Flame spray pyrolysis optimization via statistics and machine learning
Not Available
SEARCH · Search NASA
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.
Not Available
This presentation discusses how Upper Subcritical Limits (USL) are key parameters to determine operational limits in nuclear criticality safety evaluations. It also discusses an example of plutonium casting operation using tantalum at LANL PF-4. The Whisper tool at Los Alamos relies on many inputs, including covariance data, leading the presentation to ask if an existing benchmark data be used in Whisper to adjust nuclear data and covariances to justify a higher USL. If not, Whisper can be used to help design an optimal new benchmark experiment. The presentation also seeks to determine what the possible impacts are on USL and operational limits for plutonium casting. In conclusion, nuclear data covariances are used for by Whisper for: GSSL adjustment of nuclear data and covariances, identification of most similar existing benchmark experiments to application, simulation of Upper Subcritical Limit of application, and input to optimization techniques for designing most appropriate new benchmark experiment(s) to meet requirements. This requires a complete set of nuclear data covariances, benchmarks and k-effective sensitivity profiles (for both benchmarks and applications). The presentation concludes by asking if end users should trust results that depend on current covariance data.
Sensitivity analysis (SA) and uncertainty quantification (UQ) are now widely recognized as essential parts of analyses for complex nuclear systems. Plenty of well-established methodologies are already available for performing SA and UQ, such as Monte Carlo analysis, adjoint-based method, and variance decomposition. The focus of this summary is on developing an efficient sampling-based approach to perform SA and UQ through RAVEN. RAVEN (Risk Analysis and Virtual control Environment) is a software framework able to perform SA and UQ, currently under development at the Idaho National Laboratory. In literature, a number of approaches have been developed to accelerate the process of SA and UQ for large complex systems, such as principal component analysis (PCA) of input space, input-output correlation analysis, and intersection subspace methods. The fundamental idea is to identify the dominant active directions in the parameter space prior to conducting SA and UQ. The premise is that the number of active directions is much less than the dimensionality of input space. In this work, the PCA or Karhunen-Loeve expansion or SVD method is implemented in RAVEN to identify the active directions. This approach is also combined with the forward sampling approaches, such as Monte Carlo, Grid, Latin Hypercube, and Response Surface Design sampling approaches that are currently available in RAVEN, to perform SA and UQ more efficiently. The summary is organized as follows. In the following section, we present the PCA-based forward sampling approach. The effectiveness and efficiency are discussed in the numerical results section. The paper then ends with a concluding summary. In this summary, we have presented PCA-based forward sampling approach to perform SA and UQ. The goal is to reduce the computational overhead for repetitive executions of complex nuclear systems that are generally required when performing SA and UQ. The control of correlations between input parameters is implemented in RAVEN to better represent the input relationships. In addition, we are currently examining our implementations with more realistic models, and we are also focusing on implementation of some new high-dimensional input reduction techniques in RAVEN. (authors)
We present a physics-based methodology for quantifying the uncertainty in a given quantity of interest (QOI) that is contributed by uncertainties in opacities in radiation transport problems. Typically, opacities are tabulated as a function of density, temperature, and photon energy group. The size of this table makes a study of uncertainties at this level challenging because of the well-known 'curse of dimensionality.' We address this by studying uncertain parameters in the underlying physical model that generates the opacity tables. At this level, there are fewer uncertain parameters but still too many to analyze directly through computationally expensive radiation transport simulations. In order to explore this large uncertain parameter space, we develop two simplified radiation transport problems that are much less computationally demanding than the target problem of interest. An emulator is created for each QOI for each simplified problem using Bayesian Multivariate Adaptive Regression Splines (BMARS). This emulator is used to create a functional relationship between the QOIs and the uncertain parameters. Sensitivity analysis is performed using the emulator to determine which parameters contribute significantly to the uncertainty. This physics-based screening process reduces the dimension of the parameter space that is then studied via the computationally expensive radiation transport calculation to generate distributions of quantities of interest. Results of this research demonstrate that the QOIs for the target problem agree for varying screening criteria determined by the sensitivity analysis, and the QOIs agree well for varying Latin Hypercube Design (LHD) sample sizes for the uncertain space. (authors)
Sensitivity analysis (SA) is a commonly used approach for identifying important parameters that dominate model behaviors. We use a newly developed software package, a Problem Solving environment for Uncertainty Analysis and Design Exploration (PSUADE), to evaluate the effectiveness and efficiency of ten widely used SA methods, including seven qualitative and three quantitative ones. All SA methods are tested using a variety of sampling techniques to screen out the most sensitive (i.e., important) parameters from the insensitive ones. The Sacramento Soil Moisture Accounting (SAC-SMA) model, which has thirteen tunable parameters, is used for illustration. The South Branch Potomac River basin near Springfield, West Virginia in the U.S. is chosen as the study area. The key findings from this study are: (1) For qualitative SA methods, Correlation Analysis (CA), Regression Analysis (RA), and Gaussian Process (GP) screening methods are shown to be not effective in this example. Morris One-At-a-Time (MOAT) screening is the most efficient, needing only 280 samples to identify the most important parameters, but it is the least robust method. Multivariate Adaptive Regression Splines (MARS), Delta Test (DT) and Sum-Of-Trees (SOT) screening methods need about 400–600 samples for the same purpose. Monte Carlo (MC), Orthogonal Array (OA) and Orthogonal Array based Latin Hypercube (OALH) are appropriate sampling techniques for them; (2) For quantitative SA methods, at least 2777 samples are needed for Fourier Amplitude Sensitivity Test (FAST) to identity parameter main effect. McKay method needs about 360 samples to evaluate the main effect, more than 1000 samples to assess the two-way interaction effect. OALH and LPτ (LPTAU) sampling techniques are more appropriate for McKay method. For the Sobol' method, the minimum samples needed are 1050 to compute the first-order and total sensitivity indices correctly. These comparisons show that qualitative SA methods are more efficient but less accurate and robust than quantitative ones.
A parameter study was performed for the Los Alamos National Laboratory Blast Tube to tune geometry to best replicate peak blast pressures from experiments. Charge placement, reflecting wall thickness, flange gap width, and reflecting wall stand-off were sampled using a Latin-Hypercube design to train a Radial Basis Function Network surrogate model. This surrogate model was then used with a Nelder - Mead Simplex minimization routine to minimize error for the surrogate predictions.
Dynamic probabilistic risk analysis (DPRA) has emerged in the nuclear industry as an effective tool to identify points of weakness in nuclear reactors and create a comprehensive approach to quantifying risk. While DPRA allows for detailed analysis, it usually requires thousands of runs due to the necessity of exploring low probability, high consequence events, which can be computationally prohibitively difficult to assess. Risk Analysis Virtual Environment (RAVEN) is a software package being developed under the Nuclear Energy Advanced Modeling and Simulation program at the Idaho National Laboratory. It is able to manage complex control logic to drive a RELAP-7 simulation and to handle sampling strategies including Monte Carlo, Latin Hypercube, Grid and Adaptive samplers. By performing a thorough investigation of the possibility space of an accident scenario with RELAP-7, RAVEN is able to identify consequences of accident scenarios that have not been encountered. The most computational resource intensive step of a RAVEN analysis is running the RELAP-7 code modeling the physical system behavior. Using surrogate models to represent the outcomes of more complex simulations rather than the full model can greatly reduce the computation effort required. In situations where the uncertainty distributions of a model are unknown, or when there are multiple sets of uncertainty distributions, DPRA generally requires rerunning the physical model for each set of uncertainty distributions used. By taking an unbiased sample of the physical model and creating a surrogate model from the results, the surrogate model can be sampled under a different set of uncertainty distributions to provide equivalent results. This paper presents an example of this approach. (authors)
The overachieving goal of the whole CMDV-MCS project is to improve understanding of warm season continental convection and to develop treatments of convection and microphysics capable of representing mesoscale convective systems (MCSs) features in large-scale models. Our tasks for this project contributing to the overachieving goal include: (1) Improve the ice nucleation formulation for MG2 and P3 cloud microphysics schemes; (2) Improve the treatment of subgrid dynamics and thermodynamics driving the ice nucleation in E3SM; and (3) Test the performance of improved ice microphysics in E3SM with observation data. In this project, we have (1) Improved the ice nucleation parameterization for MG2 and P3 in E3SM by implementing two advanced empirical parameterizations with connection to aerosols. The two deterministic heterogeneous ice nucleation parameterizations (i.e., DeMott et al., 2015; Niemand et al., 2012) were merged with the MG2 and P3 cloud microphysics schemes in E3SM. Long-term simulations were conducted to examine the impacts of these new parameterizations on simulated cloud properties; (2) Improved the treatment of subgrid dynamics and thermodynamics driving the ice nucleation in E3SM. We evaluated the double Gaussian PDF of vertical velocity simulated by the Cloud Layers Unified By Binormals (CLUBB) and the sub-column vertical velocity sampled from the Subgrid Importance Latin Hypercube Sampler (SILHS) in E3SM. We introduced the vertical velocity variance induced by topographic gravity waves for ice nucleation and droplet activation; and (3) Tested the performance of improved ice microphysics in E3SM with observation data. We tested the new treatments of ice nucleation in the single column model (SCM) mode for the stratiform mixed-phase clouds observed during 9-10 October 2004 in the DOE ARM Mixed-Phase Arctic Cloud Experiment (M-PACE) and for the convective clouds observed on 20 May 2011 in the Midlatitude Continental Convective Clouds Experiment (MC3E). Modeled ice nucleating particles (INPs) concentrations were compared against observations collected around the globe.
Representation of clouds remains among the largest uncertainties in climate models and thus climate projections. Clouds vary significantly over different climate regimes and are controlled by different dynamics and physics. Using the cloud simulator output from perturbed-parameter ensemble climate runs with prescribed monthly sea surface temperature, this study examines the performance of the Community Atmosphere Model version 4 (CAM4) in simulating clouds over different tropical regions. Perturbing 28 selected parameters shows that model performance is quite sensitive to parameter values in different cloud regimes. Carefully adjusting these parameters could lead to a better simulation of clouds over many regions compared with the default model. Latin hypercube runs that pseudo-randomly sample the 28 parameters simultaneously have much wider spread and more spatial variations than the runs with parameters varied One-At-a-Time (OAT), suggesting the importance of non-linearities and interactions among parameters associated with different physical processes. The perturbed parameters have a relatively large impact on the mean bias compared to the pattern error
RAVEN is a software framework able to perform parametric and stochastic analysis based on the response of complex system codes. The initial development was aimed at providing dynamic risk analysis capabilities to the thermohydraulic code RELAP-7, currently under development at Idaho National Laboratory (INL). Although the initial goal has been fully accomplished, RAVEN is now a multi-purpose stochastic and uncertainty quantification platform, capable of communicating with any system code. In fact, the provided Application Programming Interfaces (APIs) allow RAVEN to interact with any code as long as all the parameters that need to be perturbed are accessible by input files or via python interfaces. RAVEN is capable of investigating system response and explore input space using various sampling schemes such as Monte Carlo, grid, or Latin hypercube. However, RAVEN strength lies in its system feature discovery capabilities such as: constructing limit surfaces, separating regions of the input space leading to system failure, and using dynamic supervised learning techniques. The development of RAVEN started in 2012 when, within the Nuclear Energy Advanced Modeling and Simulation (NEAMS) program, the need to provide a modern risk evaluation framework arose. RAVEN’s principal assignment is to provide the necessary software and algorithms in order to employ the concepts developed by the Risk Informed Safety Margin Characterization (RISMC) program. RISMC is one of the pathways defined within the Light Water Reactor Sustainability (LWRS) program. In the RISMC approach, the goal is not just to identify the frequency of an event potentially leading to a system failure, but the proximity (or lack thereof) to key safety-related events. Hence, the approach is interested in identifying and increasing the safety margins related to those events. A safety margin is a numerical value quantifying the probability that a safety metric (e.g. peak pressure in a pipe) is exceeded under certain conditions. Most of the capabilities, implemented having RELAP-7 as a principal focus, are easily deployable to other system codes. For this reason, several side activates have been employed (e.g. RELAP5-3D, any MOOSE-based App, etc.) or are currently ongoing for coupling RAVEN with several different software. The aim of this document is to provide a set of commented examples that can help the user to become familiar with the RAVEN code usage.
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.
The goal of this document is to provide a set of detailed examples that can help the user to become familiar with the RAVEN code. RAVEN is capable of investigating system response and explore input space using various sampling schemes such as Monte Carlo, grid, or Latin Hypercube. How- ever, RAVEN strength lies in its system feature discovery capabilities such as: constructing limit surfaces, separating regions of the input space leading to system failure, and using dynamic super- vised learning techniques. New users should consult the RAVEN Tutorial to get started.
RAVEN is a software framework able to perform parametric and stochastic analysis based on the response of complex system codes. The initial development was aimed at providing dynamic risk analysis capabilities to the thermohydraulic code RELAP-7, currently under development at Idaho National Laboratory (INL). Although the initial goal has been fully accomplished, RAVEN is now a multi-purpose stochastic and uncertainty quantification platform, capable of communicating with any system code. In fact, the provided Application Programming Interfaces (APIs) allow RAVEN to interact with any code as long as all the parameters that need to be perturbed are accessible by input files or via python interfaces. RAVEN is capable of investigating system response and explore input space using various sampling schemes such as Monte Carlo, grid, or Latin hypercube. However, RAVEN strength lies in its system feature discovery capabilities such as: constructing limit surfaces, separating regions of the input space leading to system failure, and using dynamic supervised learning techniques. The development of RAVEN started in 2012 when, within the Nuclear Energy Advanced Modeling and Simulation (NEAMS) program, the need to provide a modern risk evaluation framework arose. RAVEN’s principal assignment is to provide the necessary software and algorithms in order to employ the concepts developed by the Risk Informed Safety Margin Characterization (RISMC) program. RISMC is one of the pathways defined within the Light Water Reactor Sustainability (LWRS) program. In the RISMC approach, the goal is not just to identify the frequency of an event potentially leading to a system failure, but the proximity (or lack thereof) to key safety-related events. Hence, the approach is interested in identifying and increasing the safety margins related to those events. A safety margin is a numerical value quantifying the probability that a safety metric (e.g. peak pressure in a pipe) is exceeded under certain conditions. Most of the capabilities, implemented having RELAP-7 as a principal focus, are easily deployable to other system codes. For this reason, several side activates have been employed (e.g. RELAP5-3D, any MOOSE-based App, etc.) or are currently ongoing for coupling RAVEN with several different software. The aim of this document is to provide a set of commented examples that can help the user to become familiar with the RAVEN code usage.
Computer simulation experiments are a valuable complement to physical experimentation for systems where there is limited access to quantities of interest. The motivation for this work is to characterize the fluid dynamic and heat transfer processes that affect melt rate of waste slurry during vitrification, which can limit the throughput rate of the melter. Heat transfer to the cold cap from the molten glass below this layer is an important factor determining melt rate. In this study, computational fluid dynamics is used to elucidate the behavior of gas-filled cavities that form underneath the cold cap in a laboratory-scale waste glass melter. The cold cap consists of a reacting feed layer above a foam layer. The cavities are a result of coalesced gases from collapsing primary foam and air from forced convection bubblers. Bubbles in the foam layer move horizontally with the circulation pattern of the melt and escape around the edges of the cold cap to the plenum above. The effects of varying thermophysical properties and foam parameters based on the bounds of high-level waste glass formulations were explored. Using the DAKOTA toolkit, a Latin hypercube space-filling design was used to randomly distribute samples in bins to achieve good coverage over the parameter space with minimal points. Using the results from the set of simulations, a Gaussian process model with a linear trend was used to create response surfaces for several key quantities of interest, including temperatures, temperature gradients, velocities, and foam extent beneath the cold cap.
Numerical models of weather and climate need to compute grid-box-averaged rates of physical processes such as microphysics. These averages are computed by integrating subgrid variability over a grid box. For this reason, an important aspect of atmospheric modeling is spatial integration over subgrid scales. The needed integrals can be estimated by Monte Carlo integration. Monte Carlo integration is simple and general but requires many evaluations of the physical process rate. To reduce the number of function evaluations, this paper describes a new, flexible method of importance sampling. It divides the domain of integration into eight categories, such as the portion that contains both precipitation and cloud, or the portion that contains precipitation but no cloud. It then allows the modeler to prescribe the density of sample points within each of the eight categories. The new method is incorporated into the Subgrid Importance Latin Hypercube Sampler (SILHS). The resulting method is tested on drizzling cumulus and stratocumulus cases. In the cumulus case, the sampling error can be considerably reduced by drawing more sample points from the region of rain evaporation.
The objective of the work presented is to perform a preliminary sensitivity analysis of particle-in-cell (PIC) model parameters when applied to settling bed, bubbling fluidized bed, and circulating fluidized bed simulations. These examples correspond to widely different flow conditions commonly seen in chemical engineering applications. Simulations were performed using the PIC method in the open-source software Multiphase Flow with Interphase eXchanges (MFiX) developed by the National Energy Technology Laboratory (NETL). As part of the non-intrusive uncertainty quantification (UQ) analysis, simulation campaigns were generated using Nodeworks. Sampling locations or settings for PIC model parameters were determined using the Latin Hypercube method. Response surfaces were created using radial basis functions (RBF), and Sobol’ indices were estimated to quantify the influence of model parameters on the quantities of interest (QoI). This study marks a first step towards systematically determining optimal ranges for model parameters used in MFiX-PIC. Based on limited experience, it is expected that these values would depend strongly on flow conditions. Given the complexity of the multiphase flow systems under analysis, a non-intrusive UQ based approach is used to identify the most influential parameters in each case. This prior knowledge will help in proposing an effective design of experiments (DoE) and determine optimal parameters through techniques such as deterministic or Bayesian calibration, which will be pursued in the future.
Many search strategies used to optimize parameters during hydrologic model calibration are inherently limited in that simulations must be performed sequentially in order for the algorithm to converge. As a result, the overall time required to reach a solution increases linearly with the number of steps necessary. In contrast, sampling strategies that attempt to approximate the posterior distribution of the parameters do not require the model simulations to be run in sequence and therefore invite the use of a massively parallel approach across heterogeneous computing environments. We apply a cloud computing methodology to this problem where a head node running on Amazon Web Services continuously produces candidate models that are distributed to arbitrary compute environments to run the simulation, in our case Clemson’s Palmetto cluster computer. Thus the time required to obtain a sufficient number of samples to produce an adequate approximation of the posterior distribution of the parameters is limited by the number of available compute sites. Monte Carlo sampling of the posterior space would be duplicative and inefficient, thus we combine Latin hypercube methods with genetic algorithms to ensure that no large regions of the parameter space are left unexplored, and that high probability regions are investigated thoroughly. We independently assess the fit to different types of data sets without the need to construct a single objective function. Sets of Pareto optimal models are particularly valuable when the quality and information content of data is poorly known or conceptual model errors exist in our simulation. We illustrate our approach for a problem where in situ strain data are used to calibrate the parameters of an oil reservoir in the Avant oil field in Oklahoma. We use well injection and shut-in to stress a deep (500m) confined formation, and observe the geomechanical response using borehole strainmeters installed in the shallow subsurface (30m). These data were then used to calibrate the poroelastic properties of the reservoir and caprock along with the geometry of an elliptic channel associated with the formation. Using our approach we are able to identify issues with model error as well as substantial tradeoffs that occur within the parameters.
This presentation provides an overview of emulation, calibration and experiment design for computer experiments. Emulation refers to building a statistical surrogate from a carefully selected and limited set of model runs to predict unsampled outputs. The standard kriging approach to emulation of complex computer models is presented. Calibration refers to the process of probabilistically constraining uncertain physics/engineering model inputs to be consistent with observed experimental data. An initial probability distribution for these parameters is updated using the experimental information. Markov chain Monte Carlo (MCMC) algorithms are often used to sample the calibrated parameter distribution. Several MCMC algorithms commonly employed in practice are presented, along with a popular diagnostic for evaluating chain behavior. Space-filling approaches to experiment design for selecting model runs to build effective emulators are discussed, including Latin Hypercube Design and extensions based on orthogonal array skeleton designs and imposed symmetry requirements. Optimization criteria that further enforce space-filling, possibly in projections of the input space, are mentioned. Designs to screen for important input variations are summarized and used for variable selection in a nuclear fuels performance application. This is followed by illustration of sequential experiment design strategies for optimization, global prediction, and rare event inference.