Search NASA⌕ Search

SEARCH · Search NASA

Results for “Network Models”

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 1,531 records · Page 85

Scaling Laws of Graph Neural Networks for Atomistic Materials Modeling

Atomistic materials modeling is a critical task with wide-ranging applications, from drug discovery to materials science, where accurate predictions of the target material property can lead to significant advancements in scientific discovery. Graph Neural Networks (GNNs) represent the state-of-the-art approach for modeling atomistic material data thanks to their capacity to capture complex relational structures. While machine learning performance has historically improved with larger models and datasets, GNNs for atomistic materials modeling remain relatively small compared to large language models (LLMs), which leverage billions of parameters and terabyte-scale datasets to achieve remarkable performance in their respective domains. To address this gap, we explore the scaling limits of GNNs for atomistic materials modeling by developing a foundational model with billions of parameters, trained on extensive datasets in terabytescale. Our approach incorporates techniques from LLM libraries to efficiently manage large-scale data and models, enabling both effective training and deployment of these large-scale GNN models. This work addresses three fundamental questions in scaling GNNs: the potential for scaling GNN model architectures, the effect of dataset size on model accuracy, and the applicability of LLM-inspired techniques to GNN architectures. Specifically, the outcomes of this study include (1) insights into the scaling laws for GNNs, highlighting the relationship between model size, dataset volume, and accuracy, (2) a foundational GNN model optimized for atomistic materials modeling, and (3) a GNN codebase enhanced with advanced LLM-based training techniques. Our findings lay the groundwork for large-scale GNNs with billions of parameters and terabyte-scale datasets, establishing a scalable pathway for future advancements in atomistic materials modeling.

Li, Chaojian [ORNL] (ORCID:0000000340309777)↗

Comment on"Sensitivity of the active fracture model parameter to fracture network orientation and injection scenarios" by Basagaoglu et al. (2009)

Basagaoglu et al. (2009) present a study on detailed unsaturated flow behavior in two-dimensional fracture networks using numerical experiments (simulations) based on the lattice-Boltzmann method. Their results are valuable for improving our understanding of unsaturated flow processes and evaluating the active fracture model (AFM) that was developed for capturing large-scale preferential flow in fractured rocks (Liu et al., 1998; 2003). As indicated in Basagaoglu et al. (2009), a previous study was conducted to evaluate the AFM with numerical experiments (Seol et al., 2003). However, the methodology used in that study and the corresponding conclusions are highly questioned for the following two reasons. First, the evaluation relies on a condition that simulated water flow processes in a fracture network are adequately represented with a continuum approach, because they draw their conclusions by comparing simulation results with those obtained from a dual-continuum model based on the AFM. No effort was made by Seol et al. (2003) to justify the validity of the continuum approach for their specific fracture network that includes a small number of fractures only. (The analyses of Basagaoglu et al. (2009) do not need the similar condition.) Second, Seol et al. (2003) use numerical dispersion to represent the matrix diffusion process. This treatment is not valid simply because numerical dispersion results from numerical errors and is not a physical process.

54 ENVIRONMENTAL SCIENCES↗

NCC Simulation Model: Simulating the operations of the network control center, phase 2

The simulation of the network control center (NCC) is in the second phase of development. This phase seeks to further develop the work performed in phase one. Phase one concentrated on the computer systems and interconnecting network. The focus of phase two will be the implementation of the network message dialogues and the resources controlled by the NCC. These resources are requested, initiated, monitored and analyzed via network messages. In the NCC network messages are presented in the form of packets that are routed across the network. These packets are generated, encoded, decoded and processed by the network host processors that generate and service the message traffic on the network that connects these hosts. As a result, the message traffic is used to characterize the work done by the NCC and the connected network. Phase one of the model development represented the NCC as a network of bi-directional single server queues and message generating sources. The generators represented the external segment processors. The served based queues represented the host processors. The NCC model consists of the internal and external processors which generate message traffic on the network that links these hosts. To fully realize the objective of phase two it is necessary to identify and model the processes in each internal processor. These processes live in the operating system of the internal host computers and handle tasks such as high speed message exchanging, ISN and NFE interface, event monitoring, network monitoring, and message logging. Inter process communication is achieved through the operating system facilities. The overall performance of the host is determined by its ability to service messages generated by both internal and external processors.

Benjamin, Norman M.↗

Accurate and rapid acoustic damage characterization in complex structures using sparse sensor networks and deep learning models

Damage diagnosis in critical components is essential for ensuring the safety and reliability of operations across industries, spanning manufacturing, aerospace, and energy. Traditional acoustic nondestructive testing methods primarily focus on detecting defects through the direct scattering of single-mode incident waves from the damage, which limit their applicability to simple structures and small inspection areas. Our earlier research demonstrated that machine learning algorithms combined with sparse sensor networks can identify critical defect signatures even from multiply scattered, multi-mode acoustic signals, indicating the potential for improved defect inspection in complex, real-world structures. In this work, we demonstrate the successful implementation of this approach in a fixed sensor configuration to rapidly and accurately detect simulated defects in a geometrically complex, real-world structure, a brake rotor hub. Three different types of defects were physically simulated on the surface of the hub, and the collected data were used to train an autoencoder-based deep learning model. Two models were tested, one using single measurements and the other using multiple measurements taking advantage of the spatial distribution of the sensor network. After training, the multi-measurement model achieved 100 % accuracy in identifying, classifying, and locating unseen, unique damages. This work illustrates the potential of the proposed method for a wide range of industrial applications.

36 MATERIALS SCIENCE↗

Quantification of neural networks uncertainties with applications to SAFARI-1 axial neutron flux profiles

Deep Neural Networks (DNNs) have been widely used as a data-driven modelling tool in nuclear engineering. However, as a Machine Learning model, Artificial Neural Network (ANN) predictions are subjected to uncertainties originating from the noise in training data, incomplete coverage of the domain, and imperfect neural network architectures. In this work, we target at quantifying the prediction/approximation uncertainties of ANNs using Monte Carlo Dropout (MCD), as well as Bayesian Neural Networks (BNNs) which are solved by variational inference. With a demonstration problem in which neural networks are used to predict the assembly axial neutron flux profiles, the results have shown that the three different neural network models (regular DNNs, DNNs solved with MCD and BNNs) can produce results that agree very well among each other and with the measurement data, on cycles that are not used in the training process. Besides the excellent generalization capability, the uncertainty bands produced by MCD and BNN agree very well, and in general, they can fully envelop the noisy measurement data points. (authors)

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Electron percolation in realistic models of carbon nanotube networks

The influence of penetrable and curved carbon nanotubes (CNT) on the charge percolation in three-dimensional disordered CNT networks have been studied with Monte-Carlo simulations. By considering carbon nanotubes as solid objects but where the overlap between their electron cloud can be controlled, we observed that the structural characteristics of networks containing lower aspect ratio CNT are highly sensitive to the degree of penetration between crossed nanotubes. Following our efficient strategy to displace CNT to different positions to create more realistic statistical models, we conclude that the connectivity between objects increases with the hard-core/soft-shell radii ratio. In contrast, the presence of curved CNT in the random networks leads to an increasing percolation threshold and to a decreasing electrical conductivity at saturation. The waviness of CNT decreases the effective distance between the nanotube extremities, hence reducing their connectivity and degrading their electrical properties. We present the results of our simulation in terms of thickness of the CNT network from which simple structural parameters such as the volume fraction or the carbon nanotube density can be accurately evaluated with our more realistic models.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Simulated evolution of fractures and fracture networks subject to thermal cooling: A coupled discrete element and heat conduction model

Advancement of EGS requires improved prediction of fracture development and growth during reservoir stimulation and long-term operation. This, in turn, requires better understanding of the dynamics of the strongly coupled thermo-hydro-mechanical (THM) processes within fractured rocks. We have developed a physically based rock deformation and fracture propagation simulator by using a quasi-static discrete element model (DEM) to model mechanical rock deformation and fracture propagation induced by thermal stress and fluid pressure changes. We also developed a network model to simulate fluid flow and heat transport in both fractures and porous rock. In this paper, we describe results of simulations in which the DEM model and network flow & heat transport model are coupled together to provide realistic simulation of the changes of apertures and permeability of fractures and fracture networks induced by thermal cooling and fluid pressure changes within fractures. Various processes, such as Stokes flow in low velocity pores, convection-dominated heat transport in fractures, heat exchange between fluid-filled fractures and solid rock, heat conduction through low-permeability matrices and associated mechanical deformations are all incorporated into the coupled model. The effects of confining stresses, developing thermal stress and injection pressure on the permeability evolution of fracture and fracture networks are systematically investigated. Results are summarized in terms of implications for the development and evolution of fracture distribution during hydrofracturing and thermal stimulation for EGS.

15 GEOTHERMAL ENERGY↗

Emergent Subpopulation Behavior Uncovered with a Community Dynamic Metabolic Model of Escherichia coli Diauxic Growth

ABSTRACT Microbes have adapted to greatly variable environments in order to survive both short-term perturbations and permanent changes. A classical and yet still actively studied example of adaptation to dynamic environments is the diauxic shift of Escherichia coli , in which cells grow on glucose until its exhaustion and then transition to using previously secreted acetate. Here we tested different hypotheses concerning the nature of this transition by using dynamic metabolic modeling. To reach this goal, we developed an open source modeling framework integrating dynamic models (ordinary differential equation systems) with structural models (metabolic networks) which can take into account the behavior of multiple subpopulations and smooth flux transitions between time points. We used this framework to model the diauxic shift, first with a single E. coli model whose metabolic state represents the overall population average and then with a community of two subpopulations, each growing exclusively on one carbon source (glucose or acetate). After introduction of an environment-dependent transition function that determined the balance between subpopulations, our model generated predictions that are in strong agreement with published data. Our results thus support recent experimental evidence that diauxie, rather than a coordinated metabolic shift, would be the emergent pattern of individual cells differentiating for optimal growth on different substrates. This work offers a new perspective on the use of dynamic metabolic modeling to investigate population heterogeneity dynamics. The proposed approach can easily be applied to other biological systems composed of metabolically distinct, interconverting subpopulations and could be extended to include single-cell-level stochasticity. IMPORTANCE Escherichia coli diauxie is a fundamental example of metabolic adaptation, a phenomenon that is not yet completely understood. Further insight into this process can be achieved by integrating experimental and computational modeling methods. We present a dynamic metabolic modeling approach that captures diauxie as an emergent property of subpopulation dynamics in E. coli monocultures. Without fine-tuning the parameters of the E. coli core metabolic model, we achieved good agreement with published data. Our results suggest that single-organism metabolic models can only approximate the average metabolic state of a population, therefore offering a new perspective on the use of such modeling approaches. The open source modeling framework that we provide can be applied to model general subpopulation systems in more-complex environments and can be extended to include single-cell-level stochasticity.

59 BASIC BIOLOGICAL SCIENCES↗

Convolutional Neural Networks for Heat Flux Model Validation on NSTX-U

To demonstrate the use of embedded thermocouples in new National Spherical Tokamak eXperiment Upgrade (NSTX-U) graphite plasma-facing components (PFCs), a convolutional neural network (CNN) has been trained using the ANSYS simulations to predict the scrape-off layer (SOL) heat flux width, λ q , given various machine operational parameters and diagnostic data as inputs. Here, the proof-of-concept CNN was trained on the thermocouple data generated by the approximated NSTX-U heat loads applied to real PFC designs in ANSYS. Once trained, the CNN is capable of high precision reconstruction of parameterized heat flux profiles expected in NSTX-U. In addition, to test the system's ability to cope with noise and systematic error, pseudonoise was injected into the simulated data. CNN can accurately predict the incident heat flux despite this noise and error.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Comprehensive Material Characterization and Simultaneous Model Calibration for Improved Computational Simulation Credibility

Computational simulation is increasingly relied upon for high-consequence engineering decisions, and a foundational element to solid mechanics simulations is a credible material model. Our ultimate vision is to interlace material characterization and model calibration in a real-time feedback loop, where the current model calibration results will drive the experiment to load regimes that add the most useful information to reduce parameter uncertainty. The current work investigated one key step to this Interlaced Characterization and Calibration (ICC) paradigm, using a finite load-path tree to incorporate history/path dependency of nonlinear material models into a network of surrogate models that replace computationally-expensive finite-element analyses. Our reference simulation was an elastoplastic material point subject to biaxial deformation with a Hill anisotropic yield criterion. Training data was generated using either a space-filling or adaptive sampling method, and surrogates were built using either Gaussian process or polynomial chaos expansion methods. Surrogate error was evaluated to be on the order of 10 ⁻5 and 10 ⁻3 percent for the space-filling and adaptive sampling training data, respectively. Direct Bayesian inference was performed with the surrogate network and with the reference material point simulator, and results agreed to within 3 significant figures for the mean parameter values, with a reduction in computational cost over 5 orders of magnitude. These results bought down risk regarding the surrogate network and facilitated a successful FY22-24 full LDRD proposal to research and develop the complete ICC paradigm.

36 MATERIALS SCIENCE↗

Thermodynamic Modeling of Complex Solid Solutions in the Lu-H-N System via Graph Neural Network Accelerated Monte Carlo Simulations

Metal hydrides are important across diverse applications, such as hydrogen storage, batteries, gas sensors, nuclear reactions, and high-temperature superconductivity. Previous computational studies of metal hydrides under extreme pressures, e.g., 𝑂⁡(10 2 ) ⁢GPa, usually treat them as stoichiometric compounds without considering interstitial lattice disorder. As pressures become more moderate in the 𝑂⁡(10 0 ) ⁢GPa and below range, hydrogen disorder at interstitial lattice sites becomes prominent, e.g., in the N-doped Lu hydride that was recently claimed superconducting near 1 GPa. Further adding compositional complexity from alloying and/or multielement interstitial occupation makes elucidating pressure- and temperature-dependent observables intractable by first-principles calculations alone. We therefore propose a lattice graph neural-network surrogate modeling approach to predict configuration- and pressure-dependent equation-of-state properties. Their efficiency permits Monte Carlo simulations to calculate Gibbs energies and pressure-dependent phase diagrams, thereby revealing insights into the synthesis conditions required for achieving desired phase equilibria. We demonstrate this concept for the compositionally complex cubic Lu(H,N,Va) 3 system where three constituents (hydrogen, nitrogen and vacancy) have disordered multielement interstitial occupancies and insights into pressure-dependent phase equilibria are critically needed, e.g., N-doping levels can significantly lower dehydrogenation temperatures and provide a new strategy to optimize hydrogen-storage alloys. This work can improve the thermodynamic understanding of the Lu-H-N system and help rational synthesis of N-doped Lu hydrides, but more generally demonstrates an efficient approach to model pressure-dependent thermodynamics of multicomponent solid solutions.

Monte Carlo methods↗

Development of Upscaling Techniques and Construction of Calibrated Models for Fractured Rocks Using Discrete Fracture Network Approaches

The Pahute Mesa Corrective Action Unit on the Nevada National Security Site (NNSS) contains several fractured aquifers that can potentially provide high-permeability pathways for radionuclides to migrate away from underground nuclear testing locations. Although the geometric properties of fractures, such as length and orientation, can be found using geophysical methods and borehole image analyses, the hydraulic property of fractures (which are primarily influenced by apertures) are often unknown or have a high degree of uncertainty attached to their values. The aperture values have a profound impact on groundwater velocity and are critical to understanding transport scales at Pahute Mesa. Fracture aperture values and their distributions also have dominant control over radionuclide transport with potential for long-range correlations or scale-invariant dynamics. Understanding of the small-scale transport variability in fractures could lead to identification of distinct characteristics on larger scales. Furthermore, upscaling techniques need to be developed to adequately predict transport behavior at field scales by capturing the transport trends of smaller scale networks. This report explores these vital modeling needs to understand the role of fractures in Pahute Mesa. Discrete fracture network (DFN) models are the basis for constructing a well-calibrated model to ascertain aperture values of fractures in lava-flow aquifers. These models are also used to develop and compare the usefulness of several random-walk-based models to perform transport upscaling.

54 ENVIRONMENTAL SCIENCES↗

From systems biology to photosynthesis and whole-plant modeling: a conceptual model for integrating multi-scale networks

Network analysis is now a common statistical tool for molecular biologists. Network algorithms are readily used to model gene, protein and metabolic correlations providing insight into pathways driving biological phenomenon. One output from such an analysis is a candidate gene list that can be responsible, in part, for the biological process of interest. The question remains, however, as to whether molecular network analysis can be used to inform process models at higher levels of biological organization. In our previous work, transcriptional networks derived from three plant species were constructed, interrogated for orthology and then correlated to photosynthetic inhibition at elevated temperature. One unique aspect of that study was the link from co-expression networks to net photosynthesis. In this addendum, we propose a conceptual model where traditional network analysis can be linked to whole-plant models thereby informing predictions on key processes such as photosynthesis, nutrient uptake and assimilation, and C partitioning.

59 BASIC BIOLOGICAL SCIENCES↗

Fault detection and diagnosis using neural network approaches

Neural networks can be used to detect and identify abnormalities in real-time process data. Two basic approaches can be used, the first based on training networks using data representing both normal and abnormal modes of process behavior, and the second based on statistical characterization of the normal mode only. Given data representative of process faults, radial basis function networks can effectively identify failures. This approach is often limited by the lack of fault data, but can be facilitated by process simulation. The second approach employs elliptical and radial basis function neural networks and other models to learn the statistical distributions of process observables under normal conditions. Analytical models of failure modes can then be applied in combination with the neural network models to identify faults. Special methods can be applied to compensate for sensor failures, to produce real-time estimation of missing or failed sensors based on the correlations codified in the neural network.

Kramer, Mark A.↗

A validated methodology for the prediction of heating and cooling energy demand for buildings within the Urban Heat Island: Case-study of London

This paper describes a method for predicting air temperatures within the Urban Heat Island at discreet locations based on input data from one meteorological station for the time the prediction is required and historic measured air temperatures within the city. It uses London as a case-study to describe the method and its applications. The prediction model is based on Artificial Neural Network (ANN) modelling and it is termed the London Site Specific Air Temperature (LSSAT) predictor. The temporal and spatial validity of the model was tested using data measured 8 years later from the original dataset; it was found that site specific hourly air temperature prediction provides acceptable accuracy and improves considerably for average monthly values. It thus is a very reliable tool for use as part of the process of predicting heating and cooling loads for urban buildings. This is illustrated by the computation of Heating Degree Days (HDD) and Cooling Degree Hours (CDH) for a West-East Transect within London. The described method could be used for any city for which historic hourly air temperatures are available for a number of locations; for example air pollution measuring sites, common in many cities, typically measure air temperature on an hourly basis. (author)

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Modelling the Value of Networked, Small-Hydro Generators for Storage and Dispatch Capability on the Grid

Natel Energy, Inc. (Natel), Alameda, California, has developed a new type of fish-friendly, low-head hydropower turbine, the hydroEngine (R), that is designed to provide reliable, dispatchable power and ancillary services at significantly reduced civil works costs. For the U.S. Department of Energy's (DOE) Small Business Voucher for Hydropower, Third Round, Natel and NREL will model and evaluate the potential value streams of a portfolio of networked small hydropower generators. The project is designed to advance the understanding of the value of small hydro generation to the grid of the future. It will help support dialog with potential customers and stakeholders regarding the value of the Natel's technologies and inform future R&D efforts in this area.

13 HYDRO ENERGY↗

Solar Transposition Modeling via Deep Neural Networks With Sky Images

This article presents a notable advance toward the development of a new method of increasing the single-axis tracking photovoltaic (PV) system power output by improving the determination and near-term prediction of the optimum module tilt angle. The tilt angle of the plane receiving the greatest total irradiance changes with Sun position and atmospheric conditions including cloud formation and movement, aerosols, and particulate loading, as well as varying albedo within a module's field of view. In this article, we present a multi-input convolutional neural network that can create a profile of plane-of-array irradiance versus surface tilt angle over a full 180° arc from horizon to horizon. As input, the neural network uses the calculated solar position and clear-sky irradiance values, along with sky images. The target irradiance values are provided by the multiplanar irradiance sensor (MPIS). In order to account for varying irradiance conditions, the MPIS signal is normalized by the theoretical clear-sky global horizontal irradiance. Using this information, the neural network outputs an N -dimensional vector, where N is the number of points to approximate the MPIS curve via Fourier resampling. The output vector of the model is smoothed with a Gaussian kernel to account for error in the downsamping and subsequent upsampling steps, as well as to smooth the unconstrained output of the model. Furthermore, these profiles may be used to perform near-term prediction of angular irradiance, which can then inform the movement of a PV tracker.

42 ENGINEERING↗

A Hierarchical Modeling for Reactive Power Optimization With Joint Transmission and Distribution Networks by Curve Fitting

Here, in order to solve the reactive power optimization with joint transmission and distribution networks, a hierarchical modeling method is proposed in this paper. It allows the reactive power optimization of transmission and distribution networks to be performed separately, leading to a master–slave structure and improves traditional centralized modeling methods by alleviating the big data problem in a control center. Specifically, the transmission-distribution-network coordination issue of the hierarchical modeling method is investigated. First, a curve-fitting approach is developed to provide a cost function of the slave model for the master model, which reflects the impacts of each slave model. Second, the transmission and distribution networks are decoupled at feeder buses, and all the distribution networks are coordinated by the master reactive power optimization model to achieve the global optimality. Finally, numerical results on two test systems verify the effectiveness of the proposed hierarchical modeling and curve-fitting methods.

24 POWER TRANSMISSION AND DISTRIBUTION↗