SEARCH · Engineering Papers
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.
First attempt at quantifying uncertainties in ocean remote sensing reflectance satellite measurements. Based on 1000 iterations of Monte Carlo. Data source is a SeaWiFS 4-day composite, 2003. The uncertainty is for remote sensing reflectance (Rrs) at 443 nm.
Uncertainty visualization plays a critical role in transforming ensemble simulation data into actionable insights by effectively communicating various dimensions of uncertainty within a system. The emergence of artificial intelligence-driven surrogate models trained on multirun ensemble data offers a transformative opportunity to replace computationally intensive simulations with fast estimates, enabling users to explore data spaces with unprecedented depth and interactivity. However, integrating ensemble data and surrogate models into decision-making workflows and tools introduces novel challenges for uncertainty visualization. These include reconciling and clearly communicating the unique uncertainties associated with ensembles and their surrogate model estimates, and leveraging these approximations to inform actionable decisions. This work explores these challenges in the context of high-dimensional data visualization, bridging discrete datasets with their continuous representations and addressing the complexities of systems that support iterative navigation between input and output spaces. We evaluate the role of uncertainty visualization in fostering intuitive, actionable interactions and identify critical hurdles in advancing this frontier of computational simulation.
Despite the increasing use of large-scale Land Surface Models (LSMs) in predicting hydrological responses in extreme conditions, there's a critical gap in understanding the uncertainties in these predictions. This study addresses this gap through a detailed diagnostic evaluation of the uncertainties arising from meteorological forcing selection and model parametrization in hydrological simulations of the Community Land Model version 5 (CLM5). CLM5 is configured at a spatial scale of about 12-km to simulate runoff processes for 464 headwater watersheds, selected from the Catchment Attributes for Large-Sample Studies (CAMELS) dataset to be representative of physiographic and climatic gradients across the conterminous United States. For each watershed, CLM5 is driven by five commonly used gridded forcing datasets in combination with a large ensemble (> 1200) of key CLM5 hydrologic parameters. Our results suggest that uncertainty in CLM5 runoff simulations resulting from both forcing and parametric sources is markedly higher in arid regions, e.g., Great Plains and Midwest regions. Uncertainty in low flow is dominated by parametric uncertainty, while the selection of meteorological forcing contributes more dominantly to high flow and seasonal flows during fall and spring. Our analysis also demonstrates that the selection of forcing datasets and the metrics used to calibrate CLM5 significantly impact the model’s predictive accuracy in extreme event severity for both floods and droughts. Overall, the results from this study highlight the need to understand and account for forcing and parametric uncertainties in CLM5 simulations, particularly for hazard and risk assessments addressing hydrologic extremes.
Many laboratory experiments generate data that are characterized by a form of uncertainty that differs from noise: experimental procedure uncertainty. The origin of this type of uncertainty is the difficulty in controlling a subset of experimental conditions in such a way that the experiment is perfectly reproducible within noise. In this report, we describe a Gaussian Process modeling-based method that accounts for experimental procedure uncertainty. The method accounts for variations in conditions from experiment to experiment that are independent between experiments, but correlated within each experiment, and incorporates the resulting uncertainty into predictions at future experimental settings. The method is discussed in the context of a specific chemistry application in which Raman scattering spectra are measured from three-component mixtures.
Recent drinking water regulations have imposed the need for per- and polyfluoroalkyl substances (PFAS) remediation. In response, treatment facilities may be required to retrofit existing treatment schemes to treat PFAS below maximum contaminant levels (MCLs). Adsorption technologies such as granular activated carbon (GAC) and ion exchange (IX) have been demonstrated to be effective; however, there are limited techno-economic metrics available which provide guidance on technology selection and design for diverse PFAS-containing source water conditions. Process systems engineering (PSE) tools which can traditionally perform these analyses are hindered by the data availability, model validity, and understanding of treatment phenomena for emerging contaminants. This work employs published data regressions, statistical models, process models, techno-economic analyses, and other process systems tools in a model-based uncertainty framework to consider the limitations of emerging contaminant research. Through this analysis framework, economic results are provided as probabilistic distributions based on the uncertainty of the models and diverse conditions that treatment facilities experience.
We present a new software, HYPPO, that enables the automatic tuning of hyperparameters of various deep learning (DL) models. Unlike other hyperparameter optimization (HPO) methods, HYPPO uses adaptive surrogate models and directly accounts for uncertainty in model predictions to find accurate and reliable models that make robust predictions. Using asynchronous nested parallelism, we are able to significantly alleviate the computational burden of training complex architectures and quantifying the uncertainty. HYPPO is implemented in Python and can be used with both TensorFlow and PyTorch libraries. We demonstrate various software features on time-series prediction and image classification problems as well as a scientific application in computed tomography image reconstruction. Finally, we show that (1) we can reduce by an order of magnitude the number of evaluations necessary to find the most optimal region in the hyperparameter space and (2) we can reduce by two orders of magnitude the throughput for such HPO process to complete.
Representing uncertainty in model inputs often means finding a balance between uncertainty and physical reality. Developing wide distributions may seem conservative in principle, but this approach may lead to unrealistic model results. Characterizing the current state of knowledge of stochastic inputs presents many challenges, especially if the data available are limited or have limited relevance to the site. Relationships among these inputs may also be important to represent but are typically complex or difficult to define. Often special adjustments must be made to account for reduced credibility in particular data. If parameters are strongly related to one another, a correlation structure may be developed for input to the model. Other techniques such as regression models may be used to incorporate relationships between the information available and the desired parameters. The process of developing distributions must consider details of the model in terms of what the distribution is meant to represent. This paper uses the example of a probabilistic fate and transport model for hexahydro-1,3,5-trinitro-1,3,5-triazine (RDX) in the regional aquifer at Los Alamos National Laboratory (LANL). For many parameters, a single draw is applied to all space and time over which the model is run, for a single iteration. This simplification is often made for many reasons, and can often be beneficial, but also adds additional complexity in the distribution development process. Defining the distributional goals as they relate to the modeling process is an important step, which should take place prior to evaluation of the data. Usually, the distributions developed are meant to characterize the average value of the parameter over the spatial and temporal domain of the model. Distribution development requires consideration of many sources of information on the parameter where available, ideally from multiple references. Examples of different sources include data from different references but also from different conditions, measurement methods, or experimental types. Depending on these conditions and the reliability or relevance of particular references, different sources of data may each contribute valuable information but have varying relevance to the site. In these cases, weighting data unequally is a useful way to incorporate this information. As an example, aqueous dispersivity data are available for a variety of rock types. Only a few values are available for the desired rock type, and this is not enough to develop a distribution. Therefore, dispersivity values from other rock types are included in distribution development but are down-weighted such that the best data have the most influence on the distribution developed. In another case, K{sub d} distributions in the model are meant to represent a known composition of multiple soil types. Data from these materials are weighted accordingly to develop a distribution for the weighted average K{sub d} across soil types. Other cases include varying reliability of different sources, and weighting data according to the confidence in these sources. In some cases, input parameters are correlated with one another. An example is advective porosity, which is positively correlated with total porosity and must be less than total porosity. Paired data with both parameters must exist to discern the relationship between the two parameters and if it is necessary to build a correlation structure into the model. In general, correlated parameters may be represented in the model by a multivariate distribution, or perhaps more desirably, capturing the correlation within the developed distributions which may be treated as independent from one another. In the example of porosity, this can be done by transforming advective porosity into a proportion of total porosity which may be drawn independently from the distribution of total porosity. This paper explores how complex data and correlations can be incorporated to meet the distributional goals of the model, using the RDX regional model as a detailed example. (authors)
Launch vehicle systems are designed and developed using both heritage and new hardware. Design modifications to the heritage hardware to fit new functional system requirements can impact the applicability of heritage reliability data. Risk estimates for newly designed systems must be developed from generic data sources such as commercially available reliability databases using reliability prediction methodologies, such as those addressed in MIL-HDBK-217F. Failure estimates must be converted from the generic environment to the specific operating environment of the system in which it is used. In addition, some qualification of applicability for the data source to the current system should be made. Characterizing data applicability under these circumstances is crucial to developing model estimations that support confident decisions on design changes and trade studies. This paper will demonstrate a data-source applicability classification method for suggesting epistemic component uncertainty to a target vehicle based on the source and operating environment of the originating data. The source applicability is determined using heuristic guidelines while translation of operating environments is accomplished by applying statistical methods to MIL-HDK-217F tables. The paper will provide one example for assigning environmental factors uncertainty when translating between operating environments for the microelectronic part-type components. The heuristic guidelines will be followed by uncertainty-importance routines to assess the need for more applicable data to reduce model uncertainty.
A real-time method was demonstrated for determining accurate uncertainty levels of stability and control derivatives estimated using recursive least squares and time-domain data. The method uses a recursive formulation of the residual autocorrelation to account for colored residuals, which are routinely encountered in aircraft parameter estimation and change the predicted uncertainties. Simulation data and flight test data for a subscale jet transport aircraft were used to demonstrate the approach. Results showed that the corrected uncertainties matched the observed scatter in the parameter estimates, and did so more accurately than conventional uncertainty estimates that assume white residuals. Only small differences were observed between batch estimates and recursive estimates at the end of the maneuver. It was also demonstrated that the autocorrelation could be reduced to a small number of lags to minimize computation and memory storage requirements without significantly degrading the accuracy of predicted uncertainty levels.
This paper considers calculation of prediction intervals (PIs) by neural networks (NNs) for quantifying uncertainty in regression tasks, so as to provide fast, accurate and robust emulators to accelerate scientific simulations. We propose a novel method to learn lower and upper bounds of the PI using independent NNs without defining an exclusive loss. Our method requires no distributional assumption, does not introduce extra hyper-parameters, and can effectively identify out-of-distribution samples and quantify their uncertainty. We demonstrate advantages of our method using a benchmark problem and two real-world scientific applications.
Sensitivity analysis estimates the relative contribution of the uncertainty in input values to the uncertainty of model outputs. Partial Rank Correlation Coefficient (PRCC) and Standardized Rank Regression Coefficient (SRRC) are methods of conducting sensitivity analysis on nonlinear simulation models like the Integrated Medical Model (IMM). The PRCC method estimates the sensitivity using partial correlation of the ranks of the generated input values to each generated output value. The partial part is so named because adjustments are made for the linear effects of all the other input values in the calculation of correlation between a particular input and each output. In SRRC, standardized regression-based coefficients measure the sensitivity of each input, adjusted for all the other inputs, on each output. Because the relative ranking of each of the inputs and outputs is used, as opposed to the values themselves, both methods accommodate the nonlinear relationship of the underlying model. As part of the IMM v4.0 validation study, simulations are available that predict 33 person-missions on ISS and 111 person-missions on STS. These simulated data predictions feed the sensitivity analysis procedures. The inputs to the sensitivity procedures include the number occurrences of each of the one hundred IMM medical conditions generated over the simulations and the associated IMM outputs: total quality time lost (QTL), number of evacuations (EVAC), and number of loss of crew lives (LOCL). The IMM team will report the results of using PRCC and SRRC on IMM v4.0 predictions of the ISS and STS missions created as part of the external validation study. Tornado plots will assist in the visualization of the condition-related input sensitivities to each of the main outcomes. The outcomes of this sensitivity analysis will drive review focus by identifying conditions where changes in uncertainty could drive changes in overall model output uncertainty. These efforts are an integral part of the overall verification, validation, and credibility review of IMM v4.0.
The Forest Light (FLIGHT) radiative transfer model was used to examine the uncertainty of vegetation structure measurements from NASA's planned ICESat-2 photon counting light detection and ranging (LiDAR) instrument across a synthetic Larix forest gradient in the taiga-tundra ecotone. The simulations demonstrate how measurements from the planned spaceborne mission, which differ from those of previous LiDAR systems, may perform across a boreal forest to non-forest structure gradient in globally important ecological region of northern Siberia. We used a modified version of FLIGHT to simulate the acquisition parameters of ICESat-2. Modeled returns were analyzed from collections of sequential footprints along LiDAR tracks (link-scales) of lengths ranging from 20 m-90 m. These link-scales traversed synthetic forest stands that were initialized with parameters drawn from field surveys in Siberian Larix forests. LiDAR returns from vegetation were compiled for 100 simulated LiDAR collections for each 10 Mg · ha(exp -1) interval in the 0-100 Mg · ha(exp -1) above-ground biomass density (AGB) forest gradient. Canopy height metrics were computed and AGB was inferred from empirical models. The root mean square error (RMSE) and RMSE uncertainty associated with the distribution of inferred AGB within each AGB interval across the gradient was examined. Simulation results of the bright daylight and low vegetation reflectivity conditions for collecting photon counting LiDAR with no topographic relief show that 1-2 photons are returned for 79%-88% of LiDAR shots. Signal photons account for approximately 67% of all LiDAR returns, while approximately 50% of shots result in 1 signal photon returned. The proportion of these signal photon returns do not differ significantly (p greater than 0.05) for AGB intervals greater than 20 Mg · ha(exp -1). The 50m link-scale approximates the finest horizontal resolution (length) at which photon counting LiDAR collection provides strong model fits and minimizes forest structure uncertainty in the synthetic Larix stands. At this link-scale AGB greater than 20 Mg · ha(exp -1) has AGB error from 20-50% at the 95% confidence level. These results suggest that the theoretical sensitivity of ICESat-2 photon counting LiDAR measurements alone lack the ability to consistently discern differences in inferred AGB at 10 Mg · ha(exp -1) intervals in sparse forests characteristic of the taiga-tundra ecotone.
In this work, multi-model Monte Carlo estimators are developed to reduce uncertainty in quantities of interest (QoIs) extracted from crystal plasticity simulations of additively manufactured (AM) metals. A significant concern in AM parts is uncertainty in mechanical properties caused in part by complex microstructures that arise from the AM process. Quantifying uncertainty in microstructure-sensitive behavior using experiments alone is costly, especially when mechanical allowables must be established. Quantitative relationships among microstructure, micromechanical metrics like slip accumulation, crack initiation, and failure are also difficult to capture with limited experiments. Crystal plasticity material models instead enable computational prediction of micromechanical stress and strain fields given a discretized microstructure. However, high-fidelity finely discretized crystal plasticity simulations are computationally expensive, while lower-fidelity models are less accurate and generally biased, making uncertainty quantification and reduction computationally difficult as well. Multi-model Monte Carlo methods leverage correlations between high- and low-fidelity models to produce unbiased estimators for QoIs with reduced uncertainty relative to standard Monte Carlo. Crystal plasticity QoIs considered in this work include yield strength and the mean and extreme values of micromechanical fields that are relevant to crack initiation. Multi-model Monte Carlo estimators are developed for each individual QoI and several groups of QoIs. The results of this work establish relationships among model correlations, sample allocation, and uncertainty reduction for different combinations of QoIs and demonstrate a trend of less uncertainty reduction as QoIs become more sensitive to local microstructure. Limitations from using pilot samples to estimate model covariances and train low-fidelity models are also addressed. The uncertainty reduction achieved by multi-model Monte Carlo is an important step toward using computational mechanics models to predict microstructure-sensitive crack initiation and failure in AM parts.
Compound riverine and coastal flooding is usually driven by complex interactions among meteorological, hydrological, and ocean extremes. However, existing efforts to model this phenomenon often do not integrate hydrological processes across atmosphere–land–river–ocean systems, leading to substantial uncertainties that have not been fully examined. To bridge this gap, we leverage the new capabilities of the Energy Exascale Earth System Model (E3SM) that enable a multi-component framework that integrates coastally refined atmospheric, terrestrial, and oceanic components. We evaluate compound uncertainties arising from two-way land–river–ocean coupling in E3SM and track the cascading meteorological and hydrological uncertainties through ensemble simulations over the Delaware River basin and estuary during Hurricane Irene (2011). Our findings highlight the importance of two-way river–ocean coupling to compound flood modeling and demonstrate E3SM's capability in capturing compound flood extent near the coast, with a hit rate over 0.75. Our study shows the growing uncertainties that transition from atmospheric forcings to flood distribution and severity. Furthermore, an analysis based on artificial neural networks is used to assess the roles of hydrological drivers, such as infiltration and soil moisture, in the generation of compound flooding. The response of compound floods to tropical cyclones (TCs) is found to be susceptible to these often overlooked drivers. For instance, the flooded area could increase more than 2-fold (∼2.4) if Hurricane Irene were preceded by an extreme antecedent soil moisture condition (AMC). The results not only support the use of a multi-component framework for interactive flooding processes, but also underscore the necessity of broader definitions of compound flooding that encompass the simultaneous occurrence of intense precipitation, storm surge, and high AMC during TCs.
Weather forecasting nowadays often requires some estimation of uncertainties associated with the output of meteorological models, in order to better inform decision making, especially in the context of intense weather events. Ensemble prediction systems provide such information through sets of possible scenarios which are designed to represent various uncertainty sources, including model uncertainties. A wide variety of methods have been proposed to estimate model uncertainties, among which perturbation methods targeting uncertain processes are a promising research field. Here, we focus on the representation of small-scale variability by process-oriented perturbation schemes applied to two key physical processes, namely turbulence and shallow convection. The perturbations are applied to a single-column version of the convection-permitting AROME model, in three idealized boundary-layer cases. Large-eddy simulations (LESs) of the same cases serve as a reference for the subgrid variability that has to be represented, and the results are also compared to those given by the Stochastically Perturbed Parametrization Tendencies (SPPT) method, which is a method commonly used by weather forecast centres to represent model uncertainty. The spread produced by our process-oriented perturbations of turbulence and shallow convection does not represent all the small-scale variability implied by the LESs for temperature and humidity. However, it is of a similar order of magnitude for the wind, thanks to perturbations generated by the stochastic turbulence scheme. The dispersion is structurally different from what is obtained with SPPT. It is non-negligible in the lower levels, where SPPT perturbations are usually suppressed because of numerical instabilities, indicating a possible complementarity between the schemes.
Flight testing has been the historical standard for determining aircraft airworthiness. However, increases in the cost of flight testing and the accuracy of inexpensive CFD encourage the adoption of certification by analysis to reduce or replace flight testing. A framework is introduced to predict the performance in the special case of a modification to an existing, previously certified aircraft. This framework uses a combination of existing flight tests or high fidelity data of the original aircraft as well as lower fidelity data from CFD or wind tunnel testing of the original and modified configurations to create 6-DOF flight dynamics models. Two methods are presented which generate an updated flight dynamics model and estimate the model form uncertainty for the modified aircraft configuration using knowledge of the original aircraft. This updated dynamics model and uncertainty estimate are then used to conduct non-deterministic simulations with wind turbulence included. The framework is applied to an example aircraft system to demonstrate the ability to predict the performance and associated model from the uncertainty of modified aircraft configurations.
Machine learning (ML) models, if trained to data sets of high-fidelity quantum simulations, produce accurate and efficient interatomic potentials. Active learning (AL) is a powerful tool to iteratively generate diverse data sets. In this approach, the ML model provides an uncertainty estimate along with its prediction for each new atomic configuration. If the uncertainty estimate passes a certain threshold, then the configuration is included in the data set. Here we develop a strategy to more rapidly discover configurations that meaningfully augment the training data set. The approach, uncertainty-driven dynamics for active learning (UDD-AL), modifies the potential energy surface used in molecular dynamics simulations to favor regions of configuration space for which there is large model uncertainty. The performance of UDD-AL is demonstrated for two AL tasks: sampling the conformational space of glycine and sampling the promotion of proton transfer in acetylacetone. The method is shown to efficiently explore the chemically relevant configuration space, which may be inaccessible using regular dynamical sampling at target temperature conditions.
This paper describes the first uncertainty analysis for drilling stability using a frequency-domain drilling stability model. The stability model inputs include: the modal parameters for the torsional-axial vibration mode from the twist drill-holder-spindle axial frequency response function; and the mechanistic coefficients that relate the torque and thrust force to chip area for the selected drill-workpiece material combination. Furthermore, Monte Carlo simulation is applied to propagate the input uncertainties to output uncertainty in the predicted stability map, which separates stable from unstable (chatter) zones in the spindle speed-chip width parameter space. Additionally, the mean stability boundary and its 95% confidence intervals are determined for five cases: varying all four inputs simultaneously and varying them individually. This enables the individual sensitivities to be compared. Experimental results from drilling tests are included for comparison to the prediction. Additionally, Matlab code is provided to implement the stability model and Monte Carlo uncertainty analysis.