Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “large scale network”

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.

At least 1,261 records · Page 70
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Measurement-informed Dynamic Aggregation of Distribution Systems

This paper proposes a measurement-informed dynamic aggregation methodology in order to create equivalent representations of distribution systems that are compatible with large-scale transmission analysis. By optimizing an equivalent feeder parameters using time-series measurements of active power, reactive power, and voltage at the Point of Interconnection (POI), the approach yields simplified yet dynamically accurate equivalents. Implemented in PSCAD with models of photovoltaic–battery systems, three-phase motors, and static loads, the method employs hybrid differential evolution and bounded least-squares optimization laying the foundation for for real-time state estimation and optimized sensor placement in distribution networks.

Ahmed, Kazi Ishrak [University of Tennessee, Knoxv↗
DOE OSTI2025

Micropolar deep material network

This study extends the Deep Material Network (DMN), a physics-informed machine learning framework, to predict the homogenized mechanical response of composite materials with micropolar (Cosserat-type) constitutive behavior. This extension incorporates microstructure-dependent size effects, enabling accurate, efficient, and size-aware predictions for composites with complex internal architectures. While traditional, direct numerical simulation micropolar models effectively capture size effects by introducing extra local degrees of freedom, they bring significant computational challenges, particularly for multiscale analyses relevant to engineering applications. The micropolar DMN developed in this paper achieves high accuracy while significantly reducing computation time compared to micropolar direct numerical simulations. This advancement enables multiscale analyses and parameter studies that were previously impractical, such as high-cycle fatigue simulations and comprehensive investigations of internal length scale effects notably in size-dependent plastic response and the optimization of lattice structures. By uniting microstructure-sensitive modeling, physics-driven learning, and scalable surrogate modeling, the micropolar DMN paves the way for accelerated material design, large-scale parametric studies, and the reliable incorporation of size-dependent effects across a wide range of engineering applications, including optimization and next-generation composite design.

36 MATERIALS SCIENCE↗
DOE OSTI2020

ContourNet: Salient Local Contour Identification for Blob Detection in Fusion Plasma Simulation Data

We present ContourNet, a deep learning approach to identify salient local isocontours as blobs in large-scale 5D gyrokinetic tokamak simulation data. Blobs-regions of high turbulence that run along the edge wall down toward the diverter and can damage the tokamak-are non-well-defined features but have been empirically localized by isocontours in 2D normalized fluctuating density fields. The key of our study is to train ContourNet to follow the empirical rules to detect blobs over the time-varying simulation data. The architecture of ContourNet is a convolutional neural segmentation network: the inputs are the density field and a rasterized isocontour; the output is a set of isocontour encircling blobs. At the training stage, we feed the network with manually identified isocontours and propagated labels. At the inference stage, we extract isocontours from the segmented blob regions. Results show that our approach can achieve both high accuracy and performance, which enables scientists to understand the blob dynamics influencing the confinement of the plasma.

Blob detection↗
DOE OSTI2021

Reconstructing porous media using generative flow networks

One area of intense scientific interest for the study of sandstones, carbonates, and shale at the pore scale is the use of limited image and petrophysical data to generate multiple realizations of a rock’s pore structure. Such images aid efforts to quantify uncertainty in petrophysical properties, including porosity–permeability transforms. We develop and evaluate a deep learning-based method to synthesize porous media volumes using a so-called generative flow model trained on x-ray microscope images of rock texture and pore structure. These models are optimized on a log-likelihood objective and they synthesize large and realistic three-dimensional images. Further, we demonstrate the rapid generation of sandstone image volumes that display realism as gauged by quantitative comparison of topological features using Minkowski functionals of porosity, specific surface area, and the Euler–Poincaré characteristic (i.e., pore connectivity). We also evaluate the single-phase permeability using Navier–Stokes and lattice Boltzmann methods and show that transport properties of the generated samples match measured trends.

58 GEOSCIENCES↗
DOE OSTI2024

Modifications to Solar Titan-130 Combustion Systems for Efficient, High Turndown Operation

The project team of Southwest Research Institute® (SwRI®), Solar Turbines Incorporated (Solar), the Electric Power Research Institute (EPRI), the University of California, Irvine (UCI), and the Georgia Institute of Technology (Georgia Tech) investigated methods to allow higher efficiency part-load operation of a Solar Titan 130 gas turbine. The objective was to develop a low-emission combustion system capable of sustaining combustion and avoiding lean blowout during high turndown operation, which would allow the gas turbine to operate as efficiently as possible at part load. Currently, electric utility markets are beginning to experience substantial increases in renewable energy generation. Some of these renewable energy sources have highly variable output in an uncontrolled manner. In order to maintain grid stability, there is a need for power plants to ramp up power to the grid rapidly to make up for drops in renewable generation. This is often termed spinning reserve, but the size of this reserve may need to increase as renewable penetration into the electric utility market increases. Small combined heat and power (CHP) power plants provide a promising option for meeting this spinning reserve requirement. In order to operate in spinning reserve while still meeting the heat requirements for the CHP, the gas turbine needs to operate efficiently at very low loads. Efficient, high turndown operations in this engine are limited by the lean flammability limit of the premixed combustion system. This project sought enhance the lean operability range of the Titan 130 combustor. First, the project team participated in a brainstorming activity and ultimately selected two concepts to explore: fuel augmentation with hydrogen (H2) to improve the stability at lean operating conditions and modifications to the fuel nozzle to improve the emissions performance at lean operating conditions. Analytical and laboratory investigations were accomplished by UCI to investigate the efficacy of H2 addition at improving lean blow out (LBO) limits and the resulting emissions. These investigations used a variety of chemical reactor network (CRN) and CFD models, validated against laboratory data, to model the impact of H 2 and inform the experimental efforts accomplished by SwRI and Solar. Ultimately, both the CRN and CFD models yielded generally good agreement with the experimental data below a particular temperature threshold. Atmospheric tests of a full-scale T130 annular combustor were performed at SwRI facilities in San Antonio, Texas, to investigate the use of H 2 addition. For these tests, the T130 combustion system remained largely unchanged; minor modifications were performed to the fuel ducting to allow for the safe use of H 2 . The test ultimately demonstrated that the addition of H 2 to the fuel mixture significantly increased the AFR ratio at which the combustor could operate. This improvement to the LBO limit should allow for less use of compressor bleed and less throttling needed by the inlet guide vanes (IGV). This in turn could result in more efficient operation of the gas turbine at lower load points. The second modification explored in this work was a direct modification to the T130 injector. The project team hypothesized that modifications to the pilot of the T130 injector could provide lower emissions at high turn-down operations. These modifications were manufactured and explored by the team at Solar. High pressure rig tests, originally slated to occur at SwRI, were ultimately accomplished by Solar to maintain overall project budget and mitigate cost growth attributable to supply chain issues and inflation. The pressurized rig tests ultimately showed that the SwRI Project No. 18.24153 - DE-EE0008415 Page 2 Final Technical Report January 24, 2024 modifications did not significantly alter the performance of the combustion system at the high turn-down conditions; both the modified injectors and the baseline configuration exhibited elevated emissions comparted to the full-load operating condition. A final set of studies performed by EPRI investigated the benefit-cost of flexible CHP as well as a grid interconnection study for the California Independent System Operator (CAISO) grid. These studies considered: traditional CHP with no spinning reserve available for on-demand grid support, 50% flexible CHP where 50% of the machine’s capacity is consumed by on-site baseload operations while providing an additional 50% capacity for on-demand grid support, and 70% flexible CHP where 70% of capacity is consumed on-site by baseload operations and 30% is available for on-demand grid support. In all cases, the analyses showed a benefit-to-cost ratio greater than unity implying a positive net present value for all configurations. However, the traditional CHP showed the most economic benefit. These results are sensitive to several factors, many of which are not fully known and may vary over time. Thus site owners must be convinced that taking up the increased costs and risks from flexible CHP would be worth implementing. As the grid in California and across the country transition to incorporate larger renewable energy generation, flexible CHP can provide much needed operating reserves and dispatchability. Alternative fuel options, such as hydrogen blending and biofuels, may also lower carbon intensities of CHP. Flexible CHP should be examined in the evolving market to understand innovative business models, changes market rules and services, and new technologies.

20 FOSSIL-FUELED POWER PLANTS↗
NASA NTRS2003

[Development and Use of Hidrosig]

The NASA portion of this joint NSF-NASA grant consists of objective 2 and a part of objective 3. A major effort was made on objective 2, and it consisted of developing a numerical GIs environment called Hidrosig. This major research tool is being developed by the University of Colorado for conducting river-network-based scaling analyses of coupled water-energy-landform-vegetation interactions including water and energy balances, and floods and droughts, at multiple space-time scales.Objective 2: To analyze the relevant remotely sensed products from satellites, radars and ground measurements to compute the transported water mass for each complete Strahler stream using an 'assimilated water balance equation' at daily and other appropriate time scales. This objective requires analysis of concurrent data sets for Precipitation (PPT), Evapotranspiration (ET) and stream flows (Q) on river networks. To solve this major problem, our decision was to develop Hidrosig, a new Open-Source GIs software. A research group in Colombia, South America, developed the first version of Hidrosig, and Ricardo Mantilla was part of this effort as an undergraduate student before joining the graduate program at the University of Colorado in 2001. Hydrosig automatically extracts river networks from large DEMs and creates a "link-based" data structure, which is required to conduct a variety of analyses under objective 2. It is programmed in Java, which is a multi-platform programming language freely distributed by SUN under a GPL license. Some existent commercial tools like Arc-Info, RiverTools and others are not suitable for our purpose for two reasons. First, the source code is not available that is needed to build on the network data structure. Second, these tools use different programming languages that are not most versatile for our purposes. For example, RiverTools uses an IDL platform that is not very efficient for organizing diverse data sets on river networks. Hidrosig establishes a clear data organization framework that allows a simultaneous analysis of spatial fields along river network structures involving Horton- Strahler framework. Software tools for network extraction from DEMs and network-based analysis of geomorphologic and topologic variables were developed during the first year and a part of second year.

Gupta, Vijay K.↗
NASA NTRSDate not supplied

Quantum Hardware-Enabled Molecular Dynamics via Transfer Learning

The ability to perform ab initio molecular dynamics simulations using potential energy surfaces provided by quantum computers would open the door to virtually exact dynamics for a variety of chemical and biochemical systems, with impacts on catalysis and biophysics. Nonetheless, performing molecular dynamics on surfaces produced by quantum hardware has been hampered by the noisy energies typically produced by quantum computers and challenges associated with computing gradients and scaling to large systems interest. A recent set of advances in machine learning, known as transfer learning, provides a new path forward for molecular dynamics simulations on quantum hardware. Transfer learning offers a workaround, where one first trains models on larger, less accurate classical datasets and then refines them on smaller, more accurate quantum datasets. We explore this approach by training machine learning models to predict a molecule's potential energy based on its geometric structure using Behler-Parrinello neural networks. When successfully trained, the model enables energy gradient predictions necessary for dynamic simulations. To reduce the quantum resources needed, the model is initially trained with data derived from classical density functional theory and subsequently refined with a smaller dataset obtained from a variational quantum eigensolver optimization of the unitary coupled cluster ansatz. We show that this approach significantly reduces the size of the needed quantum training dataset while capturing the high accuracies needed within quantum chemistry simulations. The success of this two-step training method opens more opportunities to apply machine learning models on quantum data, a significant stride towards efficient quantum-classical hybrid computational models.

quantum computing↗
DOE OSTI2022

Probabilistic Neural Computing with Stochastic Devices

Abstract The brain has effectively proven a powerful inspiration for the development of computing architectures in which processing is tightly integrated with memory, communication is event‐driven, and analog computation can be performed at scale. These neuromorphic systems increasingly show an ability to improve the efficiency and speed of scientific computing and artificial intelligence applications. Herein, it is proposed that the brain's ubiquitous stochasticity represents an additional source of inspiration for expanding the reach of neuromorphic computing to probabilistic applications. To date, many efforts exploring probabilistic computing have focused primarily on one scale of the microelectronics stack, such as implementing probabilistic algorithms on deterministic hardware or developing probabilistic devices and circuits with the expectation that they will be leveraged by eventual probabilistic architectures. A co‐design vision is described by which large numbers of devices, such as magnetic tunnel junctions and tunnel diodes, can be operated in a stochastic regime and incorporated into a scalable neuromorphic architecture that can impact a number of probabilistic computing applications, such as Monte Carlo simulations and Bayesian neural networks. Finally, a framework is presented to categorize increasingly advanced hardware‐based probabilistic computing technologies.

Misra, Shashank↗
DOE OSTI2025

Scaling microstructural processes in the sintering of ionic ceramics

A multi-scale framework, combining a multiphase field formulation and large deformation mechanics, was developed as a stepping stone to perform the data analytics of the microstructural level kinetics of a sintering solid. Relevant microstructural information from this framework, such as grain, stress, and porosity statistics, was scaled up to describe the macroscopic level sintering kinetics. Here, the developed formulation was applied to describe the electric field assisted sintering of Y 2 O 3 . Microstructural inhomogeneities in a multi-granular solid result in the formation of a field of compressive stress networks, which interleave with low compression and weakly tensile regions, defining a scaffolding for sintering concentration regions to develop. A Poisson effect-induced lateral stress network is also naturally self-induced as a result of the mechanical constraints imposed by the sintering apparatus. For long sintering times, localized shear stresses enhancing mass flow along grain boundaries and internal surfaces develop. Three-sided pores are removed by either vacancy transport to the surrounding pores, or move towards the external surfaces through grain boundary diffusion. Four- and higher order-sided pores stabilize because an equal amount of vacancies are gained and lost through the connecting grain boundaries. Grain dewetting contributes to pore coalescence, suggesting that pore kinetics and grain growth are coupled and should be analyzed in concert. The combined sintering and grain growth kinetics define six regimes of sintering behavior: (1) T, the transient regime; (2) E$_Υ$, the surface energy dominated, early sintering regime, where the grain growth exponent, p = 1, and the stress concentration factor, $f$ ~ $1/\hat{ρ}^{4.6}$; (3) E S , the stress dominated, early sintering regime, where p = 1 and $f$ ~ $1/\hat{ρ}^{4}$; (4) I$_Υ$, the surface energy dominated, intermediate sintering regime, where p = 2 and $f$ ~ $1/\hat{ρ}^{4.6}$; (5) I S , the stress dominated, intermediate sintering regime, where p = 2 and $f$ ~ $1/\hat{ρ}^{4}$; and (6) L, the late sintering regime, where p = 3 and $f$ ~ 1. At the macroscopic level, the rapid densification and suppression of grain growth observed in the electric field assisted sintering process is a consequence of the compounding effects of the underlying stress-, transport-, and interfacial-energy-induced energy minimization kinetics, as predicted by the multi-scale framework.

11 - NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2026

Imaging supermoiré relaxation in helical trilayer graphene

In twisted van der Waals materials, local atomic relaxation can alter the underlying electronic structure. Characterizing lattice reconstruction and its susceptibility to strain is essential for understanding emergent electronic states, especially in multilayers in which interference between moiré lattices yields larger supermoiré patterns whose energy is highly sensitive to local stacking. Here, in this study, we image spatial modulations in the electronic character of helical trilayer graphene, which indicate relaxation into a superstructure of large domains with uniform moiré periodicity. We show that the supermoiré domain size is increased by strain and can be altered in the same device while preserving the local properties within each domain. Finally, we observe a higher conductance at the domain boundaries, consistent with predictions that they host counterpropagating edge modes. Our work provides a real-space visualization of moiré-periodic domains, reveals two independently tunable length scales and demonstrates strain engineering as a route towards designing correlated topological networks at the supermoiré scale.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Massively scalable workflows for quantum chemistry: BigChem and ChemCloud

Electronic structure theory, i.e., quantum chemistry, is the fundamental building block for many problems in computational chemistry. Here we present a new distributed computing framework (BigChem), which allows for an efficient solution of many quantum chemistry problems in parallel. BigChem is designed to be easily composable and leverages industry-standard middleware (e.g., Celery, RabbitMQ, and Redis) for distributed approaches to large scale problems. BigChem can harness any collection of worker nodes, including ones on cloud providers (such as AWS or Azure), local clusters, or supercomputer centers (and any mixture of these). BigChem builds upon MolSSI packages, such as QCEngine to standardize the operation of numerous computational chemistry programs, demonstrated here with Psi4, xtb, geomeTRIC, and TeraChem. BigChem delivers full utilization of compute resources at scale, offers a programable canvas for designing sophisticated quantum chemistry workflows, and is fault tolerant to node failures and network disruptions. We demonstrate linear scalability of BigChem running computational chemistry workloads on up to 125 GPUs. Finally, we present ChemCloud, a web API to BigChem and successor to TeraChem Cloud. ChemCloud delivers scalable and secure access to BigChem over the Internet.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Satellite-based evaluation of AeroCom model bias in biomass burning regions

Global models are widely used to simulate biomass burning aerosol (BBA). Exhaustive evaluations on model representation of aerosol distributions and properties are fundamental to assess health and climate impacts of BBA. Here we conducted a comprehensive comparison of Aerosol Comparisons between Observations and Models (AeroCom) project model simulations with satellite observations. A total of 59 runs by 18 models from three AeroCom Phase-III experiments (i.e., biomass burning emissions, CTRL16, and CTRL19) and 14 satellite products of aerosols were used in the study. Aerosol optical depth (AOD) at 550 nm was investigated during the fire season over three key fire regions reflecting different fire dynamics (i.e., deforestation-dominated Amazon, Southern Hemisphere Africa where savannas are the key source of emissions, and boreal forest burning in boreal North America). The 14 satellite products were first evaluated against AErosol RObotic NETwork (AERONET) observations, with large uncertainties found. But these uncertainties had small impacts on the model evaluation that was dominated by modeling bias. Through a comparison with Polarization and Directionality of the Earth’s Reflectances measurements with the Generalized Retrieval of Aerosol and Surface Properties algorithm (POLDER-GRASP), we found that the modeled AOD values were biased by –93 % to 152 %, with most models showing significant underestimations even for the state-of-the-art aerosol modeling techniques (i.e., CTRL19). By scaling up BBA emissions, the negative biases in modeled AOD were significantly mitigated, although it yielded only negligible improvements in the correlation between models and observations, and the spatial and temporal variations in AOD biases did not change much. For models in CTRL16 and CTRL19, the large diversity in modeled AOD was in almost equal measures caused by diversity in emissions, lifetime, and the mass extinction coefficient (MEC). We found that in the AeroCom ensemble, BBA lifetime correlated significantly with particle deposition (as expected) and in turn correlated strongly with precipitation. Additional analysis based on Cloud-Aerosol LIdar with Orthogonal Polarization (CALIOP) aerosol profiles suggested that the altitude of the aerosol layer in the current models was generally too low, which also contributed to the bias in modeled lifetime. Modeled MECs exhibited significant correlations with the Ångström exponent (AE, an indicator of particle size). Comparisons with the POLDER-GRASP-observed AE suggested that the models tended to overestimate the AE (underestimated particle size), indicating a possible underestimation of MECs in models. The hygroscopic growth in most models generally agreed with observations and might not explain the overall underestimation of modeled AOD. Our results imply that current global models contain biases in important aerosol processes for BBA (e.g., emissions, removal, and optical properties) that remain to be addressed in future research.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
NASA NTRS2022

Satellite-Based Evaluation of AeroCom Model Bias in Biomass Burning Regions

Global models are widely used to simulate biomass burning aerosol (BBA). Exhaustive evaluations on model representation of aerosol distributions and properties are fundamental to assess health and climate impacts of BBA. Here we conducted a comprehensive comparison of Aerosol Comparisons between Observations and Models (AeroCom) project model simulations with satellite observations. A total of 59 runs by 18 models from three AeroCom Phase-III experiments (i.e., biomass burning emissions, CTRL16, and CTRL19) and 14 satellite products of aerosols were used in the study. Aerosol optical depth (AOD) at 550 nm was investigated during the fire season over three key fire regions reflecting different fire dynamics (i.e., deforestation-dominated Amazon, Southern Hemisphere Africa where savannas are the key source of emissions, and boreal forest burning in boreal North America). The 14 satellite products were first evaluated against AErosol RObotic NETwork (AERONET) observations, with large uncertainties found. But these uncertainties had small impacts on the model evaluation that was dominated by modeling bias. Through a comparison with Polarization and Directionality of the Earth’s Reflectances measurements with the Generalized Retrieval of Aerosol and Surface Properties algorithm (POLDER-GRASP), we found that the modeled AOD values were biased by −93 % to 152 %, with most models showing significant underestimations even for the state-of-the-art aerosol modeling techniques (i.e., CTRL19). By scaling up BBA emissions, the negative biases in modeled AOD were significantly mitigated, although it yielded only negligible improvements in the correlation between models and observations, and the spatial and temporal variations in AOD biases did not change much. For models in CTRL16 and CTRL19, the large diversity in modeled AOD was in almost equal measures caused by diversity in emissions, lifetime, and the mass extinction coefficient (MEC). We found that in the AeroCom ensemble, BBA lifetime correlated significantly with particle deposition (as expected) and in turn correlated strongly with precipitation. Additional analysis based on Cloud-Aerosol LIdar with Orthogonal Polarization (CALIOP) aerosol profiles suggested that the altitude of the aerosol layer in the current models was generally too low, which also contributed to the bias in modeled lifetime. Modeled MECs exhibited significant correlations with the Ångström exponent (AE, an indicator of particle size). Comparisons with the POLDER-GRASP-observed AE suggested that the models tended to overestimate the AE (underestimated particle size), indicating a possible underestimation of MECs in models. The hygroscopic growth in most models generally agreed with observations and might not explain the overall underestimation of modeled AOD. Our results imply that current global models contain biases in important aerosol processes for BBA (e.g., emissions, removal, and optical properties) that remain to be addressed in future research.

biomass burning aerosols↗
DOE OSTI2022

Toward designing effective exascale scientific computing workflows: experiences and best practices

Many fields within scientific computing have embraced advances in big-data analysis and machine learning, which often requires the deployment of large, distributed and complicated workflows that may combine training neural networks, performing simulations, running inference, and performing database queries and data analysis in asynchronous, parallel and pipelined execution frameworks. Such a shift has brought into focus the need for scalable, efficient workflow management solutions with reproducibility, error and provenance handling, traceability, and checkpoint-restart capabilities, among other needs. Here, we discuss challenges and best-practices for deploying exascale-generation computational science workflows on resources at the Oak Ridge Leadership Computing Facility (OLCF). We present our experiences with large-scale deployment of distributed workflows on the Summit supercomputer, including for bioinformatics and computational biophysics, materials science, and deep learning model optimization. We also present problems and solutions created by working within a Python-centric software base on traditional HPC systems, and discuss steps that will be required before the convergence of HPC, AI, and data science can be fully realized. Our results point to a wealth of exciting new possibilities for harnessing this convergence to tackle new scientific challenges.

Coletti, Mark↗
DOE OSTI2022

Boundary characterization using 3D mapping of geometrically necessary dislocations in AM Ta microstructure

Additive manufacturing (AM) of high strength metallic materials produces microstructures and defects that deviate substantially from those created through conventional manufacturing methods. It has recently been demonstrated that with application of extreme energy densities, a high-temperature refractory metal, tantalum (Ta), can be printed to a fully dense state with exceptionally low porosity. Here we employ a combination of three-dimensional electron backscattered diffraction (EBSD) TriBeam technology and crystallographic geometrically necessary dislocation (GND) theory to characterize the microstructure and defect boundaries of this unusual AM product. Two-dimensional (2D) EBSD and these three-dimensional (3D) measurements indicate that the microstructure of AM Ta is highly oriented <111> along the build direction, yet at the same time contains large crystallographic orientation gradients that span mm’s across the build. Crystallographic GND density analysis of the 3D microstructure reveals that highly misoriented subboundaries exist within this strongly textured microstructure that have large dislocation densities of 1×10 16 m –2 , just as large as those comprising the high-angle grain boundaries (HAGBs) in the same material. The 3D crystallographic GND density mapping reveals these subboundaries are part of a complex, finely spaced network that extends throughout the entire microstructure. Here, the orientation of these boundaries can be related to the scan strategy used during printing. TEM measurements corroborate an extremely high dislocation density at the microscale and indicate a cell-like dislocation network structure existing in the AM Ta at the sub-μm scale.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Topological structure of complex predictions

Abstract Current complex prediction models are the result of fitting deep neural networks, graph convolutional networks or transducers to a set of training data. A key challenge with these models is that they are highly parameterized, which makes describing and interpreting the prediction strategies difficult. We use topological data analysis to transform these complex prediction models into a simplified topological view of the prediction landscape. The result is a map of the predictions that enables inspection of the model results with more specificity than dimensionality-reduction methods such as tSNE and UMAP. The methods scale up to large datasets across different domains. We present a case study of a transformer-based model previously designed to predict expression levels of a piece of DNA in thousands of genomic tracks. When the model is used to study mutations in the BRCA1 gene, our topological analysis shows that it is sensitive to the location of a mutation and the exon structure of BRCA1 in ways that cannot be found with tools based on dimensionality reduction. Moreover, the topological framework offers multiple ways to inspect results, including an error estimate that is more accurate than model uncertainty. Further studies show how these ideas produce useful results in graph-based learning and image classification.

Computer Science↗
DOE OSTI2024

FPGA-Accelerated Range-Limited Molecular Dynamics

Long timescale Molecular Dynamics (MD) simulation of small molecules is crucial in drug design and basic science. To accelerate a small data set that is executed for a large number of iterations, high-efficiency is required. Recent work in this domain has demonstrated that among COTS devices only FPGA-centric clusters can scale beyond a few processors. The problem addressed here is that, as the number of on-chip processors has increased from fewer than 10 into the hundreds, previous intra-chip routing solutions are no longer viable. We find, however, that through various design innovations, high efficiency can be maintained. These include replacing the previous broadcast networks with ring-routing and then augmenting the rings with out-of-order and caching mechanisms. Others are adding a level of hierarchical filtering and memory recycling. Two novel optimized architectures emerge, together with a number of variations. These are validated, analyzed, and evaluated. We find that in the domain of interest speed-ups over GPUs are achieved. Finally, the potential impact is that this system promises to be the basis for scalable long timescale MD with commodity clusters.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Meta-Learning Enhanced Physics-Informed Graph Attention Convolutional Network for Distribution Power System State Estimation

Promptly perceiving distribution system states is challenged by frequent topology changes and uncertain power injections. To address these issues, a Meta-learning enhanced physics-informed graph attention convolutional network (Meta-PIGACN) model is proposed to handle topological variability in distribution system state estimation (DSSE). Specifically, physics information is integrated into the graph convolutional network, enabling a physics-informed edge-weighting process that incorporates physical information to control the aggregation of neighboring nodes. Besides, the graph attention mechanism automatically adjusts the importance of different neighboring nodes, allowing the capture and preservation of inherent system features across varying topologies, thereby improving state estimation accuracy. Furthermore, meta-learning is proposed to acquire empirical knowledge across multiple topologies so that the model can rapidly adapt to new configurations through iterative gradient descent updates even in large-scale systems. In conclusion, the simulation results based on the 33/118/1746-node distribution systems show the high accuracy and efficiency of the proposed model.

24 POWER TRANSMISSION AND DISTRIBUTION↗