Search NASA⌕ Search

SEARCH · Search NASA

Results for “Model generalization”

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 253 records · Page 14
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Exotic phases in finite-density ℤ 3 theories

Lattice ℤ 3 theories with complex actions share many key features with finite- density QCD including a sign problem and $\mathcal{CK}$ symmetry. Complex ℤ 3 spin and gauge models exhibit a generalized Kramers-Wannier duality mapping them onto chiral ℤ 3 spin and gauge models, which are simulatable with standard lattice methods in large regions of parameter space. The Migdal-Kadanoff real-space renormalization group (RG) preserves this duality, and we use it to compute the approximate phase diagram of both spin and gauge ℤ 3 models in dimensions one through four. Chiral ℤ 3 spin models are known to exhibit a Devil’s Flower phase structure, with inhomogeneous phases that can be thought of as ℤ 3 analogues of chiral spirals. Out of the large class of models we study, we find that only chiral spin models and their duals have a Devil’s Flower structure with an infinite set of inhomogeneous phases, a result we attribute to Elitzur’s theorem. We also find that different forms of the Migdal-Kadanoff RG produce different numbers of phases, a violation of the expectation for universal behavior from a real-space RG. We discuss extensions of our work to ℤ N models, SU(N) models and nonzero temperature.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Development and Verification of an Improved Wake-Added Turbulence Model in FAST.Farm

We introduce a generalized wake-added turbulence (WAT) model in the multiphysics, multiturbine simulation tool FAST.Farm. The WAT model introduces additional small-scale turbulence that represents the breakdown of vortical structures and shear layers in the wake. The article describes the development, implementation, calibration, and verification of the model. The novelties of the model include support for wake asymmetry, buildup of WAT across the wind farm, and secondary effects of wake-induced turbulence (e.g., wake meandering) driven by smaller-scale turbulence structures that arise from wake breakdown. Large-eddy simulations were run to support the calibration of the WAT parameters and verification of the model. Previous studies hypothesized that the lack of WAT modeling was the source of underprediction of fatigue loads, in particular for cases at low turbulence intensities and/or stable atmospheric boundary layers. This study confirms that the newly implemented WAT model enhances the loads predictions in these cases.

17 WIND ENERGY↗
DOE OSTI2026

The Value of Forecasters‐in‐the‐Loop in Real‐Time Flood Forecasting in the Age of Machine Learning

Machine learning (ML) applications in hydrological forecasting are increasingly prevalent and show great potential. However, many previous studies have only evaluated performance through reanalysis or retrospective simulations compared to simplified baselines. This study provides the first assessment of ML performance against actual operational forecasting systems operated by the California Nevada River Forecast Center (CNRFC), which combines the Community Hydrologic Prediction System (CHPS) with forecasters-in-the-loop. Results demonstrate that forecasters-in-the-loop systems consistently outperform ML models in both general forecasts and flood alerting across lead times up to 96 hr, even when ML models use observed forcings, while CNRFC operational process relies on biased weather forecasts. Our analysis reveals that forecaster expertise maintains forecast reliability despite inaccurate precipitation inputs, with human-guided systems showing superior performance degradation characteristics at extended lead times. These findings highlight the irreplaceable value of human expertise in operational forecasting and caution against overstating current ML capabilities in real-world applications.

Tran, Vinh Ngoc [Univ. of Michigan, Ann Arbor, MI ↗
DOE OSTI2025

Conin

SAND2025-07645O Conin is a Python library that supports constrained analysis of probabilistic graphical models (PGMs). It enables constrained inference and learning for hidden Markov models, Bayesian networks, dynamic Bayesian networks, and Markov networks. Conin interfaces with the pgmpy library to specify general probabilistic graphical models with a variety of optimization solvers to support learning and inference. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

Hart, William [Sandia National Lab. (SNL-CA), Live↗
DOE OSTI2025

Using convolutional neural networks to accelerate three-dimensional coherent synchrotron radiation computations

Calculating the effects of coherent synchrotron radiation (CSR) is one of the most computationally expensive tasks in accelerator physics. Here, we use convolutional neural networks (CNNs), along with a latent conditional diffusion (LCD) model, trained on physics-based simulations to speed up calculations. Specifically, we produce the 3D CSR wakefields generated by electron bunches in circular orbit in the steady-state condition. Two datasets are used for training and testing the models: wakefields generated by three-dimensional Gaussian electron distributions and wakefields from a sum of up to 25 three-dimensional Gaussian distributions. The CNNs are able to accurately produce the 3D wakefields ∼250–1000 times faster than the numerical calculations, while the LCD achieves a gain of a factor of ∼34. We also test the extrapolation and out-of-distribution generalization ability of the models. They generalize well on distributions with larger spreads than what they were trained on but struggle with smaller spreads.

43 PARTICLE ACCELERATORS↗
DOE OSTI2024

Pressure-Gradient-Based RANS Model for Predicting Separation in Transitional and Turbulent Flows

Predicting flow separation poses a significant challenge for RANS models, particularly in transitional flows over airfoils. We propose a novel improvement to RANS models to predict incipient separation in both transitional and fully turbulent flows. Our approach modifies the eddy viscosity model in regions indicated by a pressure-gradient criterion that accounts for intermittency - determining whether the boundary layer is laminar or turbulent. This model demonstrates robust generalization across various airfoil shapes and Reynolds numbers. Applied to the NREL Phase VI wind turbine rotor, our model shows improved aerodynamic performance predictions compared to the baseline RANS model.

k-omega SST↗
DOE OSTI2024

Adaptively remeshed multiphysical modeling of resistance forge welding with experimental validation of residual stress fields and measurement processes

Welding processes used in the production of pressure vessels impart residual stresses in the manufactured component. Computational modeling is critical to predicting these residual stress fields and understanding how they interact with notches and flaws to impact pressure vessel durability. Here, in this work, we present a finite element model for a resistance forge weld and validate it using laboratory measurements. Extensive microstructural changes, near-melt temperatures, and large localized deformations along the weld interface pose significant challenges to Lagrangian finite element modeling. The proposed modeling approach overcomes these roadblocks in order to provide a high-fidelity simulation that can predict the residual stress state in the manufactured pressure vessel; a rich microstructural constitutive model accounts for material recrystallization dynamics, a frictional-to-tied contact model is coordinated with the constitutive model to represent interfacial bonding, and adaptive remeshing is employed to alleviate severe mesh distortion. An interrupted-weld approach is applied to the simulation to facilitate comparison to displacement measures. Several techniques are employed for residual stress measurement in order to validate the finite element model: neutron diffraction, the contour method, and the slitting method. Model-measurement comparisons are supplemented with detailed simulations that reflect the configurations of the residual-stress measurement processes themselves. The model results show general agreement with experimental measurements, and we observe some similarities in the features around the weld region. Factors that contribute to model-measurement differences are identified. Finally, we conclude with some discussion of the model development and residual stress measurement strategies, including how to best leverage the efforts put forth here for other weld problems.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Performance Comparison of Machine Learning Models for Ultrasonic Nondestructive Evaluation of Alkali-Silica Reaction in Concrete

Alkali-silica reaction (ASR) causes concrete degradation, leading to cracking, rebar corrosion, and reduced structural integrity, which raises safety concerns. Ultrasonic nondestructive evaluation (NDE) effectively assesses concrete properties and monitors ASR progression. However, its deployment and analysis require specialized expertise and subjective interpretation. As computational power increases, artificial intelligence (AI) and machine learning (ML) algorithms are increasingly being used to automate NDE data analysis across various industries for AI-assisted automation. Regulatory agencies are adapting to this technological shift, prompting a need to evaluate current ML technologies’ capabilities and limitations in assessing concrete material properties and damage. This report presents a comparative analysis of four ML regression models for predicting concrete material damage induced by ASR expansion using long-term ultrasonic data monitoring. The models investigated include linear regression (LR), support vector regression (SVR), shallow neural networks (NN), and deep neural networks (DNN). LR, SVR, and shallow NN models use features extracted from ultrasonic signals, whereas the DNN model processes time-domain ultrasonic signals and frequency spectra directly. The study systematically compared the models’ performance from various perspectives, including model input, prediction performance, and generalization ability. The findings indicate significant variability in model performance, with some ML algorithms achieving very high or very low prediction accuracy depending on the preprocessing and feature engineering (extraction and selection) applied. Key insights include the observation that shallow ML models (LR, SVR, and shallow NNs) require meticulous preprocessing and feature extraction to achieve high accuracy. In contrast, the DNN model, although it bypasses the need for feature engineering, necessitates extensive preprocessing to mitigate noise and computational demands. The SVR model emerged as the top performer among the shallow models, and the DNN model exhibited superior performance on specific datasets but struggled with generalization across specimens from different batches. Additionally, the SVR model is sensitive to temperature variations, whereas the DNN model is robust in this regard. Using recurrent neural networks is recommended for future ASR expansion prediction studies. Recurrent neural networks’ inherent ability to capture temporal dependencies and long-term patterns makes them well suited for analyzing sequential ultrasonic monitoring data. Overall, the results and conclusions of this study could provide insights into the capabilities and effectiveness of ML when applied to ultrasonic NDE data and help identify best practices for using ML for ultrasonic NDE of concrete material properties.

36 MATERIALS SCIENCE↗
DOE OSTI2025

A Review and Comparison of Different Sensitivity Analysis Techniques in Practice

There exist many methods for sensitivity analysis readily available to the practitioner. While each seeks to help the modeler answer the same general question – How do sources of uncertainty or changes in the model inputs relate to uncertainty in the output? – different methods are associated with different assumptions, constraints, and required resources, leading to conclusions that may vary in interpretability and level of detail. Thus, it is crucial that the practitioner selects the desired sensitivity analysis method judiciously, making sure to match the selected approach to the specifics of their problem and to their desired objectives. In this chapter, we provide a practical overview of a collection of widely used, widely available sensitivity analysis methods. We focus on global sensitivity approaches, which seek to characterize how uncertainty in the model output may be allocated to sources of uncertainty in model inputs across the entire input space. Generally, this will require the practitioner to specify a probability distribution over the input space. On the other hand, methods for local sensitivity analysis do not require this specification but they have more limited utility, providing insight into sources of uncertainty associated only with a particular, specified location in the input space. Our hope is that this chapter may serve as a decision-making tool for practitioners, helping to guide the selection of a sensitivity analysis approach that will best fit their needs. To support this goal, we have selected a suite of approaches to cover, which, while not exhaustive, we believe provides a flexible and robust sensitivity analysis toolkit. All methods included are widely used and available in standard software packages.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Global energy balance simulations of shattered pellet injection

The dynamics of shattered pellet injection (SPI) shutdowns are simulated using a time-dependent global energy balance model, based on a modification of the KPRAD framework. The new SPI particle source in the model calculates the ablation of individual pellet fragments that enter the plasma as a temporally resolved plume, thus capturing the effects of earlier fragments on the ablation of those that follow, which has a significant impact on the overall assimilation. Despite the reduced physics and the global averaging of all quantities, results from a large number of DIII-D, KSTAR, and JET experiments are well reproduced, including the plasma cooling timescales, particle assimilation, and current quench (CQ) rates. Cooling timescales and CQ rates are in good agreement for pellets containing as little as ∼1% neon, while particle assimilations are most accurate for neon fractions above ∼15% by number of atoms. Below this, the assimilation tends to be overestimated due to the lack of radial transport in the particle balance, which becomes important in the low-Z limit. Predictive simulations of mixed-composition dual-SPI shutdowns in ITER are compared against those with the 3D non-linear magnetohydrodynamic code JOREK and are found to reproduce overall trends observed in the higher-fidelity modeling across a range of injection scenarios. The general success of the model points to the critical role of energy balance in determining SPI particle assimilation and the subsequent disruption dynamics and highlights the value of these simulations for experimental interpretation and for optimizing the deployment of computationally expensive, higher-fidelity models.

energy balance model↗
DOE OSTI2026

SAIGE-GPU: accelerating genome- and phenome-wide association studies using GPUs

Genome-wide association studies (GWAS) at biobank scale are computationally intensive, especially for admixed populations requiring robust statistical models. SAIGE is a widely used method for generalized linear mixed-model GWAS but is limited by its CPU-based implementation, making phenome-wide association studies impractical for many research groups. We developed SAIGE-GPU, a GPU-accelerated version of SAIGE that replaces CPU-intensive matrix operations with GPU-optimized kernels. The core innovation is distributing genetic relationship matrix calculations across GPUs and communication layers. Applied to 2068 phenotypes from 635 969 participants in the Million Veteran Program, including diverse and admixed populations, SAIGE-GPU achieved a 5-fold speedup in mixed model fitting on supercomputing infrastructure and cloud platforms. We further optimized the variant association testing step through multi-core and multi-trait parallelization. Deployed on Google Cloud Platform and Azure, the method provided substantial cost and time savings. Source code and binaries are available for download at https://github.com/saigegit/SAIGE/tree/SAIGE-GPU-1.3.3. A code snapshot is archived at Zenodo for reproducibility (DOI: [10.5281/zenodo.17642591]). SAIGE-GPU is available in a containerized format for use across HPC and cloud environments and is implemented in R/C++ and runs on Linux systems.

Rodriguez, Alex [Argonne National Laboratory (ANL)↗
DOE OSTI2024

Exotic edge states of C 3 high-fold fermions in honeycomb lattices

A generalization of the graphene honeycomb model to the case where each site in the honeycomb lattice contains a n -fold degenerate set of eigenstates of the C 3 symmetry has been recently proposed to describe several systems, including triangulene crystals and photonic lattices. These generalized honeycomb models are defined by ( n a , n b ) , the number of C 3 eigenstates in the a and b sites of the unit cell, resulting in n a + n b bands. Thus, the (1,1) case gives the coventional honeycomb model that describes the two low-energy bands in graphene. Generalizations, such as (2,1), (2,2), and (3,3) display several nontrivial features, such as coexisting graphenelike Dirac cones with flat bands, both at zero and finite energy, as well as robust degeneracy points where a flat band and a parabolic band meet at the Γ point. Here we explore the edge states of this class of crystals, using as reference triangulene crystals, and we find several types of edge states absent in the conventional (1,1) honeycomb case, associated to the nontrivial features of the two-dimensional bands of the high-fold case. First, we find dispersive edge states associated to the finite-energy flat bands, that occur both at the armchair and zigzag termination. Second, in the case of noncentrosymmetric triangulene crystals that lead to a S = 1 Dirac band, we have a bonding-antibonding pair of dispersive edge states, localized in the same edge so that their energy splitting is reduced as their localization increases, opposite to the conventional behavior of pairs of states localized in opposite edges. Third, for the (3,3) case, that hosts a gap separating a pair of flat conduction and valence bands, we find nondispersive edge states with E = 0 in all edge terminations. Published by the American Physical Society 2024

Madail, L. (ORCID:0000000208457748)↗
DOE OSTI2024

Generalization error guaranteed auto-encoder-based nonlinear model reduction for operator learning

Many physical processes in science and engineering are naturally represented by operators between infinite-dimensional function spaces. The problem of operator learning, in this context, seeks to extract these physical processes from empirical data, which is challenging due to the infinite or high dimensionality of data. An integral component in addressing this challenge is model reduction, which reduces both the data dimensionality and problem size. In this paper, we utilize low-dimensional nonlinear structures in model reduction by investigating Auto-Encoder-based Neural Network (AENet). AENet first learns the latent variables of the input data and then learns the transformation from these latent variables to corresponding output data. Our numerical experiments validate the ability of AENet to accurately learn the solution operator of nonlinear partial differential equations. Furthermore, we establish a mathematical and statistical estimation theory that analyzes the generalization error of AENet. Finally, our theoretical framework shows that the sample complexity of training AENet is intricately tied to the intrinsic dimension of the modeled process, while also demonstrating the robustness of AENet to noise.

Auto-encoder↗
DOE OSTI2024

A supersymmetric SYK model with a curious low energy behavior

We consider N = 2,4 supersymmetric SYK models that have a peculiar low energy behavior, with the entropy going like S = S 0 + (constant)T a , where a ≠ 1. The large N equations for these models are a generalization of equations that have been previously studied as an unjustified truncation of the planar diagrams describing the BFSS matrix quantum mechanics or other related matrix models. Here we reanalyze these equations in order to better understand the low energy physics of these models. We find that the scalar fields develop large expectation values which explore the low energy valleys in the potential. The low energy physics is dominated by quadratic fluctuations around these values. These models were previously conjectured to have a spin glass phase. We did not find any evidence for this phase by using the usual diagnostics, such as searching for replica symmetry breaking solutions.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2026

Validating a Dynamic PWR Safety and Security Model?

Nuclear power plants (NPPs) are assessed for safety and security using separate models that cannot capture how an attacker's decisions and a plant's response unfold together in real time, leaving regulators and operators without a complete picture of true plant vulnerability. Traditional probabilistic risk assessment (PRA) methods treat adversarial events as fixed initiators with predetermined outcomes, and are structurally incapable of representing the time-dependent interplay between physical security events, safety system response, and operator mitigative actions. At Idaho National Laboratory (INL), I contributed to the development and validation of Modeling and Analysis for Safety and Security using the Dynamic EMRALD Framework (MASS-DEF). Where static PRA relies on event-tree logic that cannot evolve mid-scenario, MASS-DEF couples a time-dependent dynamic PRA tool EMRALD (Event Modeling Risk Assessment using Linked Diagrams) with attack simulation software, allowing attacker behavior, plant system states, and operator actions to interact across time. My work focused on validating a general Pressurized Water Reactor (PWR) model. I traced model logic against PWR plant to identified errors in logic and confirm accuracy. I then built and tested attack scenarios against a general PWR model to verify that the model produced expected outcomes across all logical pathways. I also contributed a section to a related technical paper applying the same EMRALD platform to radiation dose modeling. Results show that MASS-DEF can quantitatively demonstrate that many plants exceed their regulatory security thresholds. This demonstrated margin provides a technically defensible basis for reducing the number of guards without compromising regulatory compliance. Physical security costs represent roughly 10% of annual operating budgets, making such reductions directly meaningful to INL's mission of sustaining existing commercial NPPs. This internship strengthened my understanding of nuclear systems, probabilistic modeling, and technical writing, and has solidified my pursuit of a career at a national laboratory.

98 - NUCLEAR DISARMAMENT, SAFEGUARDS, AND PHYSICAL↗
DOE OSTI2026

Performance evaluation of CMIP6 models on the Arctic-Siberian Plain teleconnection affecting the East Asian heat waves

The frequency and intensity of summer heat waves in East Asia have increased sharply in recent decades, significantly impacting public health and the economy. The Arctic-Siberian Plain (ASP) teleconnection pattern has been identified as a key driver, with ASP warming amplifying atmospheric circulation patterns conducive to extreme temperatures. This study evaluates the ability of Coupled Model Inter-comparison Project phase 6 models to simulate the ASP pattern across interannual variability (IAV) and intra-seasonal variability (ISV) timescales using the Common Basis Function method. The multi-model mean shows statistically significant pattern correlations with ERA5 reanalysis, with correlation coefficients of 0.90 and 0.99 for IAV and ISV, respectively. While the ASP pattern is generally well captured, models exhibit substantial inter-model diversity in the intensity and position of anticyclonic anomalies over the ASP and East Asia. Models with ASP pattern variability similar to reanalysis better reproduce extreme East Asian temperatures, whereas those over- or underestimating ASP variability exhibit lower skill. These performance differences are related to differences in simulating key variables associated with the development of the ASP pattern. Our findings highlight the role of the ASP pattern in modulating extreme heat events, as models with improved ASP simulations align more closely with observed temperature extremes. Refining ASP representations in models could enhance seasonal heat wave predictions, improving climate adaptation strategies.

Arctic-Siberian Plain (ASP)↗
DOE OSTI2024

Froggatt-Nielsen meets the SMEFT

We study the matching of Froggatt-Nielsen theories of flavour onto the Standard Model Effective Field Theory (SMEFT), upon integrating out a heavy Beyond-the-Standard-Model (BSM) scalar ‘flavon’ whose vacuum expectation value breaks an Abelian flavour symmetry at energies Λ FN well above the electroweak scale, Λ FN > Λ SM . We include matching contributions to the infrared d SM = 6 (Warsaw basis) SMEFT sourced from ultraviolet contact terms suppressed up to order 1/${\Lambda}_{\textrm{UV}}^2$ in the Froggatt-Nielsen Lagrangian, where Λ UV > Λ FN is an arbitrary ultraviolet scale where further unspecified BSM particles are dynamical. This includes tree-level (one-loop) ultraviolet diagrams with d FN = 6 (5) effective vertices. We first do so with a toy model, but then generalize our findings to arbitrary Frogatt-Nielsen charges. Our results indicate a rich and non-trivial signature of Froggatt-Nielsen theories on the (otherwise) model-independent operators of the SMEFT, and we briefly speculate on extending our analysis to broader classes of BSM flavour models, e.g. non-Abelian and/or gauged theories. We thus take an important step towards determining how to use rapidly developing theoretical and experimental SMEFT technologies to gain unambiguous insight into the SM’s longstanding fermion flavour puzzle.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

A generalized wind turbine cross section as a reduced-order model to gain insights in blade aeroelastic challenges

In this work, we present an approach to study the aeroelastic stability of a wind turbine by focusing on the dynamics of a blade cross section. We present a methodology to obtain a reduced-order model of the blade dynamics in the form of generalized cross-sectional quantities that approximates the aerodynamic and structural properties of the full blade. The motivation for the work is to gain a physical understanding of the influence of aerodynamic models such as dynamic wake and dynamic stall on the frequency and damping of the structure using a reduced-order model with low computational cost. The model may be coupled to two-dimensional computational fluid dynamics softwares or engineering unsteady airfoil aerodynamics models accounting for dynamic wake and dynamic stall. In the latter case, we can obtain monolithic state-space forms of the aeroelastic system of equations, which simplifies the determination of the modal parameters and therefore the study of stability. The work investigates wind turbines in operation or at standstill, where vortex-induced vibrations and stall-induced vibrations, respectively, might be an issue. The implementation is made available as part of the open-source Python package WELIB and as part of the open-source unsteady aerodynamic driver of OpenFAST.

17 WIND ENERGY↗