Search NASA⌕ Search

SEARCH · Search NASA

Results for “Graph embedding”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

231 records · Page 13
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

NuGraph2 with context-aware inputs: physics-inspired improvements in semantic segmentation

Graph neural networks have recently shown strong promise for event reconstruction tasks in Liquid Argon Time Projection Chambers, yet their performance remains limited for underrepresented classes of particles, such as Michel electrons. In this work, we investigate physics-informed strategies to improve semantic segmentation within the NuGraph2 architecture. We explore three complementary approaches: (i) enriching the input representation with context-aware features derived from detector geometry and track continuity, (ii) introducing auxiliary decoders to capture class-level correlations, and (iii) incorporating energy-based regularization terms motivated by Michel electron energy distributions. Experiments on MicroBooNE public datasets show that physics-inspired feature augmentation yields the largest gains, particularly boosting Michel electron precision and recall by disentangling overlapping latent space regions. In contrast, auxiliary decoders and energy-regularization terms provided limited improvements, partly due to the hit-level nature of NuGraph2, which lacks explicit particle- or event-level representations. Our findings highlight that embedding physics context directly into node-level inputs is more effective than imposing task-specific auxiliary losses, and suggest that future hierarchical architectures such as NuGraph3, with explicit particle- and event-level reasoning, will provide a more natural setting for advanced decoders and physics-based regularization. The code for this work is publicly available on Github at https://github.com/vitorgrizzi/nugraph_phys/tree/main_phys.

Other Experiments↗
DOE OSTI2024

Graph-learning approach to combine multiresolution seismic velocity models

SUMMARY The resolution of velocity models obtained by tomography varies due to multiple factors and variables, such as the inversion approach, ray coverage, data quality, etc. Combining velocity models with different resolutions can enable more accurate ground motion simulations. Toward this goal, we present a novel methodology to fuse multiresolution seismic velocity maps with probabilistic graphical models (PGMs). The PGMs provide segmentation results, corresponding to various velocity intervals, in seismic velocity models with different resolutions. Further, by considering physical information (such as ray path density), we introduce physics-informed probabilistic graphical models (PIPGMs). These models provide data-driven relations between subdomains with low (LR) and high (HR) resolutions. Transferring (segmented) distribution information from the HR regions enhances the details in the LR regions by solving a maximum likelihood problem with prior knowledge from HR models. When updating areas bordering HR and LR regions, a patch-scanning policy is adopted to consider local patterns and avoid sharp boundaries. To evaluate the efficacy of the proposed PGM fusion method, we tested the fusion approach on both a synthetic checkerboard model and a fault zone structure imaged from the 2019 Ridgecrest, CA, earthquake sequence. The Ridgecrest fault zone image consists of a shallow (top 1 km) high-resolution shear-wave velocity model obtained from ambient noise tomography, which is embedded into the coarser Statewide California Earthquake Center Community Velocity Model version S4.26-M01. The model efficacy is underscored by the deviation between observed and calculated traveltimes along the boundaries between HR and LR regions, 38 per cent less than obtained by conventional Gaussian interpolation. The proposed PGM fusion method can merge any gridded multiresolution velocity model, a valuable tool for computational seismology and ground motion estimation.

Geochemistry & Geophysics↗
DOE OSTI2021

HAM: Hotspot-Aware Manager for Improving Communications with 3D-Stacked Memory

merging High-Performance Computing (HPC) workloads, such as graph analytics, machine learning, and big data science, are data-intensive. Data-intensive workloads usually present fine-grained memory accesses with limited or no data locality, and thus incur frequent cache misses and low utilization of memory bandwidth. 3D-stacked memory devices such as Hybrid Memory Cube (HMC) and High Bandwidth Memory (HBM) can provide significantly higher bandwidth than conventional memory modules. However, the traditional interfaces and optimization methods for JEDEC DDR devices do not allow to fully exploit the potential performance of 3D-stacked memory with the massive amount of irregular memory accesses of data-intensive applications. In this paper, we propose a novel Hotspot-Aware Manager (HAM) infrastructure for 3D-stacked memory devices capable of optimizing memory access streams via request aggregation, hotspot detection, and in-memory prefetching. %and an associated hotspot-aware page policy. We present the HAM design and implementation, and simulate it on a system using RISC-V embedded cores with attached HMC devices. We extensively evaluate HAM with over 12 benchmarks and applications representing diverse irregular memory access patterns. The results show that, on average, HAM reduces redundant requests by 37.51\% and increases the prefetch buffer hit rate by 4.2 times, compared to a baseline streaming prefetcher. On the selected benchmark set, HAM provides performance gains of 21.81\% in average (up to 34.28\%) and power savings of 35.07\% over a standard 3D-stacked memory.

Wang, Xi↗
DOE OSTI2012

Transfer matrix computation of generalized critical polynomials in percolation

Percolation thresholds have recently been studied by means of a graph polynomial PB(p), henceforth referred to as the critical polynomial, that may be defined on any periodic lattice. The polynomial depends on a finite subgraph B, called the basis, and the way in which the basis is tiled to form the lattice. The unique root of P B (p) in [0, 1] either gives the exact percolation threshold for the lattice, or provides an approximation that becomes more accurate with appropriately increasing size of B. Initially P B (p) was defined by a contraction-deletion identity, similar to that satisfied by the Tutte polynomial. Here, we give an alternative probabilistic definition of P B (p), which allows for much more efficient computations, by using the transfer matrix, than was previously possible with contraction-deletion. We present bond percolation polynomials for the (4, 82), kagome, and (3, 122) lattices for bases of up to respectively 96, 162, and 243 edges, much larger than the previous limit of 36 edges using contraction-deletion. We discuss in detail the role of the symmetries and the embedding of B. For the largest bases, we obtain the thresholds p c (4, 82) = 0.676 803 329 · · ·, p c (kagome) = 0.524 404 998 · · ·, p c (3, 122) = 0.740 420 798 · · ·, comparable to the best simulation results. We also show that the alternative definition of P B (p) can be applied to study site percolation problems.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2013

Transfer matrix computation of critical polynomials for two-dimensional Potts models

We showed, In our previous work, that critical manifolds of the q-state Potts model can be studied by means of a graph polynomial P B (q, v), henceforth referred to as the critical polynomial. This polynomial may be defined on any periodic two-dimensional lattice. It depends on a finite subgraph B, called the basis, and the manner in which B is tiled to construct the lattice. The real roots v = e K — 1 of P B (q, v) either give the exact critical points for the lattice, or provide approximations that, in principle, can be made arbitrarily accurate by increasing the size of B in an appropriate way. In earlier work, P B (q, v) was defined by a contraction-deletion identity, similar to that satisfied by the Tutte polynomial. Here, we give a probabilistic definition of P B (q, v), which facilitates its computation, using the transfer matrix, on much larger B than was previously possible.We present results for the critical polynomial on the (4, 8 2 ), kagome, and (3, 12 2 ) lattices for bases of up to respectively 96, 162, and 243 edges, compared to the limit of 36 edges with contraction-deletion. We discuss in detail the role of the symmetries and the embedding of B. The critical temperatures v c obtained for ferromagnetic (v > 0) Potts models are at least as precise as the best available results from Monte Carlo simulations or series expansions. For instance, with q = 3 we obtain v c (4, 8 2 ) = 3.742 489 (4), v c (kagome) = 1.876 459 7 (2), and v c (3, 12 2 ) = 5.033 078 49 (4), the precision being comparable or superior to the best simulation results. More generally, we trace the critical manifolds in the real (q, v) plane and discuss the intricate structure of the phase diagram in the antiferromagnetic (v < 0) region.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2011

Simrank: Rapid and sensitive general-purpose k-mer search tool

Terabyte-scale collections of string-encoded data are expected from consortia efforts such as the Human Microbiome Project (http://nihroadmap.nih.gov/hmp). Intra- and inter-project data similarity searches are enabled by rapid k-mer matching strategies. Software applications for sequence database partitioning, guide tree estimation, molecular classification and alignment acceleration have benefited from embedded k-mer searches as sub-routines. However, a rapid, general-purpose, open-source, flexible, stand-alone k-mer tool has not been available. Here we present a stand-alone utility, Simrank, which allows users to rapidly identify database strings the most similar to query strings. Performance testing of Simrank and related tools against DNA, RNA, protein and human-languages found Simrank 10X to 928X faster depending on the dataset. Simrank provides molecular ecologists with a high-throughput, open source choice for comparing large sequence sets to find similarity.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS1986

Models of polarized infrared emission from bipolar nebulae

Many stars with circumstellar dust shells show a high degree of linear polarization (Sato et al. 1985). We are developing a model which assumes that the polarization arises from scattering by circumstellar dust. Our model assumes a geometry in which the star is surrounded by an optically thin spherical dust shell and embedded within an optically thick disk. This geometry is consistent with that proposed for objects with bipolar molecular outflow. This is important because many bipolar flow objects have also been observed to be highly polarized. The high degree of linear polarization is produced because the disk differentially attenuates the light from the star. The light incident from the point source is attenuated by a factor of exp(-tau/cos theta) where theta is the angle between a ray from the point source to the scatterer and a ray normal to the disk; tau is the optical depth at the wavelength of interest. Hence, the light scattered from the regions directly above and below the disk give the largest contribution to the total flux. The scattering angle for light from these regions is near 90 deg., so the light is strongly polarized and, in the Rayleigh scattering regime, is polarized parallel to the disk. The Stokes parameters for the scattered light from each particle in the shell are calculated by using the scattering matrix elements generated by a Mie scattering program. After the Stokes parameters for each particle are computed they are summed to give the Stokes parameters for the entire shell. Two graphs are presented which show the intensity and polarization spectrum generated by our model using the optical constants for astronomical silicates as defined by Draine and Lee (1984).

Burns, M. S.↗
DOE OSTI2012

Effective Software Design and Development for the New Graph Architecture HPC Machines

Software applications need to change and adapt as modern architectures evolve. Nowadays advancement in chip design translates to increased parallelism. Exploiting such parallelism is a major challenge in modern software engineering. Multicore processors are about to introduce a significant change in the way we design and use fundamental data structures. In this work we describe the design and programming principles of a software library of highly concurrent scalable and nonblocking data containers. In this project we have created algorithms and data structures for handling fundamental computations in massively multithreaded contexts, and we have incorporated these into a usable library with familiar look and feel. In this work we demonstrate the first design and implementation of a wait-free hash table. Our multiprocessor data structure design allows a large number of threads to concurrently insert, remove, and retrieve information. Non-blocking designs alleviate the problems traditionally associated with the use of mutual exclusion, such as bottlenecks and thread-safety. Lock-freedom provides the ability to share data without some of the drawbacks associated with locks, however, these designs remain susceptible to starvation. Furthermore, wait-freedom provides all of the benefits of lock-free synchronization with the added assurance that every thread makes progress in a finite number of steps. This implies deadlock-freedom, livelock-freedom, starvation-freedom, freedom from priority inversion, and thread-safety. The challenges of providing the desirable progress and correctness guarantees of wait-free objects makes their design and implementation difficult. There are few wait-free data structures described in the literature. Using only standard atomic operations provided by the hardware, our design is portable; therefore, it is applicable to a variety of data-intensive applications including the domains of embedded systems and supercomputers.Our experimental evaluation shows that our hash table design outperforms the most advanced locking solution, provided by Intel's TBB library, by 22%. When compared to more traditional locking designs we show a performance improvement by a factor of 7.92. When compared to alternative non-blocking designs, our hash table demonstrates solid performance gains in a large majority of cases, typically by a factor of 3.44.

99 GENERAL AND MISCELLANEOUS↗
NASA NTRS2012

Software to Control and Monitor Gas Streams

This software package interfaces with various gas stream devices such as pressure transducers, flow meters, flow controllers, valves, and analyzers such as a mass spectrometer. The software provides excellent user interfacing with various windows that provide time-domain graphs, valve state buttons, priority- colored messages, and warning icons. The user can configure the software to save as much or as little data as needed to a comma-delimited file. The software also includes an intuitive scripting language for automated processing. The configuration allows for the assignment of measured values or calibration so that raw signals can be viewed as usable pressures, flows, or concentrations in real time. The software is based on those used in two safety systems for shuttle processing and one volcanic gas analysis system. Mass analyzers typically have very unique applications and vary from job to job. As such, software available on the market is usually inadequate or targeted on a specific application (such as EPA methods). The goal was to develop powerful software that could be used with prototype systems. The key problem was to generalize the software to be easily and quickly reconfigurable. At Kennedy Space Center (KSC), the prior art consists of two primary methods. The first method was to utilize Lab- VIEW and a commercial data acquisition system. This method required rewriting code for each different application and only provided raw data. To obtain data in engineering units, manual calculations were required. The second method was to utilize one of the embedded computer systems developed for another system. This second method had the benefit of providing data in engineering units, but was limited in the number of control parameters.

Arkin, C.↗
DOE OSTI2024

Acoustic-based monitoring and machine learning of component status for microreactor applications

This report provides a description and assessment of recent efforts to couple acoustic-based experimental measurements and characterization with machine learning models in order to enhance structural health monitoring capabilities for nuclear microreactors. With resilient embedded sensors in development by others supported by programs funded by the US Department of Energy’s Office of Nuclear Energy, the work described herein builds upon ongoing efforts to improve non-destructive testing technology that relates measured acoustic signatures to component stresses and/or structural defects, using a combination of new experimental measurements and machine learning architectures. The experimental procedure remained similar to that developed for the previous year’s demonstration of damage detection by the authors, with the same damaged sample tested under similar applied stress conditions. Notably, a new mounting fixture was designed and implemented to improve measurement consistency and a more sophisticated laser Doppler vibrometer was employed to make high-fidelity vibration measurements. Two nominally identical sets of training data were collected for each experimental setup to better understand the repeatability of the experiment and to better test the generality of trained neural network models. Additionally, we obtained new high-quality 3D mode shapes of the damaged test article at various stress and excitation levels, providing greater insights into the physical response of the sample during testing. Previously, we demonstrated that a machine learning model based on a convolutional neural network can predict structural details of an artificially introduced interface (intact, rough cut, smooth cut), and the applied torque level. In this study, we have transitioned to graph-based neural network architectures to better develop and test a flexible framework that is more suitable to being transferred away from controlled benchtop experiments and into more applied settings where less-structured data inputs may be expected. In general, performance testing of a graph neural network on frequency-domain representations of the data indicates strong and consistent identification of test conditions for datasets recorded on damaged components. With goals of predicting damage location and other changing experimental conditions using limited datasets, predictive models using a graph neural network architecture correctly predicted the applied torque level with an accuracy of 85% using only a single measurement point and predicted within one torque level in 95% of test windows. Predictions of damage location had limited success due to the symmetry and minimal number of the damage scenarios presented during model training. Results were ambiguous as to whether the model could detect the location of the artificial damage, or if it was instead learning the location of a given measurement point on the part and subsequently detecting which points were closest to the location of the damage. This finding will be factored into upcoming planned work on damaged graphite components, where new experimental tests with a larger number and variety of damage scenarios are expected to provide improved validation of recent developments in monitoring methodology.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2017

Using the D-Wave 2X Quantum Computer to Explore the Formation of Global Terrorist Networks

Social networks with signed edges (+/-) play an important role in an area of social network theory called structural balance. In these networks, edges represent relationships that are labeled as either friendly (+) or hostile (-). A signed social network is balanced only if all cycles of three or more nodes in the graph have an odd number of hostile edges. A fundamental property of a balanced network is that it can be cleanly divided into 2 factions, where all relationships within each faction are friendly, and all relationships between members of different factions are hostile. The more unbalanced a network is, the more edges will fail to adhere to this rule, making factions more ambiguous. Social theory suggests unbalanced networks should be unstable, a finding that has been supported by research on gangs, which shows that unbalanced relationships are associated with greater violence, possibly due to this increased ambiguity about factional allegiances (Nakamura et al). One way to estimate the imbalance in a network, if only edge relationships are known, is to assign nodes to factions that minimize the number of violations of the edge rule described above. This problem is known to be computationally NP-hard. However, Facchetti et al. have pointed out that it is equivalent to an Ising model with a Hamiltonian that effectively counts the number of edge rule violations. Therefore, finding the assignment of factions that minimizes energy of the equivalent Ising system yields an estimate of the imbalance in the network. Based on the Ising model equivalence of the signed-social network balance problem, we have used the D-Wave 2X quantum annealing computer to explore some aspects of signed social networks. Because connectivity in the D-Wave computer is limited to its particular native topology, arbitrary networks cannot be represented directly. Rather, they must be “embedded” using a technique in which multiple qubits are chained together with special weights to simulate a collection of nodes with the required connectivity. This limits the size of a fully connected network in the D-Wave to about 50 simulated nodes, using all of the approximately 1150 qubits in the machine. In order to keep within this limitation, while exploring a problem of potential social relevance, we constructed time series of historical network snapshots from Stanford’s Mapping Militants Project, where nodes represent militant organizations, and edges represent either alliances or rivalries between organizations. We constructed two series from different theaters – Iraq and Syria – spanning timelines from about 2000 to 2016, each with networks whose maximum size was in the 20-30 node range. Computationally, our experience suggests D-Wave technology is promising, providing fast, nearly constant scaling of computational effort in the main part of the calculation that relies on the quantum annealing cycle. However, the cost of embedding an arbitrary network of interest in the D-Wave native topology scales poorly. If the embedding cost can be amortized relative to the annealing cycle, it may be possible to gain a substantial advantage over classical computing methods, provided a large enough network can be accommodated by partitioning into subnetworks or some similar strategy. In terms of our application to networks of militant organizations, we found a rise in network imbalance in the Syrian theater that appears to correspond roughly with the entrance of the Islamic State into a milieu already populated with other groups, a phenomenon we plan to explore in more detail. In these very preliminary results, we also noticed that during at least one period where both the size and imbalance of the network increased substantially, the imbalance per edge seemed to remain fairly steady. This may suggest some adaptive behavior among the participating factions, which may also warrant further exploration.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Single-Point Extrapolation to the Complete Basis Set Limit through Deep Learning

Machine learning (ML) offers an attractive method for making predictions about molecular systems while circumventing the need to run expensive electronic structure calculations. Once trained on ab initio data, the promise of ML is to deliver accurate predictions of molecular properties that were previously computationally infeasible. In this work, we develop and train a graph neural network model to correct the basis set incompleteness error (BSIE) between a small and large basis set at the RHF and B3LYP levels of theory. Our results show that, when compared to fitting to the total potential, an ML model fitted to correct the BSIE is better at generalizing to systems not seen during training. We test this ability by training on single molecules while evaluating on molecular complexes. We also show that ensemble models yield better behaved potentials in situations where the training data is insufficient. However, even when only fitting to the BSIE, acceptable performance is only achieved when the training data sufficiently resemble the systems one wants to make predictions on. As a result, the test error of the final model trained to predict the difference between the cc-pVDZ and cc-pV5Z potential is 0.184 kcal/mol for the B3LYP density functional, and the ensemble model accurately reproduces the large basis set interaction energy curves on the S66x8 dataset.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
NASA NTRS2004

Reannealed Fiber Bragg Gratings Demonstrated High Repeatability in Temperature Measurements

Fiber Bragg gratings (FBGs) are formed by periodic variations of the refractive index of an optical fiber. These periodic variations allow an FBG to act as an embedded optical filter, passing the majority of light propagating through a fiber while reflecting back a narrow band of the incident light. The peak reflected wavelength of the FBG is known as the Bragg wavelength. Since the period and width of the refractive index variation in the fiber determines the wavelengths that are transmitted and reflected by the grating, any force acting on the fiber that alters the physical structure of the grating will change the wavelengths that are transmitted and reflected by it. Both thermal and mechanical forces acting on the grating will alter its physical characteristics, allowing the FBG sensor to detect both the temperature variations and the physical stresses and strains placed upon it. This ability to sense multiple physical forces makes the FBG a versatile sensor. To assess the feasibility of using Bragg gratings as temperature sensors for propulsion applications, researchers at the NASA Glenn Research Center evaluated the performance of Bragg gratings at elevated temperatures for up to 300 C. For these purposes, commercially available polyimide-coated high-temperature gratings were used that were annealed by the manufacturer to 300 C. To assure the most thermally stable gratings at the operating temperatures, we reannealed the gratings to 400 C at a very slow rate for 12 to 24 hr until their reflected optical powers were stabilized. The reannealed gratings were then subjected to periodic thermal cycling from room temperature to 300 C, and their peak reflected wavelengths were monitored. The setup shown is used for reannealing and thermal cycling the FBGs. Signals from the photodetectors and the spectrum analyzer were fed into a computer equipped with LabVIEW software. The software synchronously monitored the oven/furnace temperature and the optical spectrum analyzer as well as processed the data. Experimental results presented in the following graph show typical wavelength versus temperature dependence of a reannealed FBG through six thermal cycles (80 hr). The average standard deviation of the temperature-to-wavelength relationship ranged from 1.86 to 2.92 C over the six thermal cycles each grating was subjected to. This is an error of less than 1.0 percent of full scale throughout the entire evaluation temperature range from ambient to 300 C.

Adamovsky, Grigory↗
DOE OSTI2013

Sensor Network Demonstration for In Situ Decommissioning - 13332

Florida International University's (FIU's) Applied Research Center is currently supporting the Department of Energy's (DOE) Environmental Management Office of D and D and Facility Engineering program. FIU is supporting DOE's initiative to improve safety, reduce technical risks, and limit uncertainty within D and D operations by identifying technologies suitable to meet specific facility D and D requirements, assessing the readiness of those technologies for field deployment, and conducting feasibility studies and large scale demonstrations of promising technologies. During FY11, FIU collaborated with Savannah River National Laboratory in the development of an experimental test site for the demonstration of multiple sensor systems for potential use in the in situ decommissioning process. In situ decommissioning is a process in which the above ground portion of a facility is dismantled and removed, and the underground portion is filled with a cementious material such as grout. In such a scenario, the question remains on how to effectively monitor the structural health of the grout (cracking, flexing, and sinking), as well as track possible migration of contaminants within and out of the grouted monolith. The right types of sensors can aid personnel in better understanding the conditions within the entombed structure. Without sensors embedded in and around the monolith, it will be very difficult to estimate structural integrity and contaminant transport. Yet, to fully utilize the appropriate sensors and the provided data, their performance and reliability must be evaluated outside a laboratory setting. To this end, a large scale experimental setup and demonstration was conducted at FIU. In order to evaluate a large suite of sensor systems, FIU personnel designed and purchased a pre-cast concrete open-top cube, which served as a mock-up of an in situ DOE decommissioned facility. The inside of the cube measures 10 ft x 10 ft x 8 ft. In order to ensure that the individual sensors would be immobilized during the grout pouring activities, a set of nine sensor racks were designed. The 270 sensors provided by Idaho National Laboratory (INL), Mississippi State University (MSU), University of Houston (UH), and University of South Carolina (USC) were secured to these racks based on predetermined locations. Once sensor racks were installed inside the test cube, connected and debugged, approximately 32 cubic yards of special grout material was used to entomb the sensors. MSU provided and demonstrated four types of fiber loop ring-down (FLR) sensors for detection of water, temperature, cracks, and movement of fluids. INL provided and demonstrated time differenced 3D electrical resistivity tomography (ERT), advanced tensiometers for moisture content, and thermocouples for temperature measurements. University of Houston provided smart aggregate (SA) sensors, which detect crack severity and water presence. An additional UH sensor system demonstrated was a Fiber Bragg Grating (FBG) fiber optic system measuring strain, presence of water, and temperature. USC provided a system which measured acoustic emissions during cracking, as well as temperature and pH sensors. All systems were connected to a Sensor Remote Access System (SRAS) data networking and collection system designed, developed and provided by FIU. The purpose of SRAS was to collect and allow download of the raw sensor data from all the sensor system, as well as allow upload of the processed data and any analysis reports and graphs. All this information was made available to the research teams via the Deactivation and Decommissioning Knowledge Management and Information Tool (D and D KM-IT). As a current research effort, FIU is performing an energy analysis, and transferring several sensor systems to a Photovoltaic (PV) System to continuously monitor energy consumption parameters and overall power demands. Also, One final component of this research is focusing on developing an integrated data network to capture, log and analyze sensor system data in near real time from a single interface. FIU personnel and DOE Fellows monitored the progress and condition of the sensors for a period of six months. During this time, the sensors recorded data pertaining to strain, compression, temperature, crack detection, moisture presence, fluid mobility, shock resistance, monolith movement, and electrical resistivity. In addition, FIU regularly observed the curing process of the grout and documented the cube condition via the nine racks of sensors. The sensors held up throughout the curing process, withstood the natural elements for six months, and monitored the integrity of the grout. The large scale experiment and demonstration conducted at FIU was the first of its kind to demonstrate the feasibility of state of the art sensors for in situ decommissioning applications. These efforts successfully measured the durability, performance, and precision of the sensors in question as well as monitored and recorded the curing process of the selected grout material under natural environmental conditions. The current energy analysis work is resulting in data on the constraints placed by some of the sensor systems on a power network that requires high reliability and low losses. In addition, a sensor system demonstration has determined that it is feasible to develop an integrated data network where data can be accessed in near real-time from all systems, thereby allowing for larger-scale integrated system testing to be performed. Information collected during the execution of this research project will aid decision makers in the identification of sensors to be used in nuclear facilities selected for in situ decommissioning. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2026

Deciphering the Solvation Structure of Aqueous ZnCl 2 Solutions from X-ray Absorption Spectra Using the Interpretable Graph Neural Network

Machine learning (ML) provides powerful pathways for predicting spectroscopic observables from atomic structures, but its broader impact depends on making model predictions interpretable in terms of physical and chemical principles. Here, we introduce a physics-guided graph neural network (GNN) model that predicts Zn K-edge X-ray spectroscopy (XAS) spectra of aqueous ZnCl 2 solutions. Training data are generated from ab initio XAS calculations on molecular dynamics snapshots obtained using a machine learning interatomic potential. The GNN reproduces experimental spectra across concentrations from dilute (<0.1 m) to highly concentrated (30 m, “water-in-salt”) regimes and scales efficiently to large, disordered liquid systems beyond the reach of conventional ab initio approaches. Gradient-based attribution analysis reveals that the model learns physically meaningful structure-spectrum relationships. Ligand-specific attributions reflect orbital hybridization patterns and the origin of the excitations derived from the density functional theory. Bond-length attributions recover spectral shifts consistent with multiple-scattering theory. Finally, this work bridges data-driven prediction with electronic-structure theory, establishing a general paradigm for interpretable ML that links atomic structure, electronic structure, and spectroscopic observables.

25 ENERGY STORAGE↗