Search NASA⌕ Search

SEARCH · Search NASA

Results for “Model selection”

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 811 records · Page 45

Accurate, Fast and Scalable Kernel Ridge Regression on Parallel and Distributed Systems

Kernel Ridge Regression (KRR) is a fundamental method in machine learning. Given an n-by-d data matrix as input, a traditional implementation requires Θ(n2) memory to form an n-by-n kernel matrix and Θ(n3) flops to compute the final model. These time and storage costs prohibit KRR from scaling up to large datasets. For example, even on a relatively small dataset (a 520k-by-90 input requiring 357 MB), KRR requires 2 TB memory just to store the kernel matrix. Additionally, the reason is that n usually is much larger than d for real-world applications. On the other hand, weak scaling becomes a problem: if we keep d and n/p fixed as p grows (p is # machines), the memory needed grows as Θ(p) per processor and the flops as Θ(p2) per processor. In the perfect weak scaling situation, both the memory needed and the flops grow as Θ(1) per processor (i.e. memory and flops are constant). The traditional Distributed KRR implementation (DKRR) only achieved 0.32% weak scaling efficiency from 96 to 1536 processors. In this work, we propose two new methods to address these problems: the Balanced KRR (BKRR) and K-means KRR (KKRR). These methods consider alternative ways to partition the input dataset into p different parts, generating p different models, and then selecting the best model among them. Compared to a conventional implementation, KKRR2 (optimized version of KKRR) improves the weak scaling efficiency from 0.32% to 38% and achieves a 591x speedup for getting the same accuracy by using the same data and the same hardware (1536 processors). BKRR2 (optimized version of BKRR) achieves a higher accuracy than the current fastest method using less training time for a variety of datasets. For the applications requiring only approximate solutions, BKRR2 improves the weak scaling efficiency to 92% and achieves 3505x speedup (theoretical speedup: 4096x).

97 MATHEMATICS AND COMPUTING↗

Determining Reduced Order Models for Optimal Stochastic Reduced Order Models

The use of parameterized reduced order models(PROMs) within the stochastic reduced order model (SROM) framework is a logical progression for both methods. In this report, five different parameterized reduced order models are selected and critiqued against the other models along with truth model for the example of the Brake-Reuss beam. The models are: a Taylor series using finite difference, a proper orthogonal decomposition of the the output, a Craig-Bampton representation of the model, a method that uses Hyper-Dual numbers to determine the sensitivities, and a Meta-Model method that uses the Hyper-Dual results and constructs a polynomial curve to better represent the output data. The methods are compared against a parameter sweep and a distribution propagation where the first four statistical moments are used as a comparison. Each method produces very accurate results with the Craig-Bampton reduction having the least accurate results. The models are also compared based on time requirements for the evaluation of each model where the Meta- Model requires the least amount of time for computation by a significant amount. Each of the five models provided accurate results in a reasonable time frame. The determination of which model to use is dependent on the availability of the high-fidelity model and how many evaluations can be performed. Analysis of the output distribution is examined by using a large Monte-Carlo simulation along with a reduced simulation using Latin Hypercube and the stochastic reduced order model sampling technique. Both techniques produced accurate results. The stochastic reduced order modeling technique produced less error when compared to an exhaustive sampling for the majority of methods.

97 MATHEMATICS AND COMPUTING↗

Model Calibration with Markov Chain Monte Carlo Tutorial

The purpose of this tutorial is to demonstrate how to use Markov chain Monte Carlo (MCMC) to calibrate a model. By calibration, we mean the selection of model parameters (and, when relevant, structures). A common goal in model development and diagnostics is calibration, or the identification of model structures and parameters which are consistent with data. While models can be calibrated through hand-tuning parameters or minimizing simple error metrics such as root-mean-square-error (RMSE), these approaches can underrepresent the probabilistic nature of the data-generating process, as well as the potential for multiple model configurations to be consistent with the data. Probabilistic uncertainty quantification, which is the topic of this notebook, can address these concerns. This tutorial is presented as an appendix to the e-book: Addressing Uncertainty in MultiSector Dynamics Research.

Markov chain Monte Carlo↗

Testing experience with the National Transonic Facility

The National Transonic Facility (NTF) was designed for high productivity within the constraints of operation in a cryogenic environment using nitrogen as a test gas. This environment has a significant impact on overall operation. The facility has been operational since August, 1984, and experience has been gained with respect to the facility operation and aerodynamic testing. This paper describes the experience to date with the pretest preparation and testing operations including liquid N2 supply constraints. The instrumentation system is discussed including force, pressure, angle of attack, and model deformation measurements. Selected data from models tested in the NTF, and a status of the static calibration, are also presented.

Fuller, D. E.↗

Large Eddy/Reynolds-Averaged Navier-Stokes Simulations of CUBRC Base Heating Experiments

ven with great advances in computational techniques and computing power during recent decades, the modeling of unsteady separated flows, such as those encountered in the wake of a re-entry vehicle, continues to be one of the most challenging problems in CFD. Of most interest to the aerothermodynamics community is accurately predicting transient heating loads on the base of a blunt body, which would result in reduced uncertainties and safety margins when designing a re-entry vehicle. However, the prediction of heat transfer can vary widely depending on the turbulence model employed. Therefore, selecting a turbulence model which realistically captures as much of the flow physics as possible will result in improved results. Reynolds Averaged Navier Stokes (RANS) models have become increasingly popular due to their good performance with attached flows, and the relatively quick turnaround time to obtain results. However, RANS methods cannot accurately simulate unsteady separated wake flows, and running direct numerical simulation (DNS) on such complex flows is currently too computationally expensive. Large Eddy Simulation (LES) techniques allow for the computation of the large eddies, which contain most of the Reynolds stress, while modeling the smaller (subgrid) eddies. This results in models which are more computationally expensive than RANS methods, but not as prohibitive as DNS. By complimenting an LES approach with a RANS model, a hybrid LES/RANS method resolves the larger turbulent scales away from surfaces with LES, and switches to a RANS model inside boundary layers. As pointed out by Bertin et al., this type of hybrid approach has shown a lot of promise for predicting turbulent flows, but work is needed to verify that these models work well in hypersonic flows. The very limited amounts of flight and experimental data available presents an additional challenge for researchers. Recently, a joint study by NASA and CUBRC has focused on collecting heat transfer data on the backshell of a scaled model of the Orion Multi-Purpose Crew Vehicle (MPCV). Heat augmentation effects due to the presence of cavities and RCS jet firings were also investigated. The high quality data produced by this effort presents a new set of data which can be used to assess the performance of CFD methods. In this work, a hybrid LES/RANS model developed at North Carolina State University (NCSU) is used to simulate several runs from these experiments, and evaluate the performance of high fidelity methods as compared to more typical RANS models. .

Salazar, Giovanni↗

Future Sea Level Change Under CMIP5 and CMIP6 Scenarios From the Greenland and Antarctic Ice Sheets

Projections of the sea level contribution from the Greenland and Antarctic ice sheets (GrIS and AIS) rely on atmospheric and oceanic drivers obtained from climate models. The Earth System Models participating in the Coupled Model Intercomparison Project phase 6 (CMIP6) generally project greater future warming compared with the previous Coupled Model Intercomparison Project phase 5 (CMIP5) effort. Here we use four CMIP6 models and a selection of CMIP5 models to force multiple ice sheet models as part of the Ice Sheet Model Intercomparison Project for CMIP6 (ISMIP6). We find that the projected sea level contribution at 2100 from the ice sheet model ensemble under the CMIP6 scenarios falls within the CMIP5 range for the Antarctic ice sheet but is significantly increased for Greenland. Warmer atmosphere in CMIP6 models results in higher Greenland mass loss due to surface melt. For Antarctica, CMIP6 forcing is similar to CMIP5 and mass gain from increased snowfall counteracts increased loss due to ocean warming.

Antarctica↗

Comparisons of Mixing Efficiency for the Strut Fuel Injector Obtained from Large-Eddy and Reynolds-Averaged Simulations, and Experiments

Mixing efficiency is obtained for a strut fuel injector at hypervelocity flow conditions by using large-eddy simulations (LES), Reynolds-averaged simulations (RAS), and experiments. The injector and flow conditions have been previously investigated by using RAS and experiments as a part of the Enhanced Injection and Mixing Project (EIMP) at the NASA Langley Research Center (LaRC). Because the fidelity of LES is a strong function of the grid, the mixing efficiency is obtained on two grids, the coarser of which is a factor of two coarser in each of the three dimensions with respect to the fine grid. The RAS uses the two-equation linear eddy viscosity and diffusivity modeling of Menter. In RAS, the species diffusivity model exhibits a strong dependence on the turbulent Schmidt number, which is often adjusted until some metric of engineering interest, such as the mixing efficiency, matches the experimental data. In the absence of experimental data, scale-resolving simulations, such as LES, have been proposed as surrogates for experiments that could provide the data needed to “calibrate” the turbulent Schmidt number in the RAS models. This approach is followed because LES requires significantly more computational resources (CPU, data storage, and time) than RAS, making it prohibitive for use in many engineering applications and specifically for parameter exploration or optimization. Here we examine the mixing efficiency obtained from several RAS with different values of the turbulent Schmidt number, and compare the results with those obtained from the LES and experiments. In addition, the least squares fitting approach was used to demonstrate how to obtain an estimate for the turbulent Schmidt number from LES analytically. These estimates were then used together with prior knowledge about RAS model sensitivity to select a turbulence model that was expected to best match the LES data.

LES↗

Comparisons of Mixing Efficiency for the Strut Fuel Injector Obtained from Large-Eddy and Reynolds-Averaged Simulations, and Experiments

Mixing efficiency is obtained for a strut fuel injector at hypervelocity flow conditions by using large-eddy simulations (LES), Reynolds-averaged simulations (RAS), and experiments. The injector and flow conditions have been previously investigated by using RAS and experiments as a part of the Enhanced Injection and Mixing Project (EIMP) at the NASA Langley Research Center (LaRC). Because the fidelity of LES is a strong function of the grid, the mixing efficiency is obtained on two grids, the coarser of which is a factor of two coarser in each of the three dimensions with respect to the fine grid. The RAS uses the two-equation linear eddy viscosity and diffusivity modeling of Menter. In RAS, the species diffusivity model exhibits a strong dependence on the turbulent Schmidt number, which is often adjusted until some metric of engineering interest, such as the mixing efficiency, matches the experimental data. In the absence of experimental data, scale-resolving simulations, such as LES, have been proposed as surrogates for experiments that could provide the data needed to “calibrate” the turbulent Schmidt number in the RAS models. This approach is followed because LES requires significantly more computational resources (CPU, data storage, and time) than RAS, making it prohibitive for use in many engineering applications and specifically for parameter exploration or optimization. Here we examine the mixing efficiency obtained from several RAS with different values of the turbulent Schmidt number, and compare the results with those obtained from the LES and experiments. In addition, the least squares fitting approach was used to demonstrate how to obtain an estimate for the turbulent Schmidt number from LES analytically. These estimates were then used together with prior knowledge about RAS model sensitivity to select a turbulence model that was expected to best match the LES data.

LES↗

MODELING TRANSPORT IN THE DOWN GRADIENT PORTION OF THE 200-PO-1 OPERABLE UNIT AT THE HANFORD SITE

Remedial Investigations are underway for the 200-PO-l Operable Unit (OU) at the U.S. Department of Energy's Hanford Site in Washington State. To support the baseline risk assessment and evaluation of remedial alternatives, fate and transport modeling is being conducted to predict the future concentration of contaminants of potential concern in the 200-PO-1 OU. This study focuses on modeling the 'down gradient' transport of those contaminants that migrate beyond the 3-D model domain selected for performing detailed 'source area' modeling within the 200-PO-1 OU. The down gradient portion is defined as that region of the 200-PO-1 OU that is generally outside the 200 Area (considered 'source area') of the Hanford Site. A 1-D transport model is developed for performing down gradient contaminant fate and transport modeling. The 1-D transport model is deemed adequate based on the inferred transport pathway of tritium in the past and the observation that most of the contaminant mass remains at or near the water table within the unconfined aquifer of the Hanford Formation and the Cold-Creek/Pre-Missoula Gravel unit. The Pipe Pathway feature of the GoldSim software is used to perform the calculations. The Pipe Pathway uses a Laplace transform approach to provide analytical solutions to a broad range of advection-dominated mass transport systems involving one-dimensional advection, longitudinal dispersion, retardation, decay and ingrowth, and exchanges with immobile storage zones. Based on the historical concentration distribution data for the extensive tritium plume in this area, three Pipe Pathways are deemed adequate for modeling transport of contaminants. Each of these three Pipe Pathways is discretized into several zones, based on the saturated thickness variation in the unconfined aquifer and the location of monitoring wells used for risk assessment calculation. The mass fluxes of contaminants predicted to exit the source area model domain are used as an input to the down gradient model, while the flow velocities applied are based on the present-day hydraulic gradients and estimation of hydraulic conductivity in the unconfined aquifer. The results of the calculation indicate that the future concentrations of contaminants of potential concern in the down gradient portion of the 200-PO-1 OU declines with time and distance.

54 ENVIRONMENTAL SCIENCES↗

Diagnosing Model Errors in Simulation of Solar Radiation on Inclined Surfaces

Transposition models have been widely used in the solar energy industry to simulate solar radiation on inclined PV panels. Following numerous studies comparing the performance of transposition models, this paper aims to understand the quantitative uncertainty in the state-of-the-art transposition models and the sources leading to the uncertainty. Our results show significant differences between two highly used isotropic transposition models with one substantially underestimating the diffuse plane-of-array (POA) irradiances when diffuse radiation is perfectly isotropic. In the empirical transposition models, the selection of empirical coefficients and land surface albedo can both result in uncertainty in the output. This study can be used as a guide for future development of physics-based transposition models.

14 SOLAR ENERGY↗

Material Selection and Heat Transfer Model for PTUHS Device

The Path to Ultimate Heat Sink Device (PTUHS) is a passive safety system that utilizes a radiative heat transfer valve to control the rate at which heat is removed from a nuclear reactor pressure vessel (RPV) and dispersed into surrounding soil. The material selection for the PTUHS device is investigated, where the thermophysical properties are either maximized or minimized bases on what section of the PTUHS that material is being used for. Final recommendations of material choice are then presented. The heat transfer model of the PTUHS device is solved in ABAQUS, where a normal operating conditions and SCRAM conditions are solved. These temperature maps and heat flux of the system show the system's ability, where minimal heat is lost during normal operating conditions and system failure in a SCRAM scenario due to temperatures that were reached.

42 - ENGINEERING↗

Critical considerations for wind-tunnel testing V/STOL aircraft models

Low-speed wind-tunnel testing of V/STOL aircraft concepts to determine the aerodynamic-propulsion interaction effects during the transition between hover and wingborne flight is a necessary step in the development cycle of this type of aircraft. The paper examines factors which must be dealt with to assure that the information obtained in experiments is accurate and representative of the full-scale aircraft modeled. Areas of discussion include: (1) Proper engine simulation selection and model size constraints as a result of the selection; (2) selection of reference power-off condition for determining propulsion interaction effects; (3) modeling of realistic flight conditions to obtain a comprehensive evaluation of the entire transition performance; and (4) wind-tunnel boundary interferences and how these are highly configuration dependent.

Margason, R. J.↗

Rift Valley Fever Risk Map Model and Seroprevalence in Selected Wild Ungulates and Camels from Kenya

Since the first isolation of Rift Valley fever virus (RVFV) in the 1930s, there have been multiple epizootics and epidemics in animals and humans in sub-Saharan Africa. Prospective climate-based models have recently been developed that flag areas at risk of RVFV transmission in endemic regions based on key environmental indicators that precede Rift Valley fever (RVF) epizootics and epidemics. Although the timing and locations of human case data from the 2006-2007 RVF outbreak in Kenya have been compared to risk zones flagged by the model, seroprevalence of RVF antibodies in wildlife has not yet been analyzed in light of temporal and spatial predictions of RVF activity. Primarily wild ungulate serum samples from periods before, during, and after the 2006-2007 RVF epizootic were analyzed for the presence of RVFV IgM and/or IgG antibody. Results show an increase in RVF seropositivity from samples collected in 2007 (31.8%), compared to antibody prevalence observed from 2000-2006 (3.3%). After the epizootic, average RVF seropositivity diminished to 5% in samples collected from 2008-2009. Overlaying maps of modeled RVF risk assessments with sampling locations indicated positive RVF serology in several species of wild ungulate in or near areas flagged as being at risk for RVF. Our results establish the need to continue and expand sero-surveillance of wildlife species Kenya and elsewhere in the Horn of Africa to further calibrate and improve the RVF risk model, and better understand the dynamics of RVFV transmission.

Seroprevalence↗

The optimization of model ensemble composition and size can enhance the robustness of crop yield projections

Linked climate and crop simulation models are widely used to assess the impact of climate change on agriculture. However, it is unclear how ensemble configurations (model composition and size) influence crop yield projections and uncertainty. Here, we investigate the influences of ensemble configurations on crop yield projections and modeling uncertainty from Global Gridded Crop Models and Global Climate Models under future climate change. We performed a cluster analysis to identify distinct groups of ensemble members based on their projected outcomes, revealing unique patterns in crop yield projections and corresponding uncertainty levels, particularly for wheat and soybean. Furthermore, our findings suggest that approximately six Global Gridded Crop Models and 10 Global Climate Models are sufficient to capture modeling uncertainty, while a cluster-based selection of 3-4 Global Gridded Crop Models effectively represents the full ensemble. The contribution of individual Global Gridded Crop Models to overall uncertainty varies depending on region and crop type, emphasizing the importance of considering the impact of specific models when selecting models for local-scale applications. Our results emphasize the importance of model composition and ensemble size in identifying the primary sources of uncertainty in crop yield projections, offering valuable guidance for optimizing ensemble configurations in climate-crop modeling studies tailored to specific applications.

Agriculture↗

THE SYSTEMATICS OF STRONG LENS MODELING QUANTIFIED: THE EFFECTS OF CONSTRAINT SELECTION AND REDSHIFT INFORMATION ON MAGNIFICATION, MASS, AND MULTIPLE IMAGE PREDICTABILITY

Until now, systematic errors in strong gravitational lens modeling have been acknowledged but have never been fully quantified. Here, we launch an investigation into the systematics induced by constraint selection. We model the simulated cluster Ares 362 times using random selections of image systems with and without spectroscopic redshifts and quantify the systematics using several diagnostics: image predictability, accuracy of model-predicted redshifts, enclosed mass, and magnification. We find that for models with >15 image systems, the image plane rms does not decrease significantly when more systems are added; however, the rms values quoted in the literature may be misleading as to the ability of a model to predict new multiple images. The mass is well constrained near the Einstein radius in all cases, and systematic error drops to <2% for models using >10 image systems. Magnification errors are smallest along the straight portions of the critical curve, and the value of the magnification is systematically lower near curved portions. For >15 systems, the systematic error on magnification is ∼2%. We report no trend in magnification error with the fraction of spectroscopic image systems when selecting constraints at random; however, when using the same selection of constraints, increasing this fraction up to ∼0.5 will increase model accuracy. The results suggest that the selection of constraints, rather than quantity alone, determines the accuracy of the magnification. We note that spectroscopic follow-up of at least a few image systems is crucial because models without any spectroscopic redshifts are inaccurate across all of our diagnostics.

79 ASTRONOMY AND ASTROPHYSICS↗

Model atmosphere analysis of selected luminous B stars

The general scientific goal of this program has been to determine whether the atmospheric structure of the B-type stars can be represented by the current generation of plane parallel, line-blanketed, LTE stellar atmosphere models sufficiently well to allow accurate effective temperatures and surface gravities to be deduced. The B stars cover a wide range of temperature and luminosity. For the hottest such stars (with T approximately 30,000 K) the applicability of the models may be compromised by departures from LTE in the stellar atmospheres ('non-LTE effects'). At the highest luminosities (the B 'super giants'), the models may be invalidated by departures from plane parallel geometry. Thus we seek to identify the temperature and luminosity range within which these effects are unimportant and where the models may be relied upon.

Fitzpatrick, Edward L.↗

Stimulation of a turbofan engine for evaluation of multivariable optimal control concepts

The development of control systems for jet engines requires a real-time computer simulation. The simulation provides an effective tool for evaluating control concepts and problem areas prior to actual engine testing. The development and use of a real-time simulation of the Pratt and Whitney F100-PW100 turbofan engine is described. The simulation was used in a multi-variable optimal controls research program using linear quadratic regulator theory. The simulation is used to generate linear engine models at selected operating points and evaluate the control algorithm. To reduce the complexity of the design, it is desirable to reduce the order of the linear model. A technique to reduce the order of the model; is discussed. Selected results between high and low order models are compared. The LQR control algorithms can be programmed on digital computer. This computer will control the engine simulation over the desired flight envelope.

Seldner, K.↗

Minimal realizations of Dirac neutrino mass from generic one-loop and two-loop topologies at d =5

We carry out a systematic investigation for the minimal Dirac neutrino mass models emerging from generic one-loop and two-loop topologies that arise from d = 5 effective operator with a singlet scalar, σ. To ensure that the tree-level Dirac mass, as well as Majorana mass terms at all orders, are absent for the neutrinos, we work in the framework where the Standard Model is supplemented by the well-motivated U(1)B-L gauge symmetry. At the one-loop level, we analyze six possible topologies, out of which two of them have the potential to generate desired Dirac neutrino mass. Adopting a systematic approach to select minimal models, we construct seventeen viable one-loop Dirac neutrino mass models. By embracing a similar methodical approach at the two-loop, we work out twenty-three minimal candidates. Among the forty selected economical models, the majority of the models proposed in this work are new. In our search, we also include the scenarios where the particles in the loop carry charges under the color group. Furthermore, we discuss the possible dark matter candidates within a given model, if any, without extending the minimal particle content.

79 ASTRONOMY AND ASTROPHYSICS↗