Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “representation”

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 91 records · Page 5
All sourcesNASA NTRSDOE OSTI
DOE OSTI2020

Anti-symmetric representation of the extended magnetohydrodynamic equations

We introduce the anti-symmetric representation of the extended magnetohydrodynamic (MHD) equations. In this representation, the use of the anti-symmetric flux operator (∇·ν+ν·∇) results in conservation theorems with discrete analogs. Inherently robust numerical applications are achieved with little effort, and conservation to machine precision is possible with simple numerical schemes. Starting from the two-fluid equations, we construct a single-fluid MHD model based on generalized center-of-mass variables for the mass (√ρ), momentum (√ρν), and pressure (√ p ). This model is shown to possess identical conservation properties to the two-fluid system, with the only restriction being the use of a single temperature. Common approximations to the Braginskii heat fluxes and to the gyroviscous stress tensor are cast into our representation for convenience. The discrete conservation properties are verified using the classic Orszag–Tang vortex problem. In addition to the favorable mass, momentum, and energy conservation properties, the time reversibility of the simulations is demonstrated.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2023

N -representability violations in truncated equation-of-motion coupled-cluster methods

One-electron reduced density matrices (1RDMs) from equation-of-motion (EOM) coupled-cluster with single and double excitations (CCSD) calculations are analyzed to assess their N-representability (i.e., whether they are derivable from a physical N-electron state). We identify EOM-CCSD stationary states whose 1RDMs violate either ensemble-state N-representability conditions or pure-state conditions known as generalized Pauli constraints. As such, these 1RDMs do not correspond to any physical N-electron state. Unphysical states are also encountered in the course of time-dependent EOM-CC simulations; when an external field drives transitions between a pair of stationary states with pure-state N-representable 1RDMs, the 1RDM of the time-dependent state can violate ensemble-state conditions. Furthermore, these observations point to potential challenges in interpreting the results of time-dependent EOM-CCSD simulations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

Extended Applications of Subgrid Representation in the 2D/1D Method

Recent efforts in MPACT have focused on improving the performance of the 2D/1D subplane implementation to help target computational performance goals. Here, we build on previous efforts that targeted the use of subgrid treatments to improve the accuracy of control rod representation, presenting three additional applications of subgrid treatments with the goal of reducing the computational burden of simulations. These subgrid applications include treatment of spacer grids, thermal feedback, and axial reflector material representation. With these approaches, a single method of characteristics (MOC) plane can contain several different materials axially that are represented explicitly via subgrids on the coarse mesh finite difference (CMFD) mesh but are axially homogenized on the MOC mesh. This allows for a substantial reduction in the number of MOC planes needed in the calculation through the introduction of an approximate treatment, particularly with regard to the self-shielded cross sections and MOC-informed radial current coupling coefficients in CMFD. Several test problems ranging from single rod to quarter core are used to assess the solution accuracy and performance of these various subgrid representations. Overall, the accuracy of the approximations seems very reasonable, with extremely small differences in eigenvalue observed and maximum pin power errors in the 0.5% to 1.0% range. Several cases show substantial value in the compromise between accuracy and computational performance. Others highlight the new computational hurdles that future research will aim to resolve.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2026

Particle trajectory representation learning with masked point modeling

Liquid argon time projection chambers (LArTPCs) offer millimeter-scale 3D images of particle trajectories, enabling precision studies of neutrino oscillation, detection of supernova and solar neutrinos, searches for exotic dark matter, and proton decay. Current approaches utilize supervised machine learning models, requiring extensive simulations of particle physics and detector response that can introduce bias. Self-supervised learning (SSL), a machine learning approach that learns useful representations of unlabeled data from the data itself, has significantly advanced how large datasets are utilized for representation learning; however, its potential for applications to sensory data in high precision particle physics experiments remains largely unexplored. We introduce the Point-based liquid argon masked autoencoder (PoLAr-MAE), a self-supervised framework that learns physically meaningful representations directly from unlabeled LArTPC images. PoLAr-MAE achieves remarkable data efficiency for a point-level segmentation task, outperforming fully supervised methods in low data regimes. Linear classifiers on model outputs demonstrate robust performance across multiple downstream tasks. Our results position sensor-level SSL as a practical foundation model strategy for LArTPCs.

Young, Samuel [Stanford Univ., CA (United States)]↗
DOE OSTI2023

Multimodal representation learning for predicting molecule–disease relations

Motivation: Predicting molecule–disease indications and side effects is important for drug development and pharmacovigilance. Comprehensively mining molecule–molecule, molecule–disease and disease–disease semantic dependencies can potentially improve prediction performance. Methods: We introduce a Multi-Modal REpresentation Mapping Approach to Predicting molecular-disease relations (M2REMAP) by incorporating clinical semantics learned from electronic health records (EHR) of 12.6 million patients. Specifically, M2REMAP first learns a multimodal molecule representation that synthesizes chemical property and clinical semantic information by mapping molecule chemicals via a deep neural network onto the clinical semantic embedding space shared by drugs, diseases and other common clinical concepts. To infer molecule–disease relations, M2REMAP combines multimodal molecule representation and disease semantic embedding to jointly infer indications and side effects. Results: We extensively evaluate M2REMAP on molecule indications, side effects and interactions. Results show that incorporating EHR embeddings improves performance significantly, for example, attaining an improvement over the baseline models by 23.6% in PRC-AUC on indications and 23.9% on side effects. Further, M2REMAP overcomes the limitation of existing methods and effectively predicts drugs for novel diseases and emerging pathogens. Availability and implementation: The code is available at https://github.com/celehs/M2REMAP, and prediction results are provided at https://shiny.parse-health.org/drugs-diseases-dev/.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2022

Pairwise connected tensor network representation of path integrals

It has been recently shown how the tensorial nature of real-time path integrals (PIs) involving the Feynman-Vernon influence functional can be utilized with matrix product states, taking advantage of the finite length of the bath-induced memory. Tensor networks (TNs) promise to provide a unified language to express the structure of a PI. A generalized TN specifically incorporating the pairwise interaction structure of the influence functional and its invariance with respect to the average forward-backward position or the sojourn value in the form of the blip representation is derived and implemented. This pairwise connected TNPI (PC-TNPI) is illustrated through applications to typical spin-boson problems and explorations of the differences caused by the exact form of the spectral density. The storage and performance scalings are reported, showing the compactness of the representation and the efficiency of the contraction process. Finally, taking advantage of the compressed representation, the viability of using PC-TNPI for simulating multistate problems is demonstrated. The PC-TNPI structure can be shown to yield other TN algorithms currently in use. Consequently, it should be possible to use it as a starting point for deriving other optimized procedures.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Compressive Neural Representations of Volumetric Scalar Fields

Here, we present an approach for compressing volumetric scalar fields using implicit neural representations. Our approach represents a scalar field as a learned function, wherein a neural network maps a point in the domain to an output scalar value. By setting the number of weights of the neural network to be smaller than the input size, we achieve compressed representations of scalar fields, thus framing compression as a type of function approximation. Combined with carefully quantizing network weights, we show that this approach yields highly compact representations that outperform state-of-the-art volume compression approaches. The conceptual simplicity of our approach enables a number of benefits, such as support for time-varying scalar fields, optimizing to preserve spatial gradients, and random-access field evaluation. We study the impact of network design choices on compression performance, highlighting how simple network architectures are effective for a broad range of volumes.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Session Introduction: Graph Representations and Algorithms in Biomedicine

Connectivity is a fundamental property of biological systems: on the cellular level, proteins interact with each other to form protein-protein interaction networks (PPIs); on the organism level, neurons are arranged in a network; and on a community-level, species can have complex relationships with one another that drive the development and balance of an ecosystem. Graphs, representations of systems consisting of entities as vertices and their connections as edges, are a useful structure to characterize many such systems. Such models can be used to understand biological systems that naturally have a network structure, including PPIs, biological neurons, and ecosystems. In today’s information age, graph representations and algorithms (often in combination with machine learning techniques) are used to organize massive amounts of related data, much of which may be heterogeneous or unstructured, and identify patterns that represent novel biological insights. PSB’s 2023 session “Graph Representations and algorithms in Biomedicine,” encompasses modern developments in graph theory and its applications to various fields of biomedicine. This session includes a wide range of research - knowledge graphs built from text-mined health data, heterogeneous networks using multi-omic databases, and graphs refined to represent uncertainty or improve memory usage.

Chrisman, Brianna S.↗
DOE OSTI2021

Effects of raster terrain representation on GIS shortest path analysis

Spatial analysis extracts meaning and insights from spatially referenced data, where the results are highly dependent on the quality of the data used and the manipulations on the data when preparing it for analysis. Users should understand the impacts that data representations may have on their results in order to prevent distortions in their outcomes. We study the consequences of two common data preparations when locating a linear feature performing shortest path analysis on raster terrain data: 1) the connectivity of the network generated by connecting raster cells to their neighbors, and 2) the range of the attribute scale for assigning costs. Such analysis is commonly used to locate transmission lines, where the results could have major implications on project cost and its environmental impact. Experiments in solving biobjective shortest paths show that results are highly dependent on the parameters of the data representations, with exceedingly variable results based on the choices made in reclassifying attributes and generating networks from the raster. Based on these outcomes, we outline recommendations for ensuring geographic information system (GIS) data representations maintain analysis results that are accurate and unbiased.

Medrano, F. Antonio (ORCID:000000025913632X)↗
DOE OSTI2023

Understanding the Inner-Workings of Language Models Through Representation Dissimilarity

We use model stitching to understand the internal representations of language models. Similar to vision models, we find that "more is better," and representations learned with more data and larger width can improve the performance of weaker models via stitching. We likewise find that certain architecture choices, using GeLU vs SoLU activation functions, influence the quality of learned representations. Finally, model stitching (as opposed to other model diagnostic methods, like mode connectivity) can localize the different generalization strategies of text classifiers under domain shift to certain hidden layers.

Brown, Davis R.↗
DOE OSTI2020

Joint Analysis of Program Data Representations using Machine Learning for Improved Software Assurance and Development Capabilities

We explore the use of multiple deep learning models for detecting flaws in software programs. Current, standard approaches for flaw detection rely on a single representation of a software program (e.g., source code or a program binary). We illustrate that, by using techniques from multimodal deep learning, we can simultaneously leverage multiple representations of software programs to improve flaw detection over single representation analyses. Specifically, we adapt three deep learning models from the multimodal learning literature for use in flaw detection and demonstrate how these models outperform traditional deep learning models. We present results on detecting software flaws using the Juliet Test Suite and Linux Kernel.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Learned implicit representations of aerosol chemistry and physics for enhancing the predictability of water cycle extreme events

Focal Area(s): Focal area 2: Predictive modeling through the use of AI-derived model components; and Focal area 3: physics-guided AI. Science Challenge: Comprehensive models of many geophysical processes require a large number of state variables, disqualifying them for inclusion in Earth System Models (ESMs) for the foreseeable future. This white paper proposes the paradigm shift from the use of state variables representing explicit physical quantities to the use of machine-learning to create and integrate a compact, implicit representation of a physical system with a smaller number of state variables. As an initial application, we propose to create parsimonious machine-learned surrogate models of gas- and aerosol-phase chemistry and physics with the purpose of improving the accuracy of the representation of cloud formation and cloud-aerosol interaction in the E3SM model. This type of improved representation of cloud microphysics is critical for enhancing the predictability of water cycle extremes.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Understanding and improving model representation of aerosol optical properties for a Chinese haze event measured during KORUS-AQ

KORUS-AQ was an international cooperative air quality field study in South Korea that measured local and remote sources of air pollution affecting the Korean Peninsula during May–June 2016. Some of the largest aerosol mass concentrations were measured during a Chinese haze transport event (24 May). Air quality forecasts using the WRF-Chem model with aerosol optical depth (AOD) data assimilation captured AOD during this pollution episode but overpredicted surface particulate matter concentrations in South Korea, especially PM 2.5 , often by a factor of 2 or larger. Analysis revealed multiple sources of model deficiency related to the calculation of optical properties from aerosol mass that explain these discrepancies. Using in situ observations of aerosol size and composition as inputs to the optical properties calculations showed that using a low-resolution size bin representation (four bins) underestimates the efficiency with which aerosols scatter and absorb light (mass extinction efficiency). Besides using finer-resolution size bins (8–16 bins), it was also necessary to increase the refractive indices and hygroscopicity of select aerosol species within the range of values reported in the literature to achieve better consistency with measured values of the mass extinction efficiency (6.7 m 2 g –1 observed average) and light-scattering enhancement factor (ƒ(RH)) due to aerosol hygroscopic growth (2.2 observed average). Furthermore, an evaluation of the optical properties obtained using modeled aerosol properties revealed the inability of sectional and modal aerosol representations in WRF-Chem to properly reproduce the observed size distribution, with the models displaying a much wider accumulation mode. Other model deficiencies included an underestimate of organic aerosol density (1.0 g cm –3 in the model vs. observed average of 1.5 g cm –3 ) and an overprediction of the fractional contribution of submicron inorganic aerosols other than sulfate, ammonium, nitrate, chloride, and sodium corresponding to mostly dust (17 %–28 % modeled vs. 12 % estimated from observations). These results illustrate the complexity of achieving an accurate model representation of optical properties and provide potential solutions that are relevant to multiple disciplines and applications such as air quality forecasts, health impact assessments, climate projections, solar power forecasts, and aerosol data assimilation.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Breakups are complicated: an efficient representation of collisional breakup in the superdroplet method

Abstract. A key constraint of particle-based methods for modeling cloud microphysics is the conservation of total particle number, which is required for computational tractability. The process of collisional breakup poses a particular challenge to this framework, as breakup events often produce many droplet fragments of varying sizes, which would require creating new particles in the system. This work introduces a representation of collisional breakup in the so-called “superdroplet” method which conserves the total number of superdroplets in the system. This representation extends an existing stochastic collisional-coalescence scheme and samples from a fragment size distribution in an additional Monte Carlo step. This method is demonstrated in a set of idealized box model and single-column warm-rain simulations. We further discuss the effects of the breakup dynamic and fragment size distribution on the particle size distribution, hydrometeor population, and microphysical process rates. Box model experiments serve to characterize the impacts of properties such as coalescence efficiency and fragmentation function on the relative roles of collisional breakup and coalescence. The results demonstrate that this representation of collisional breakup can produce a stationary particle size distribution, in which breakup and coalescence rates are approximately equal, and that it recovers expected behavior such as a reduction in precipitate-sized particles in the column model. The breakup algorithm presented here contributes to an open-source pythonic implementation of the superdroplet method, PySDM, which will facilitate future research using particle-based microphysics.

de Jong, Emily (ORCID:0000000253104554)↗
DOE OSTI2025

Assessing the climate impact of an improved volcanic sulfate aerosol representation in E3SM

Accurately simulating historical surface temperature variations is essential for evaluating climate models, yet many struggle to reproduce the mid-20th-century temperature trends associated with significant volcanic eruptions. This study examines the impact of volcanic sulfate aerosol representation on these biases using the Energy Exascale Earth System Model (E3SM). The standard CMIP6 protocol prescribes volcanic forcing through radiative perturbations, omitting volcanic aerosol–cloud interactions (VACIs). Here, we implement an emission-based approach with an updated volcanic eruption inventory that directly incorporates volcanic sulfur dioxide (SO 2 ) emissions, enabling a more process-based representation of volcanic forcing. This approach leads to improved surface temperature variability and a modest reduction in cold biases between 1940 and 1980 compared to the CMIP6 setup. Additionally, we assess cloud property responses to a more realistic volcanic sulfate aerosol representation, which weakens cloud-induced cooling during periods of lower volcanic activity. However, despite these refinements, a significant temperature cold bias remains, indicating that further improvements in atmospheric chemistry, aerosol microphysics, cloud processes, and model parameterizations are needed to fully resolve this issue in E3SM.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Anomaly Detection in Sequential Health Care Data using Higher-Order Network Representation

The ever-increasing use of information technology (IT) in health care presents new challenges to patient care. Information errors arising from the use of Health IT and their implications on care delivery and patient outcomes have been widely reported. Information errors can lead to changes in clinical decisions, care processes, and care outcomes, among others. An altered care process, for example, could interrupt the sequence of care, which could lead to changes in care decisions and/or changes in care outcomes. We define interruptions in the care process as anomalies in the care sequence. In this paper, we present a new approach based on the higher-order network (HON) representation to detect anomalies in sequential health care data using electronic health records. The results show that there are higher-order dependencies in health care data; and that the use of HON representation is more effective than the first-order network representation for detecting anomalies in sequential health care data.

Niu, Haoran↗
DOE OSTI2021

Geothermal Representation in Power System Models

Power system models generally fail to capture the range of characteristics geothermal resources provide and the value they potentially contribute to decarbonization and reliability of future electricity grids as firm, dispatchable, non-combustion power resources. This study reviews the results of power system modeling efforts to investigate geothermal deployment potential in the United States, including the U.S. DOE GeoVision analysis and ongoing modeling and analysis efforts to support planning and development of future grids with 100% renewable energy in California. Several themes are identified that could be implemented immediately to improve the accuracy of geothermal representation in power system models: consistency of model inputs, modeling of baseload and dispatchable geothermal resources, accurate valuation of grid services, improved representation of capacity factor, use of contemporary LCOE estimates, improved understanding of the evolution of geothermal value, and use of accurate resource potential constraints. Many of the models reviewed produced significantly different amounts of geothermal resource selection - even when modeling the same region and time period. This highlights the variability of inputs and assumptions among models, so creating a consistent set of geothermal inputs is a first step toward more accurate representation of geothermal in models. Research opportunities are identified that could help improve geothermal data inputs in modeling efforts, including analyses of historical data, sensitivity to model inputs, and comparative value of geothermal generators as baseload or dispatchable resources. Outcomes of such research can inform the geothermal community about how best to guide geothermal development toward wider deployment in support of future electricity grids through improved understanding of the evolution of geothermal value over time and the characteristics that contribute to that value.

capacity expansion models↗
DOE OSTI2023

NeRVI: Compressive neural representation of visualization images for communicating volume visualization results

We present NeRVI, a new deep-learning approach that compresses a large collection of visualization images generated from time-varying data for communicating volume visualization results. Based on an image-based implicit neural representation, our approach represents tens of thousands of high-resolution rendering images parametrized by different parameters via a hybrid model of multilayer perceptrons and convolutional neural networks. Here, our model predicts images and corresponding masks, and the masks are utilized for loss computation and network training to capture fine structural details and small components. In conjunction with model quantization and weight encoding, NeRVI yields highly compact compressive neural representations while preserving the image fidelity well. We demonstrate the effectiveness of NeRVI with isosurface rendering and direct volume rendering images generated from multiple data sets and compare NeRVI with other state-of-the-art deep learning-based (InSituNet, SIREN, NeRF, and NeRV) methods. Quantitative and qualitative results show that NeRVI provides an alternative solution that augments domain scientists' ability to manage, represent, and communicate scientific visualization output.

97 MATHEMATICS AND COMPUTING↗