by computational limitations from size, weight, and power (SWaP) constraints as well as data availability. Automated anomaly detection methods tend to perform poorly under these constraints, while current operational approaches can introduce delays in response time due to the manual, retrospective processes for understanding system failures. As a result, presently deployed space systems, and those deployed in the near future, face situations where mission operations might be delayed or only be able to operate under degraded capabilities. Here, we examine a near-term lightweight solution that provides real-time detection capabilities for rare events and assess state-of-the-art anomaly detection techniques against real SOH telemetry from space platforms. This report describes our methodology and research, which could support more automated capabilities for comprehensive space operations as well as for other resource-constrained edge applications.
