Search NASA⌕ Search

SEARCH · Search NASA

Results for “Bayesian Testing”

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 613 records · Page 34

GEOMETRIC AND DYNAMICAL MODELS OF REVERBERATION MAPPING DATA

We present a general method to analyze reverberation (or echo) mapping data that simultaneously provides estimates for the black hole mass and for the geometry and dynamics of the broad-line region (BLR) in active galactic nuclei (AGNs). While previous methods yield a typical scale size of the BLR or a reconstruction of the transfer function, our method directly infers the spatial and velocity distribution of the BLR from the data, from which a transfer function can be easily derived. Previous echo mapping analysis requires an independent estimate of a scaling factor known as the virial coefficient to infer the mass of the black hole, but this is not needed in our more direct approach. We use the formalism of Bayesian probability theory and implement a Markov Chain Monte Carlo algorithm to obtain estimates and uncertainties for the parameters of our BLR models. Fitting of models to the data requires knowledge of the continuum flux at all times, not just the measured times. We use Gaussian Processes to interpolate and extrapolate the continuum light curve data in a fully consistent probabilistic manner, taking the associated errors into account. We illustrate our method using simple models of BLR geometry and dynamics and show that we can recover the parameter values of our test systems with realistic uncertainties that depend upon the variability of the AGN and the quality of the reverberation mapping observing campaign. With a geometry model we can recover the mean radius of the BLR to within {approx}0.1 dex random uncertainty for simulated data with an integrated line flux uncertainty of 1.5%, while with a dynamical model we can recover the black hole mass and the mean radius to within {approx}0.05 dex random uncertainty, for simulated data with a line profile average signal-to-noise ratio of 4 per spectral pixel. These uncertainties do not include modeling errors, which are likely to be present in the analysis of real data, and should therefore be considered as lower limits to the accuracy of the method.

79 ASTRONOMY AND ASTROPHYSICS↗

Bayesian Spectral Moment Estimation and Uncertainty Quantification

We present a Bayesian spectral fitting method developed for spectroscopic data analysis, particularly (but not solely) in the context of fusion energy research. The presented techniques are particularly valuable to estimating moments and corresponding uncertainties whenever the spectra result from line-integrated measurements in nonuniform plasmas, for which the approximation of atomic line shapes being ideal Gaussians gives poor estimates. We decompose multiple, potentially overlapping spectral lines into a sum of Gauss-Hermite polynomials, whose properties allow efficient truncation and uncertainty quantification, often with only three free parameters per atomic emission line. Tests with both synthetic and experimental data demonstrate the effectiveness and robustness where more standard nonlinear fitting routines may experience difficulties. A parallelized version of our implementation is publicly released under an open source license.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Calculating system reliability with SRFYDO

SRFYDO is a process for estimating reliability of complex systems. Using information from all applicable sources, including full-system (flight) data, component test data, and expert (engineering) judgment, SRFYDO produces reliability estimates and predictions. It is appropriate for series systems with possibly several versions of the system which share some common components. It models reliability as a function of age and up to 2 other lifecycle (usage) covariates. Initial output from its Exploratory Data Analysis mode consists of plots and numerical summaries so that the user can check data entry and model assumptions, and help determine a final form for the system model. The System Reliability mode runs a complete reliability calculation using Bayesian methodology. This mode produces results that estimate reliability at the component, sub-system, and system level. The results include estimates of uncertainty, and can predict reliability at some not-too-distant time in the future. This paper presents an overview of the underlying statistical model for the analysis, discusses model assumptions, and demonstrates usage of SRFYDO.

97 MATHEMATICS AND COMPUTING↗

A Multifidelity Ensemble Kalman Filter with Reduced Order Control Variates

This work develops a new multi delity ensemble Kalman lter (MFEnKF) algorithm based on linear control variate framework. The approach allows for rigorous multi delity extensions of the EnKF, where the uncertainty in coarser delities in the hierarchy of models represent control variates for the uncertainty in ner delities. Small ensembles of high delity model runs are com- plemented by larger ensembles of cheaper, lower delity runs, to obtain much improved analyses at only small additional computational costs. We investigate the use of reduced order models as coarse delity control variates in the MFEnKF, and provide analyses to quantify the improvements over the traditional ensemble Kalman lters. We apply these ideas to perform data assimilation with a quasi-geostrophic test problem, using direct numerical simulation and a corresponding POD-Galerkin reduced order model. Numerical results show that the two- delity MFEnKF provides better analyses than existing EnKF algorithms at comparable or reduced computational costs.

97 MATHEMATICS AND COMPUTING↗

A Bayesian approach to evaluation of soil biogeochemical models

Abstract. To make predictions about the carbon cycling consequences of rising global surface temperatures, Earth system scientists rely on mathematical soil biogeochemical models (SBMs). However, it is not clear which models have better predictive accuracy, and a rigorous quantitative approach for comparing and validating the predictions has yet to be established. In this study, we present a Bayesian approach to SBM comparison that can be incorporated into a statistical model selection framework. We compared the fits of linear and nonlinear SBMs to soil respiration data compiled in a recent meta-analysis of soil warming field experiments. Fit quality was quantified using Bayesian goodness-of-fit metrics, including the widely applicable information criterion (WAIC) and leave-one-out cross validation (LOO). We found that the linear model generally outperformed the nonlinear model at fitting the meta-analysis data set. Both WAIC and LOO computed higher overfitting risk and effective numbers of parameters for the nonlinear model compared to the linear model, conditional on the data set. Goodness of fit for both models generally improved when they were initialized with lower and more realistic steady-state soil organic carbon densities. Still, testing whether linear models offer definitively superior predictive performance over nonlinear models on a global scale will require comparisons with additional site-specific data sets of suitable size and dimensionality. Such comparisons can build upon the approach defined in this study to make more rigorous statistical determinations about model accuracy while leveraging emerging data sets, such as those from long-term ecological research experiments.

54 ENVIRONMENTAL SCIENCES↗

The Langmuir probe system in the Wendelstein 7-X test divertor

Here, the design and evaluation of the Langmuir probe system used in the first divertor operation phase of Wendelstein 7-X is described. The probes are integrated into the target plates and have individually facetted surfaces to keep the angle of incidence of the magnetic field within an appropriate range for different magnetic configurations. Multiple models for the derivation of plasma parameters from current–voltage characteristics are introduced. These are analyzed with regard to their assumptions and limitations, generalized, and adapted to our use case. A detailed comparison is made to determine the most suitable model. It is found that the choice of model has a large impact, for example, resulting in a change in the inferred temperatures of up to a factor two. This evaluation is implemented in a Bayesian modeling framework and automated to allow for joint analysis with other diagnostics and a replacement of ad hoc assumptions. We rigorously treat parameter uncertainties, revealing strong correlations between them. General and flexible model formulations permit an expansion to additional effects.

47 OTHER INSTRUMENTATION↗

Robustness of Deep Learning Classification to Adversarial Input on GPUs: Asynchronous Parallel Accumulation Is a Source of Vulnerability

The ability of machine learning (ML) classification models to resist small, targeted input perturbations—known as adversarial attacks—is a key measure of their safety and reliability. We show that floating-point non associativity (FPNA) coupled with asynchronous parallel programming on GPUs is sufficient to result in misclassification, without any perturbation to the input. Additionally, we show that this misclassification is particularly significant for inputs close to the decision boundary and that standard adversarial robustness results may be overestimated up to 4.6 when not considering machine-level details. We first study a linear classifier, before focusing on standard Graph Neural Network (GNN) architectures and datasets used in robustness assessments. We develop a novel black-box attack using Bayesian optimization to discover external workloads that can change the instruction scheduling which bias the output of reductions on GPUs and reliably lead to misclassification. Motivated by these results, we present a new learnable permutation (LP) gradient-based approach to learning floating-point operation orderings that lead to misclassifications. The LP approach provides a worst-case estimate in a computationally efficient manner, avoiding the need to run identical experiments tens of thousands of times over a potentially large set of possible GPU states or architectures. Finally, using instrumentation-based testing, we investigate parallel reduction ordering across different GPU architectures under external background workloads, when utilizing multi-GPU virtualization, and when applying power capping. Our results demonstrate that parallel reduction ordering varies significantly across architectures under the first two conditions, substantially increasing the search space required to fully test the effects of this parallel scheduler-based vulnerability. These results and the methods developed here can help to include machine-level considerations into adversarial robustness assessments, which can make a difference in safety and mission critical applications.

Shanmugavelu, Sanjif [Maxeler Technologies, a Groq↗

Fossil Fuel Emission Verification Modeling at LLNL

We have an established project at LLNL to develop the tools needed to constrain fossil fuel carbon dioxide emissions using measurements of the carbon-14 isotope in atmospheric samples. In Figure 1 we show the fossil fuel plumes from Los Angeles and San Francisco for two different weather patterns. Obviously, a measurement made at any given location is going to depend on the weather leading up to the measurement. Thus, in order to determine the GHG emissions from some region using in situ measurements of those GHGs, we use state-of-the-art global and regional atmospheric chemistry-transport codes to simulate the plumes: the LLNL-IMPACT model (Rotman et al., 2004) and the WRFCHEM community code (http://www.wrf-model.org/index.php). Both codes can use observed (aka assimilated) meteorology in order to recreate the actual transport that occurred. The measured concentration of each tracer at a particular spatio-temporal location is a linear combination of the plumes from each region at that location (for non-reactive species). The challenge is to calculate the emission strengths for each region that fit the observed concentrations. In general this is difficult because there are errors in the measurements and modeling of the plumes. We solve this inversion problem using the strategy illustrated in Figure 2. The Bayesian Inference step combines the a priori estimates of the emissions, and their uncertainty, for each region with the results of the observations, and their uncertainty, and an ensemble of model predicted plumes for each region, and their uncertainty. The result is the mathematical best estimate of the emissions and their errors. In the case of non-linearities, or if we are using a statistical sampling technique such as a Markov Chain Monte Carlo technique, then the process is iterated until it converges (ie reaches stationarity). For the Bayesian inference we can use both a direct inversion capability, which is fast but requires assumptions of linearity and Gaussianity of errors, or one of several statistical sampling techniques, which are computationally slower but do not require either linearity or Gaussianity (Chow, et al., 2008; Delle Monache, et al., 2008). The emission regions we are using are based on the air-basins defined by the California Air Resources Board (CARB), see Figure 3. The only difference is that we have joined some of the smaller air basins together. The results of a test using 4 days of simulated observations using our ensemble retrieval system are shown in Figure 3 (right). The main source of the variation between the different model configurations arises from the uncertainty in the atmospheric boundary layer parameterization in the WRF model. We are currently developing a capability to constrain the boundary layer height in our carbon-14 work either by weighting the ensemble member results by the accuracy of their boundary layer height (using commercial aircraft observations), or as part of the retrieval process using an ensemble Kalman filter (EnKF) capability.

54 ENVIRONMENTAL SCIENCES↗

Probabilistic constrained Bayesian inversion for transpiration cooling

Abstract To enable safe operations in applications such as rocket combustion chambers, the materials require cooling to avoid material damage. Here, transpiration cooling is a promising cooling technique. Numerous studies investigate possibilities to simulate and evaluate the complex cooling mechanism. One naturally arising question is the amount of coolant required to ensure a safe operation. To study this, we introduce an approach that determines the posterior probability distribution of the Reynolds number using an inverse problem and constraining the maximum temperature of the system under parameter uncertainties. Mathematically, this chance inequality constraint is dealt with by a generalized polynomial chaos expansion of the system. The posterior distribution will be evaluated by different Markov chain Monte Carlo based methods. A novel method for the constrained case is proposed and tested among others on two‐dimensional transpiration cooling models.

97 MATHEMATICS AND COMPUTING↗

A BAYESIAN APPROACH TO COMPARING COSMIC RAY ENERGY SPECTRA

A common problem in ultra-high energy cosmic ray physics is the comparison of energy spectra. The question is whether the spectra from two experiments or two regions of the sky agree within their statistical and systematic uncertainties. We develop a method to directly compare energy spectra for ultra-high energy cosmic rays from two different regions of the sky in the same experiment without reliance on agreement with a theoretical model of the energy spectra. The consistency between the two spectra is expressed in terms of a Bayes factor, defined here as the ratio of the likelihood of the two-parent source hypothesis to the likelihood of the one-parent source hypothesis. Unlike other methods, for example {chi}{sup 2} tests, the Bayes factor allows for the calculation of the posterior odds ratio and correctly accounts for non-Gaussian uncertainties. The latter is particularly important at the highest energies, where the number of events is very small.

79 ASTRONOMY AND ASTROPHYSICS↗

Galaxy zoo builder: Morphological dependence of spiral galaxy pitch angle

ABSTRACT Spiral structure is ubiquitous in the Universe, and the pitch angle of arms in spiral galaxies provide an important observable in efforts to discriminate between different mechanisms of spiral arm formation and evolution. In this paper, we present a hierarchical Bayesian approach to galaxy pitch angle determination, using spiral arm data obtained through the Galaxy Builder citizen science project. We present a new approach to deal with the large variations in pitch angle between different arms in a single galaxy, which obtains full posterior distributions on parameters. We make use of our pitch angles to examine previously reported links between bulge and bar strength and pitch angle, finding no correlation in our data (with a caveat that we use observational proxies for both bulge size and bar strength which differ from other work). We test a recent model for spiral arm winding, which predicts uniformity of the cotangent of pitch angle between some unknown upper and lower limits, finding our observations are consistent with this model of transient and recurrent spiral pitch angle as long as the pitch angle at which most winding spirals dissipate or disappear is larger than 10°.

Lingard, Timothy↗

Solid-state lighting life prediction using extended Kalman filter

Solid-state lighting (SSL) luminaires containing light emitting diodes (LEDs) have the potential of seeing excessive temperatures when being transported across country or being stored in non-climate controlled warehouses. They are also being used in outdoor applications in desert environments that see little or no humidity but will experience extremely high temperatures during the day. This makes it important to increase our understanding of what effects high temperature exposure for a prolonged period of time will have on the usability and survivability of these devices. The U.S. Department of Energy has made a long term commitment to advance the efficiency, understanding and development of solid-state lighting (SSL) and is making a strong push for the acceptance and use of SSL products to reduce overall energy consumption attributable to lighting. Traditional light sources “burn out” at end-of-life. For an incandescent bulb, the lamp life is defined by B50 life. However, the LEDs have no filament to “burn”. The LEDs continually degrade and the light output decreases eventually below useful levels causing failure. Presently, the TM-21 test standard is used to predict the L70 life of SSL Luminaires from LM-80 test data. The TM-21 model uses an Arrhenius Equation with an Activation Energy, Pre-decay factor and Decay Rates. Several failure mechanisms may be active in a luminaire at a single time causing lumen depreciation. The underlying TM-21 Arrhenius Model may not capture the failure physics in presence of multiple failure mechanisms. Correlation of lumen maintenance with underlying physics of degradation at system-level is needed. In this paper, a Kalman Filter and Extended Kalman Filters have been used to develop a 70% Lumen Maintenance Life Prediction Model for a LEDs used in SSL luminaires. This model can be used to calculate acceleration factors, evaluate failure-probability and identify ALT methodologies for reducing test time. Ten-thousand hour LM-80 test data for various LEDs have been used for model development. System state has been described in state space form using the measurement of the feature vector, velocity of feature vector change and the acceleration of the feature vector change. System state at each future time has been computed based on the state space at preceding time step, system dynamics matrix, control vector, control matrix, measurement matrix, measured vector, process noise and measurement noise. The future state of the lumen depreciation has been estimated based on a second order Kalman Filter model and a Bayesian Framework. The measured state variable has been related to the underlying damage using physics-based models. Life prediction of L70 life for the LEDs used in SSL luminaires from KF and EKF based models have been compared with the TM-21 model predictions and experimental data.

54 ENVIRONMENTAL SCIENCES↗

Evaluating the Potential and Challenges of an Uncertainty Quantification Method for Long Short–Term Memory Models for Soil Moisture Predictions

Recently, recurrent deep networks have shown promise to harness newly available satellite–sensed data for long–term soil moisture projections. However, to be useful in forecasting, deep networks must also provide uncertainty estimates. Here we evaluated Monte Carlo dropout with an input–dependent data noise term (MCD+N), an efficient uncertainty estimation framework originally developed in computer vision, for hydrologic time series predictions. MCD+N simultaneously estimates a heteroscedastic input–dependent data noise term (a trained error model attributable to observational noise) and a network weight uncertainty term (attributable to insufficiently constrained model parameters). Although MCD+N has appealing features, many heuristic approximations were employed during its derivation, and rigorous evaluations and evidence of its asserted capability to detect dissimilarity were lacking. To address this, we provided an in–depth evaluation of the scheme's potential and limitations. We showed that for reproducing soil moisture dynamics recorded by the Soil Moisture Active Passive (SMAP) mission, MCD+N indeed gave a good estimate of predictive error, provided that we tuned a hyperparameter and used a representative training data set. The input–dependent term responded strongly to observational noise, while the model term clearly acted as a detector for physiographic dissimilarity from the training data, behaving as intended. However, when the training and test data were characteristically different, the input–dependent term could be misled, undermining its reliability. Additionally, due to the data–driven nature of the model, data noise also influences network weight uncertainty, and therefore the two uncertainty terms are correlated. Altogether, this approach has promise, but care is needed to interpret the results.

54 ENVIRONMENTAL SCIENCES↗

Field testing and validation of a low-cost MPC for demand flexibility for grid-interactive K-12 schools

K-12 school buildings account for the highest energy consumption within the public sector. Implementing advanced HVAC controls in grid-interactive K-12 schools could bring substantial economic advantages and grid flexibility. Our previous study demonstrated that a low-cost model predictive control (MPC) solution, which coordinates multiple packaged units, can enable demand flexibility without major hardware upgrades. However, a significant gap remains between academic pilots and market-ready scalable solutions. This paper extends the previous single-site pilot to a multi-site demonstration involving three school campuses (95 total units) through a commercial technology transfer process. Addressing the challenge of verifying performance with sparse field data, we present a new statistical approach using Bayesian methods to estimate the MPC’s effect on peak demand. Unlike traditional methods, this approach robustly quantifies uncertainty in non-normal, limited datasets. The results confirm the solution’s replicability, achieving a 21.6–38.9% reduction in HVAC peak demand (10.8–22.1% at the site-level) with > 98% probability across diverse locations. Finally, we document critical barriers to scaling software-as-a-service (SaaS) solutions–such as API instability and diverse legacy systems–and offer practical strategies to accelerate the commercial adoption of grid-interactive efficient buildings.

Ham, Sang Woo↗

The look-elsewhere effect from a unified Bayesian and frequentist perspective

When searching over a large parameter space for anomalies such as events, peaks, objects, or particles, there is a large probability that spurious signals with seemingly high signi ficance will be found. This is known as the look-elsewhere effect and is prevalent throughout cosmology, (astro)particle physics, and beyond. To avoid making false claims of detection, one must account for this effect when assigning the statistical significance of an anomaly. This is typically accomplished by considering the trials factor, which is generally computed numerically via potentially expensive simulations. In this paper we develop a continuous generalization of the Bonferroni and Sidak corrections by applying the Laplace approximation to evaluate the Bayes factor, and in turn relating the trials factor to the prior-to-posterior volume ratio. Here, we use this to define a test statistic whose frequentist properties have a simple interpretation in terms of the global p-value, or statistical significance. We apply this method to various physics-based examples and show it to work well for the full range of p-values, i.e. in both the asymptotic and non-asymptotic regimes. We also show that this method naturally accounts for other model complexities such as additional degrees of freedom, generalizing Wilks' theorem. This provides a fast way to quantify statistical significance in light of the look-elsewhere effect, without resorting to expensive simulations.

79 ASTRONOMY AND ASTROPHYSICS↗

Search for Heavy Resonances in the Dimuon Channel with the DØ Detector

High-energy particle physics is often described as the quest for a fundamental theory. One can think of one theory as being more "fundamental" than a second theory if the first explains some or all of the predictions of the second in terms of processes at smaller distance scales and higher energies. Another sense in which a theory can be thought of as being more fundamental than another, is if the first theory "explains" the second; that is, if some arbitrary aspect of the second theory is a special case of a more general principle in the first theory. The Standard Model of particle physics is the most fundamental experimentally tested theory in either sense. It was developed in its final form in the 1960's and 1970's and continues to successfully predict the outcomes of high-energy physics experiments; the research presented in this thesis offers no exception to this trend ( unfortunately). However, the Standard Model only makes sense as an effective theory; that is, a theory that is not actually fundamental, but which is a low-energy limit of some underlying theory, without being dependent on the details of that underlying theory. Moreover, one important aspect of the Standard Model so far remains unexplained, namely electroweak symmetry breaking. Phase transitions occur in many places in physics, and can often be explained as a low-energy aspect of some more fundamental process. However, there is no such underlying process for electroweak symmetry breaking within the Standard Model; the introduction of the scalar Higgs field does not, in itself, explain electroweak symmetry breaking, it merely parameterizes it. Therefore, it is perhaps not surprising that the (most obvious) problem with the Standard Model, the '·gauge hierarchy problem", occurs in relation with the Higgs field. This problem, the "unnaturalness" of the Standard Model, will be discussed (briefly) in the next chapter. It so happens that the unnaturalness of the Standard Model strongly suggests the presence of new physics at an energy scale that is very close to the energies achieved with the Tevatron collider (and presumably well within the reach of the next collider, the LHC). Therefore, there is a general expectation that new physics is "just around the corner". Many, many models for new physics exist, and it is not possible to devise an optimal search strategy for all of them at once. Therefore (and possibly to reduce bias), it is preferable to search for new physics in a model-independent way. While the analysis presented in this thesis is certainly not model-independent, it uses the fact that many, often very different models share the prediction of a new mass resonance. Specifically, we search for such resonances in the dimuon channel. Since no new resonance was in fact observed, the various models only serve to parameterize the degree to which nothing was found. In chapter 2, an outline of the Standard Model is given, and some extensions to the Standard Model are described. Chapter 3 describes the accelerator complex and the D0 detector. Chapter 4 describes the simulation of both Standard Model and new physics processes and the detector simulation. Chapter 5 describes the data sample, the reconstruction algorithms, the event selection and the kinematic fit. A sample of high-PT dimuon events is obtained by applying a number of cuts, and the simulation is tuned to match the efficiencies and track resolution measured on Z→ μ + μ - data. The integrated luminosity times the trigger efficiency of the final sample is determined by comparing it to the predicted (and well-tested) Z cross section. The mass resolution is improved by a kinematic fit which uses the transverse energy balance in the event to constrain the muon track P T for each event in the final selection. In chapter 6, a final cut on the (fitted) invariant mass of the selected events is chosen in such a way that it maximizes the sensitivity to the presence of new physics for specific models and resonance masses. A counting experiment is performed for several values of the mass of a hypothetical new particle. Since no excess over the predicted background is observed, Bayesian credibility limits on the production cross-section times branching ratio of the hypothetical new particles were calculated, as well as limits on the parameters of new physics models. The uncertainties on the theory and detector simulation are taken into account. An event display of the highest reconstructed mass event is shown as well. Finally, chapter 7 gives a conclusion, some prospects of improving the result and a brief outlook.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Bayesian analysis of the backreaction models

We present a Bayesian analysis of four different types of backreaction models, which are based on the Buchert equations. In this approach, one considers a solution to the Einstein equations for a general matter distribution and then an average of various observable quantities is taken. Such an approach became of considerable interest when it was shown that it could lead to agreement with observations without resorting to dark energy. In this paper we compare the {Lambda}CDM model and the backreaction models with type Ia supernovae, baryon acoustic oscillations, and cosmic microwave background data, and find that the former is favored. However, the tested models were based on some particular assumptions about the relation between the average spatial curvature and the backreaction, as well as the relation between the curvature and curvature index. In this paper we modified the latter assumption, leaving the former unchanged. We find that, by varying the relation between the curvature and curvature index, we can obtain a better fit. Therefore, some further work is still needed--in particular, the relation between the backreaction and the curvature should be revisited in order to fully determine the feasibility of the backreaction models to mimic dark energy.

79 ASTRONOMY AND ASTROPHYSICS↗

Machine Learning Assisted Safety Modeling and Analysis of Advanced Reactors

With the advances in computational power and numerical methods, analysts can now rely on first-principle simulations to predict ultra-fine details in a variety of applications. Advances in machine learning (ML) have produced algorithms that can now learn high-level abstractions via hierarchical models. This project aims to leverage advances in ML techniques and the available high-resolution simulation data to develop a novel modeling and simulation (M\&S) methodology for reactor safety analysis. While application-agnostic ML techniques are available, complex physics constraints need to be incorporated into ML techniques to build ML-based closures for computationally efficient predictive simulations. This project intends to develop a physics-guided data-driven multi-scale methodology for M\&S of advanced reactors. The project focuses on thermal fluid (T/F) phenomena, which play major roles in advanced reactor safety. Specifically, we propose a data-driven coarse-mesh turbulence model based on local flow features for the transient analysis of thermal mixing and stratification in a sodium-cooled fast reactor (SFR). The model has a coarse-mesh setup to ensure computational efficiency, while it is trained by fine-mesh computational fluid dynamics (CFD) data with Reynolds-averaged Navier-Stokes (RANS) turbulence model to ensure accuracy. Three different neural networks are developed and tested for loss-of-flow transients in the hot pool of SFR, i.e. the densely connected convolutional neural network (DCNN), long-short-term-memory network based on proper orthogonal decomposition (POD-LSTM), and the DCNN informed by LSTM (DCNN-LSTM). The performances of these three neural networks are evaluated based on baseline models. The DCNN-LSTM model has been chosen for further hyperparameter optimization. Furthermore, based on a simplified two-dimensional case, uncertainty quantification (UQ) of the developed ML-based closure are investigated with three methods, i.e. Monte Carlo dropout, deep ensemble, and Bayesian neural network. The developed ML-based turbulent viscosity closure relation based on deep ensemble is then integrated into the system analysis module SAM and serves as a term in the conservation equations. Such a SAM-ML based procedure guarantees that the obtained results are consistent with the physical constraints of the thermal-fluid system. The SAM-ML simulation on the same loss-of-flow transient showed comparable accuracy with the CFD simulation but with a much coarser mesh setup. Last but not least, the ML-based closure improvement with the support of higher-fidelity data from large eddy simulation (LES) is discussed. As a first step towards this direction, a baseline LES simulation is performed to obtain comparable data with RANS results. Based on the early results, future investigation on further improving the ML-based closure is discussed. We believe the developed approach that combines scientific machine learning with nuclear system analysis code can benefit the advanced reactor community as more accurate safety analyses will better characterize reactor safety margins and reduce licensing efforts.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗