SEARCH · Search NASA
Results for “model accuracy”
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.
Watertight Tensor-Product Spline Reconstruction (KCNSC PDRD Technical Report)
The standard for representing geometric models in Computer Aided Design (CAD) systems is boundary representation with trimmed surfaces. However, the approximate nature of surface-surface intersections in CAD systems leaves many small gaps between mating surfaces. These gaps lead to problems for downstream analysis and manufacturing applications, requiring time consuming repair operations and further approximation of the geometry as polygonal meshes, which do not support design iteration. In this study, nVariate, Inc. evaluated its watertight reconstruction technology on a test part. The resulting watertight model has the same absolute model accuracy of the original model zero gaps, providing a representation that requires no repair and supports further design iterations in the CAD system.
Real-time infrared spectroscopy coupled with blind source separation for nuclear waste process monitoring
On-line infrared absorbance spectroscopy enables rapid measurement of solution-phase molecular species. Many spectra-to-concentration models exist for spectral data, with some models able to handle overlapping spectral bands and nonlinearities. However, model accuracy is limited by the quality of training data used in model fitting. The process spectra of nuclear waste simulants at the Savannah River Site display incongruity between training and process spectra; the glycolate spectral signature in the training data does not match the glycolate signature in Savannah River National Laboratory process data. A novel blind source separation algorithm is proposed that preprocesses spectral data so that process spectra more closely resemble training spectra, thereby improving model quantification accuracy when unexpected sources of variation appear in process spectra. The novel blind source separation preprocessing algorithm is shown to improve nitrate quantification from an R 2 of 0.934 to 0.988 and from 0.267 to 0.978 in two instances analyzing nuclear waste simulants from the Slurry Receipt Adjustment Tank and Slurry Mix Evaporator cycle at the Savannah River Site.
A High-Accuracy Power Loss Model of SiC MOSFETs in Current Source Inverter Considering Current Commutation and Parasitic Parameters
Power loss estimation of power electronic devices is important to the efficiency optimization of motor drives used in many applications. However, most existing simplified power loss models of silicon carbide (SiC) MOSFETs are not sufficiently accurate due to their neglect of parasitic parameters in the current commutation loop. In addition, the loss model of the voltage source inverter (VSI) cannot be directly applied to the current source inverter (CSI) because of differences between their commutation loops. First, the commutation processes for VSIs and CSIs are compared. The voltage and current trajectories of SiC MOSFETs in the switching transition of a CSI-based motor drive system are analyzed in detail. Based on these results, the conduction and switching losses of SiC MOSFETs in CSIs are modeled considering the current commutation details. In addition, the proposed analytical model includes parasitic inductances and capacitances in the current commutation loop. Experimental results have verified that the proposed power loss model delivers higher accuracy loss predictions than the conventional loss model.
Statistical and Machine Learning Approaches to Analyzing Pipeline Incidents in the United States (2010–2024)
This study applies machine learning methods to analyze natural gas pipeline incidents in the United States using the Pipeline and Hazardous Materials Safety Administration (PHMSA) Gas Distribution Incident Dataset (2010–2024). The dataset includes over 600 variables describing incident characteristics, infrastructure attributes, and contributing factors associated with unintentional gas releases. The objective is to assess whether these features can reliably predict the underlying cause of pipeline failures. Multinomial logistic regression and Random Forest models were developed to classify incident causes, including excavation damage, corrosion, equipment failure, and natural forces. Results show that excavation damage is both the most frequent and most predictable cause, with models achieving strong performance for this category. However, when excavation damage is excluded, model accuracy declines significantly, with some models performing near random levels. Across all approaches, severe class imbalance and limited variability in key predictors constrain predictive performance. Pipeline age and diameter emerge as the most influential variables, but they provide insufficient discriminatory power to distinguish among less frequent failure types. These findings indicate that non-excavation-related incidents are rare, heterogeneous, and weakly represented in the dataset, limiting the effectiveness of machine learning classification. Overall, this study highlights the structural limitations of the PHMSA dataset for predictive modeling and underscores the need for improved data balance and feature enrichment. The results reinforce excavation damage prevention as the most impactful strategy for reducing pipeline incidents.
Array-Based Machine Learning for Functional Group Detection in Electron Ionization Mass Spectrometry
Mass spectrometry is a ubiquitous technique capable of complex chemical analysis. The fragmentation patterns that appear in mass spectrometry are an excellent target for artificial intelligence methods to automate and expedite the analysis of data to identify targets such as functional groups. To develop this approach, we trained models on electron ionization (a reproducible hard fragmentation technique) mass spectra so that not only the final model accuracies but also the reasoning behind model assignments could be evaluated. The convolutional neural network (CNN) models were trained on 2D images of the spectra using transfer learning of Inception V3, and the logistic regression models were trained using array-based data and Scikit Learn implementation in Python. Our training dataset consisted of 21,166 mass spectra from the United States’ National Institute of Standards and Technology (NIST) Webbook. The data was used to train models to identify functional groups, both specific (e.g., amines, esters) and generalized classifications (aromatics, oxygen-containing functional groups, and nitrogen-containing functional groups). We found that the highest final accuracies on identifying new data were observed using logistic regression rather than transfer learning on CNN models. It was also determined that the mass range most beneficial for functional group analysis is 0–100 m/z. We also found success in correctly identifying functional groups of example molecules selected from both the NIST database and experimental data. Beyond functional group analysis, we also have developed a methodology to identify impactful fragments for the accurate detection of the models’ targets. The results demonstrate a potential pathway for analyzing and screening substantial amounts of mass spectral data.
Large-Scale Materials Modeling at Quantum Accuracy: Ab Initio Simulations of Quasicrystals and Interacting Extended Defects in Metallic Alloys
Ab initio electronic-structure has remained dichotomous between achievable accuracy and length-scale. Quantum many-body (QMB) methods realize quantum accuracy but fail to scale. Density functional theory (DFT) scales favorably but remains far from quantum accuracy. We present a framework that breaks this dichotomy by use of three interconnected modules: (i) invDFT: a methodological advance in inverse DFT linking QMB methods to DFT; (ii) MLXC: a machine-learned density functional trained with invDFT data, commensurate with quantum accuracy; (iii) DFT-FE-MLXC: an adaptive higher-order spectral finite-element (FE) based DFT implementation that integrates MLXC with efficient solver strategies and HPC innovations in FE-specific dense linear algebra, mixed-precision algorithms, and asynchronous compute-communication. Furthermore, we demonstrate a paradigm shift in DFT that not only provides an accuracy commensurate with QMB methods in ground-state energies, but also attains an unprecedented performance of 659.7 PFLOPS (43.1% peak FP64 performance) on 619,124 electrons using 8,000 GPU nodes of Frontier supercomputer.
Assessing the Accuracy of Property Model Predictions for Cost Optimization of Desalination Technologies
Accurate modeling of seawater thermophysical and thermodynamic properties is critical for optimizing desalination processes. This study compares three seawater property models, a Reaktoro multicomponent model, the thermophysical seawater properties library from the Massachusetts Institute of Technology, and a simplified sodium chloride model, in the context of levelized cost of water (LCOW) minimization for reverse osmosis (RO) and mechanical vapor compression systems. Process simulations and cost optimizations reveal that although all three models yield comparable LCOW and specific energy consumption (SEC) estimates under baseline conditions, deviations among their predictions increase with salinity. Relative differences in LCOW and SEC reach up to 6% and 8%, respectively. RO results show greater variability due to differences in osmotic pressure predictions, which affect pressure constraints at high recoveries. Computational performance varies substantially; specifically, Reaktoro simulations are up to 28 times slower than empirical models due to their detailed equilibrium calculations. These results suggest that empirical models offer acceptable accuracy for routine desalination process design, while Reaktoro provides advantages in scenarios requiring detailed speciation, such as scaling or pH adjustment studies. These findings underscore the importance of selecting appropriate property models based on the modeling objective of desalination applications and motivate future work integrating thermodynamic rigor with empirical efficiency.
The ellipticity parameterization for an NFW profile: An overlooked angular structure in strong lens modeling
Galaxy-scale gravitational lenses are often modeled with two-component mass profiles where one component represents the stellar mass and the second is a Navarro Frenk White (NFW) profile representing the dark matter. Outside of the spherical case, the NFW profile is costly to implement, and so it is approximated via two different methods; ellipticity can be introduced via the lensing potential (NFWp) or via the mass by approximating the NFW profile as a sum of analytical profiles (NFWm). While the NFWp method has been the default for lensing applications, it gives a different prescription of the azimuthal structure, which we show introduces ubiquitous gradients in ellipticity and boxiness in the mass distribution rather than having a constant elliptical shape. Because an unmodeled azimuthal structure has been shown to be able to bias lens model results, we explored the degree to which this azimuthal structure that was introduced can affect the model accuracy. We constructed input profiles using composite models using both the NFWp and NFWm methods and fit these mocks with a power-law elliptical mass distribution (PEMD) model with external shear. As a measure of the accuracy of the recovered lensing potential, we calculated the value of the Hubble parameter H 0 one would determine from the lensing fit. We found that the fits to the NFWp input return H 0 values that are systematically biased by about 3% lower than the NFWm counterparts. We explored whether such an effect is attributable to the mass sheet transformation (MST) by using an MST-independent quantity, ξ 2 . We show that, as expected, the NFWm mocks are degenerate with PEMD through an MST. For the NFWp, an additional bias was found beyond the MST due to the azimuthal structure exterior to the Einstein radius. We recommend modelers use an NFWm prescription in the future, such that the azimuthal structure can be introduced explicitly rather than implicitly.
Characterizing vertical upper ocean temperature structures in the European Arctic through unsupervised machine learning
In-situ observations of subsurface ocean temperatures are, in many regions, inconsistently distributed in time and space. These spatio-temporal inconsistencies in the observational network lead to difficulties in utilizing those observations effectively for ocean model evaluation or understanding larger-scale ocean characteristics. Model accuracy of subsurface ocean characteristics is especially important within regions that contain complex ocean structures. One such region is the European Arctic which not only contains several types of water masses with unique characteristics, but also wintertime sea ice coverage and complex bathymetry. This study presents an unsupervised neural networking technique that can be used in combination with traditional ocean model evaluation techniques to provide additional information on the accuracy of modeled vertical ocean temperature profiles. Self-organizing maps is an unsupervised machine learning technique that we apply to approximately twenty thousand Argo and CTD temperature profiles from 2012 to 2020 in the European Arctic to categorize the observed vertical ocean temperature structures in the top 150 m. The observed ocean profile categories, or neurons, defined by the self-organizing map show strong spatial and temporal dependencies. We then use the neuron weights, or the learned temperature profile structure of each neuron, to validate the spatial and temporal variability of modeled vertical temperature structures. This analysis gives us new insights about the model’s capabilities to reproduce specific vertical structures of the top-most ocean layer within different regions and seasons. Mapping modeled ocean temperature profiles onto the neuron-space of the observationally-defined self organized map highlights the potential of this method to advance our understanding of model deficiencies in that region.
Chlorophyll- a dynamics in the lower Amazon River: insights from in situ and hyperspectral remote sensing using OCI-PACE
Chlorophyll-a concentration (Chla) is a key indicator of phytoplankton biomass and aquatic trophic status. However, satellite-derived Chla in sediment-rich waters, such as those found in the Lower Amazon River, remains challenging. The present study characterizes in situ Chla levels and their relationships with geographic, physical, and biogeochemical parameters in the Lower Amazon. Data collected between 2014 and 2017 across four hydrological seasons included measurements of Chla, remote sensing reflectance, and water quality parameters such as total suspended sediment, conductivity, water surface temperature, dissolved oxygen, pH, dissolved organic carbon and coloured dissolved organic matter. An empirical model was developed to estimate Chla using simulated hyperspectral bands from NASA’s PACE mission, achieving high performance (R 2 = 0.76; RMSE = 0.11 μg·L −1 ). Red bands proved particularly effective for Chla retrieval, while the addition of ultraviolet bands further enhanced model accuracy. The application of the developed model to satellite imagery yielded results consistent with in situ observations for the same hydrologic season. Seasonal variation and geographic location were major factors influencing Chla dynamics. This study provides a novel contribution to Chla estimation in optically complex, highly turbid waters and highlights the potential of the PACE mission to enhance global aquatic ecosystem monitoring. In conclusion, by offering freely available hyperspectral data with high radiometric resolution, PACE represents a significant advancement in the realm of remote sensing of aquatic environments.
Developing Data-Driven Synthetic Infrastructure Models for Resilience Analysis
Research on infrastructure resilience has produced promising methods to simulate and optimize complex networks to improve performance. However, restrictions on sharing infrastructure models and the steep cost of developing and maintaining infrastructure models presents a roadblock to adoption. To overcome this limitation, this research focuses on methods to create data-driven infrastructure models that will help improve infrastructure resilience and security. The analysis couples incomplete utility data, geospatial data, machine learning, and synthetic network generation methods to rapidly develop and update infrastructure models. The methods are validated using realistic utility models and site-specific data, with a focus on Puerto Rico due to its unique infrastructure challenges and available data. This research highlights promising opportunities for the use of synthetic network generation and machine learning to create infrastructure models when very little data is available. Results demonstrate that hybrid methods, which combine sparse utility data with synthetic models, can enhance model accuracy, and machine learning can predict model attributes using training data from other models. However, the complexity of infrastructure systems means that even minor changes in network connectivity can significantly impact simulation results. Resilience analysis using synthetic infrastructure models shows that while some system behaviors are preserved, the magnitude of disruptions may not be accurately represented, indicating the need for more research and validation before using synthetic models for critical infrastructure investment decisions. The framework outlined in this report represents a significant advance to infrastructure model development and could be applied to additional domains and sites. Future research will continue to streamline and validate methods to help reduce roadblocks to resilience analysis.
Many-body reactive force field development for carbon condensation in C/O systems under extreme conditions
In this paper, we describe the development of a reactive force field for C/O systems under extreme temperatures and pressures, based on the many-body Chebyshev Interaction Model for Efficient Simulation (ChIMES). The resulting model, which targets carbon condensation under thermodynamic conditions of 6500 K and 2.5 g cm –3 , affords a balance between model accuracy, complexity, and training set generation expense. We show that the model recovers much of the accuracy of density functional theory for the prediction of structure, dynamics, and chemistry when applied to dissociative condensed phase systems at 1:1 and 1:2 C:O ratios, as well as molten carbon. Our C/O modeling approach exhibits a 104 increase in efficiency for the same system size (i.e., 128 atoms) and a linear system size scalability over standard quantum molecular dynamics methods, allowing the simulation of significantly larger systems than previously possible. We find that the model captures the condensed-phase reaction-coupled formation of carbon clusters implied by recent experiments, and that this process is susceptible to strong finite size effects. Overall, we find the present ChIMES model to be well suited for studying chemical processes and cluster formation at pressures and temperatures typical of shock waves. We expect that the present C/O modeling paradigm can serve as a template for the development of a broader high pressure–high temperature force-field for condensed phase chemistry in organic materials.
Prediction of Potential and Actual Evapotranspiration Fluxes Using Six Meteorological Data-Based Approaches for a Range of Climate and Land Cover Types
Evapotranspiration is the major component of the water cycle, so a correct estimate of this variable is fundamental. The purpose of the present research is to assess the monthly scale accuracy of six meteorological data-based models in the prediction of evapotranspiration (ET) losses by comparing the modelled fluxes with the observed ones from eight sites equipped with eddy covariance stations which differ in terms of vegetation and climate type. Three potential ET methods (Penman-Monteith, Priestley-Taylor, and Blaney-Criddle models) and three actual ET models (the Advection-Aridity, the Granger and Gray, and the Antecedent Precipitation Index method) have been proposed. The findings show that the models performances differ from site to site and they depend on the vegetation and climate characteristics. Indeed, they show a wide range of error values ranging from 0.18 to 2.78. It has been not possible to identify a single model able to outperform the others in each biome, but in general, the Advection-Aridity approach seems to be the most accurate, especially when the model calibration in not carried out. It returns very low error values close to 0.38. When the calibration procedure is performed, the most accurate model is the Granger and Gray approach with minimum error of 0.13 but, at the same time, it is the most impacted by this process, and therefore, in a context of data scarcity, it results the less recommended for ET prediction. The performances of the investigated ET approaches have been furthermore tested in case of lack of measured data of soil heat fluxes and net radiation considering using empirical relationships based on meteorological data to derive these variables. Results show that, the use of empirical formulas to derive ET estimates increases the errors up to 200% with the consequent loss of model accuracy.
A Bayesian Learning Approach to Wireless Outdoor Heatmap Construction using Deep Gaussian Process
We present a novel Bayesian learning approach to outdoor radio heatmap construction utilizing deep Gaussian process (GP). The proposed approach employs a two-layer hierarchy which consists of two cascaded Gaussian processes that are capable of modeling more complex input-output relations than standard single-layer Gaussian processes. Since deriving the exact model likelihood is challenging, a lower bound is optimized instead so that gradient descent-based methods can be performed to find out the optimal model parameters. Typically, inducing points are used in GPs to facilitate low-rank approximation of covariance (kernel) matrices for computation speedup. However, the inaccuracy induced by inducing points can accumulate when stacking multiple layers of GP which may hinder the performance of deep GP. Moreover, since inducing points need to be learned, having them at all layers of deep GP also incurs computational burden. To overcome the above challenges, in contrast to the canonical deep GP model, we use a modified architecture where a full standard GP resides in the first layer and inducing points are only introduced for the second layer. This modified architecture strikes a balance between model accuracy and training complexity. In the proposed model, the noise parameter of the first GP layer is also eliminated to improve the training efficiency as the noise parameter at the output of the second layer suffices to model the uncertainty in the output. The proposed approach is evaluated on real-world datasets, in the form of location-Received Signal Strength (RSS) pairs, collected from the Platform for Open Wireless Data-driven Experimental Research (POWDER) located at the campus of the University of Utah. Experiment results show that the proposed approach can achieve smaller prediction errors on various training and testing data configurations than DNN-based and GP-based methods.
IBR Bulk System Protection Issues and Modeling Challenges
Issue to be Address: Difficulty in predicting Inverter (IBR) generation dynamic response characteristics. Main challenges are unavailability of proprietary control details for modeling and the aggregate mixed IBRs with different dynamic response characteristics all contributing during system faults. Framework for Discussion: Identifying issues by sharing experience Reviewing options to improve IBR modeling accuracy, computation speed, use of test results. Presenting related research by EPRI and PG&E (DOE project) Desired Outcomes: Provide technical update for participants Get feedback and suggestions for ongoing research Improve research collaboration
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.
Improving Computational Efficiency of Mechanical Finite Element Method Simulations for PV Modules: Preprint
In this work we have elucidated the trade off between structural mechanics FEM model accuracy and computation time by employing lower fidelity viscoelastic models. Results indicate that computation time can easily be cut in half while only expecting a potential maximum error of 10 % by considering lower fidelity models. A novel approach to produce the Prony Series fit for viscoelastic characterization has also been presented. Employing this approach, we were able to achieve a further 10 % reduction in computation time without further sacrificing simulation accuracy.