Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Operator inference”

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 307 records · Page 17
All sourcesNASA NTRSDOE OSTI
NASA NTRS1995

Variability of moisture in the upper troposphere as inferred from TOVS satellite observations and the ECMWF model analyses in 1989

Large-scale variability of moisture in the upper troposphere is examined using TIROS (Television Infrared Observation Satellite) Operational Vertical Sounder (TOVS) satellite observations and European Centre for Medium-Range Weather Forecasts (ECMWF) model analyses from 1989 in the latitude band from 40 deg N to 40 S. To compare these dissimilar datasets, upwelling radiances were computed for the 6-7 micrometer water vapor band from the ECMWF temperature and moisture analyses, and these computed radiances were compared to the corresponding TOVS satellite observations. The ECMWF-based radiances reproduce the general locations and seasonal cycle of the TOVS-observed moisture features, particularly after an improved convective parameterization scheme was adopted by ECMWF in May 1989. However, the ECMWF analysis scheme still results in much milder lateral moisture gradients and seasonal contrasts than indicated by the TOVS observations. Seasonally, the upper troposphere in each hemisphere dries in winter and moistens in summer, but there are regions in each hemisphere that run counter to this seasonal trend, apparently depending on continental- and monsoon-scale dynamics. Dynamically, the TOVS-observed regions of significant subtropical dryness are correlated with persistent subsidence indicated by ECMWF 300-mb vertical velocity analyses. The TOVS radiance observations indicate large variations in space and time of the upper-tropospheric moisture field, which are not fully captured by the ECMWF analyses.

Salathe, Eric P., Jr.↗
NASA NTRS1998

A Proposal to Develop Interactive Classification Technology

Research for the first year was oriented towards: 1) the design of an interactive classification tool (ICT); and 2) the development of an appropriate theory of inference for use in ICT technology. The general objective was to develop a theory of classification that could accommodate a diverse array of objects, including events and their constituent objects. Throughout this report, the term "object" is to be interpreted in a broad sense to cover any kind of object, including living beings, non-living physical things, events, even ideas and concepts. The idea was to produce a theory that could serve as the uniting fabric of a base technology capable of being implemented in a variety of automated systems. The decision was made to employ two technologies under development by the principal investigator, namely, SMS (Symbolic Manipulation System) and SL (Symbolic Language) [see debessonet, 1991, for detailed descriptions of SMS and SL]. The plan was to enhance and modify these technologies for use in an ICT environment. As a means of giving focus and direction to the proposed research, the investigators decided to design an interactive, classificatory tool for use in building accessible knowledge bases for selected domains. Accordingly, the proposed research was divisible into tasks that included: 1) the design of technology for classifying domain objects and for building knowledge bases from the results automatically; 2) the development of a scheme of inference capable of drawing upon previously processed classificatory schemes and knowledge bases; and 3) the design of a query/ search module for accessing the knowledge bases built by the inclusive system. The interactive tool for classifying domain objects was to be designed initially for textual corpora with a view to having the technology eventually be used in robots to build sentential knowledge bases that would be supported by inference engines specially designed for the natural or man-made environments in which the robots would be called upon to operate.

deBessonet, Cary↗
NASA NTRS1998

ELIPS: Toward a Sensor Fusion Processor on a Chip

The paper presents the concept and initial tests from the hardware implementation of a low-power, high-speed reconfigurable sensor fusion processor. The Extended Logic Intelligent Processing System (ELIPS) processor is developed to seamlessly combine rule-based systems, fuzzy logic, and neural networks to achieve parallel fusion of sensor in compact low power VLSI. The first demonstration of the ELIPS concept targets interceptor functionality; other applications, mainly in robotics and autonomous systems are considered for the future. The main assumption behind ELIPS is that fuzzy, rule-based and neural forms of computation can serve as the main primitives of an "intelligent" processor. Thus, in the same way classic processors are designed to optimize the hardware implementation of a set of fundamental operations, ELIPS is developed as an efficient implementation of computational intelligence primitives, and relies on a set of fuzzy set, fuzzy inference and neural modules, built in programmable analog hardware. The hardware programmability allows the processor to reconfigure into different machines, taking the most efficient hardware implementation during each phase of information processing. Following software demonstrations on several interceptor data, three important ELIPS building blocks (a fuzzy set preprocessor, a rule-based fuzzy system and a neural network) have been fabricated in analog VLSI hardware and demonstrated microsecond-processing times.

Daud, Taher↗
NASA NTRS2009

An Approach to Assess Observation Impact Based on Observation-Minus-Forecast Residuals

Langland and Baker (2004) introduced an approach to assess the impact of observations on the forecasts. In that, a state-space aspect of the forecast is defined and a procedure is derived that relates changes in the aspect with changes in the initial conditions associated with the assimilation of observations) ultimately providing information about the impact of individual observations on the forecast. Some features of the approach are to be noted. The typical choice of forecast aspect employed in related works is rather arbitrary and leads to an incomplete assessment of the observing system. Furthermore, the state-space forecast aspect requires availability of a verification state that should ideally be uncorrelated with the forecast but in practice is not. Lastly, the approach involves the adjoint operator of the entire data assimilation system and as such it is constrained by the validity of this operator. In this presentation, an observation-space metric is used that, for a relatively time-homogeneous observing system, allows inferring observation impact on the forecast without some of the limitations above. Specifically, using observation-minus-forecast residuals leads to an approach with the following features: (i) it suggests a rather natural choice of forecast aspect, directly linked to the analysis system and providing full assessment of the observations; (ii) it naturally avoids introducing undesirable correlations in the forecast aspect by verifying against the observations; and (iii) it does not involve linearization and use of adjoints; therefore being applicable to any length of forecast. The state and observation-space approaches might be complementary to some degree, and involve different limitations and complexities. Illustrations are given using the NASA GEOS-5 data.

Todling, Ricardo↗
NASA NTRS2017

Inferential Framework for Autonomous Cryogenic Loading Operations

We address problem of autonomous cryogenic management of loading operations on the ground and in space. As a step towards solution of this problem we develop a probabilistic framework for inferring correlations parameters of two-fluid cryogenic flow. The simulation of two-phase cryogenic flow is performed using nearly-implicit scheme. A concise set of cryogenic correlations is introduced. The proposed approach is applied to an analysis of the cryogenic flow in experimental Propellant Loading System built at NASA KSC. An efficient simultaneous optimization of a large number of model parameters is demonstrated and a good agreement with the experimental data is obtained.

Luchinsky, Dmitry G.↗
NASA NTRS2005

The Influence of Thermodynamic Phase on the Retrieval of Mixed-Phase Cloud Microphysical and Optical Properties in the Visible and Near Infrared Region

Cloud microphysical and optical properties are inferred from the bidirectional reflectances simulated for a single-layered cloud consisting of an external mixture of ice particles and liquid droplets. The reflectances are calculated with a rigorous discrete ordinates radiative transfer model and are functions of the cloud effective particle size, the cloud optical thickness, and the values of the ice fraction in the cloud (i.e., the ratio of ice water content to total water content). In the present light scattering and radiative transfer simulations, the ice fraction is assumed to be vertically homogeneous; the habit (shape) percentage as a function of ice particle size is consistent with that used for the Moderate Resolution Imaging Spectroradiometer (MODIS) operational (Collection 4 and earlier) cloud products; and the surface is assumed to be Lambertian with an albedo of 0.03. Furthermore, error analyses pertaining to the inference of the effective particle sizes and optical thicknesses of mixed-phase clouds are performed. Errors are calculated with respect to the assumption of a cloud containing solely liquid or ice phase particles. The analyses suggest that the effective particle size inferred for a mixed-phase cloud can be underestimated (or overestimated) if pure liquid phase (or pure ice phase) is assumed for the cloud, whereas the corresponding cloud optical thickness can be overestimated (or underestimated).

Lee, Joonsuk↗
NASA NTRS2001

Robust Strategy for Rocket Engine Health Monitoring

Monitoring the health of rocket engine systems is essentially a two-phase process. The acquisition phase involves sensing physical conditions at selected locations, converting physical inputs to electrical signals, conditioning the signals as appropriate to establish scale or filter interference, and recording results in a form that is easy to interpret. The inference phase involves analysis of results from the acquisition phase, comparison of analysis results to established health measures, and assessment of health indications. A variety of analytical tools may be employed in the inference phase of health monitoring. These tools can be separated into three broad categories: statistical, rule based, and model based. Statistical methods can provide excellent comparative measures of engine operating health. They require well-characterized data from an ensemble of "typical" engines, or "golden" data from a specific test assumed to define the operating norm in order to establish reliable comparative measures. Statistical methods are generally suitable for real-time health monitoring because they do not deal with the physical complexities of engine operation. The utility of statistical methods in rocket engine health monitoring is hindered by practical limits on the quantity and quality of available data. This is due to the difficulty and high cost of data acquisition, the limited number of available test engines, and the problem of simulating flight conditions in ground test facilities. In addition, statistical methods incur a penalty for disregarding flow complexity and are therefore limited in their ability to define performance shift causality. Rule based methods infer the health state of the engine system based on comparison of individual measurements or combinations of measurements with defined health norms or rules. This does not mean that rule based methods are necessarily simple. Although binary yes-no health assessment can sometimes be established by relatively simple rules, the causality assignment needed for refined health monitoring often requires an exceptionally complex rule base involving complicated logical maps. Structuring the rule system to be clear and unambiguous can be difficult, and the expert input required to maintain a large logic network and associated rule base can be prohibitive.

Santi, L. Michael↗
DOE OSTI2022

Generation of Data-Driven Expected Energy Models for Photovoltaic Systems

Although unique expected energy models can be generated for a given photovoltaic (PV) site, a standardized model is also needed to facilitate performance comparisons across fleets. Current standardized expected energy models for PV work well with sparse data, but they have demonstrated significant over-estimations, which impacts accurate diagnoses of field operations and maintenance issues. This research addresses this issue by using machine learning to develop a data-driven expected energy model that can more accurately generate inferences for energy production of PV systems. Irradiance and system capacity information was used from 172 sites across the United States to train a series of models using Lasso linear regression. The trained models generally perform better than the commonly used expected energy model from international standard (IEC 61724-1), with the two highest performing models ranging in model complexity from a third-order polynomial with 10 parameters (Radj2 = 0.994) to a simpler, second-order polynomial with 4 parameters (Radj2=0.993), the latter of which is subject to further evaluation. Subsequently, the trained models provide a more robust basis for identifying potential energy anomalies for operations and maintenance activities as well as informing planning-related financial assessments. We conclude with directions for future research, such as using splines to improve model continuity and better capture systems with low (≤1000 kW DC) capacity.

14 SOLAR ENERGY↗
DOE OSTI2023

Locating Operational Events of the Cooling Tower of a Nuclear Reactor with a Very Local Seismic Network

Geolocation of emergent seismic signals is challenging at close distances. Here, we used three-component data from a seismic network and a targeted experiment at a research nuclear reactor to locate seismic sources. Utilizing known events collected during the targeted experiment, we were able to infer source locations with seismic amplitudes and polarization characteristics of the data. Although the resolution of the source location is not perfect, the seismic amplitudes and polarization analysis offer useful constraints. For the known events, the source region inferred with our analysis includes the true source locations. Synthetic tests indicate the resolution is largely due to limited data coverage and measurement uncertainties because the synthetic tests show similar results compared with the field data. We identified the source of the unknown event through spectrum cross correlation between the signals from the known events and an unknown event. Our findings were confirmed by operational staff at the facility. When the propagation medium properties (i.e., seismic velocity and quality factor for attenuation) are known, our analysis can be applied to continuous data from a seismic array to infer both source amplitude and location. If the medium properties are not known, a targeted experiment can be conducted to estimate them.

58 GEOSCIENCES↗
NASA NTRS2006

Algorithm for NO2 Vertical Column Retrieval from the Ozone Monitoring Instrument

We describe the operational algorithm for the retrieval of stratospheric, tropospheric, and total column densities of nitrogen dioxide NO2 from earthshine radiances measured by the Ozone Monitoring Instrument (OMI), aboard the EOS-Aura satellite. The algorithm uses the DOAS method for the retrieval of slant column NO densities. Air mass factors (AMFs) calculated from a stratospheric NO2 profile are used to make initial estimates of the vertical column density. Using data collected over a 24-h period, a smooth estimate of the global stratospheric field is constructed. Where the initial vertical column densities exceed the estimated stratospheric field, we infer the presence of tropospheric NO2, and recalculate the vertical column density (VCD) using an AMF calculated from an assumed tropospheric NO2 profile. The parameters that control the operational algorithm were selected with the aid of a set of data assembled from stratospheric and tropospheric chemical transport models. We apply the optimized algorithm to OMI data and present global maps of NO2 VCDs for the first time.

Bucsela, Eric J.↗
DOE OSTI2021

VERIFICATION OF TRISO FUEL BURNUP USING MACHINE LEARNING ALGORITHMS

Pebble Bed Reactors are fueled with fuel pebbles that are circulated multiple times through the reactor vessel before discharge. During the normal operation of a PBR, ejected pebbles are returned to the reactor or discharged depending on the fuel burnup and physical condition of the pebbles. The burnup measurement is usually based on detected radiation signatures of fission products accumulated in the pebble fuel over burnup. Previous research has shown that height of photopeaks of fission products, such as 134Cs, 137Cs, 154Eu, etc., can be used independently or in combination to infer or predict the level of burnup in the fuel. However, it remains challenging to measure such complex sources due to self-shielding effects, strong radiation background and intervening materials. Another operational challenge is the required high throughput of burnup measurement, which necessitates limited measurement time and thus impacts quality of measured gamma-ray spectra. Hence, advanced spectral analysis methods are needed to analyze the noisy gamma spectra and predict the burnup values. We propose to use machine learning (ML) method to interpret gamma-ray spectra and predict the burnup values of the pebbles. ML has achieved widespread success and adoption across a few domains that require pattern recognition and analysis in varied data types. In this work, we apply three proven ML approaches - multilayer perceptrons, convolutional neural networks, and transformers - to the task of predicting fuel burnup from measured gamma spectra, and compile a dataset of simulated spectra for training and validation of the ML models. In this paper, we will discuss the network architecture of these three ML approaches and compare the performance of the simplest of these (MLP) to a standard linear regression.

98 NUCLEAR DISARMAMENT, SAFEGUARDS, AND PHYSICAL P↗
DOE OSTI2023

AutoReP: Automatic ReLU Replacement for Fast Private Network Inference

The proliferation of the Machine-Learning-As-A-Service (MLaaS) market has brought to light a number of clients’ data privacy and security concerns. One promising solution is private inference (PI) techniques using cryptographic primitives. These techniques often come with high computation and communication overhead associated with the non-linear operator such as ReLU. Several approaches have been developed in reducing the number of ReLU operations, however, they either require a heuristic threshold selection or introduce significant accuracy drop. This work presents AutoReP, a gradient-based framework for non-linear operators reduction that aims to mitigate these concerns from a systematic perspective. AutoReP automates the process of discrete selection of ReLU and polynomial functions on neurons to accelerate PI applications. We also introduce distribution-aware polynomial approximation (DaPa) to accurately approximate ReLUs under given distribution, preserving model expressivity. Our experimental results demonstrate significant accuracy improvements of 6.12% (94.31%, 12.9K ReLU budget, CIFAR-10), 8.39% (74.92%, 12.9K ReLU budget, CIFAR-100), and 9.45% (63.69%, 55K ReLU budget, Tiny-ImageNet) over current state-of-the-art methods, e.g., SNL. Morever, AutoReP is applied to EfficientNet-B2 on ImageNet dataset, and achieved 75.55% accuracy with 176.1 × ReLU budget reduction.

Peng, Hongwu↗
DOE OSTI2021

Keras CT Segmentation v.1

SAND2020-14119 O A Tensorflow and Keras-backed framework for learned segmentation methods of 3D CT scan volumes. Supported functionality includes training models, running inference and quantifying uncertainty. The main underlying model architecture is V-Net. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

Ganter, Tyler↗
DOE OSTI2021

EXPLAINABLE AND TRUSTWORTHY DIAGNOSTICS ACHIEVABLE THROUGH PROCESS-BASED AUTOMATED REASONING

An approach has been developed that incorporates domain knowledge to obtain a more explainable and trustworthy equipment health monitoring diagnosis than might otherwise be obtained from a purely data-driven method. Physics-based models serve to constrain the realizable solution space and render a more trusted diagnosis. An automated reasoning algorithm performs backward chaining to infer a diagnosis that is consistent with logic statements that have been evaluated as true. This diagnosis is made explainable to an operator by providing the forward chaining path that elucidates for inspection and validity testing those truths implied by the diagnosis.

automated reasoning↗
DOE OSTI2026

Learned adaptive properties for mitigation of weight perturbations in embedded spiking networks

Recent years have seen an increased importance of neural network inference in edge-based scenarios, which impose size and power constraints requiring novel computing devices. These same edge scenarios may require operating over long periods of time, or exposure to extreme environments, resulting in a drift of neural network weights that cause degraded performance. In searching for ways to develop neural network approaches that perform robustly under these conditions, we propose a biologically-inspired mechanism for the dynamic adaptation of within-neuron parameters that is guided by a global context signal carrying information about perturbations and variability in incoming stimuli. Specifically, we demonstrate that adaptive voltage thresholds or neuronal time constants, when informed by a global context signal, can enable network-level mechanisms to recover from perturbed synaptic weights. Consistent with prior literature, the context-modulated approach is effective for recurrent, but not feedforward networks, by modulating network level dynamics. We demonstrate this approach successfully recovers performance in image classification tasks and spatiotemporal tracking tasks under idealized and Gaussian noise as well as for realistic perturbations from a memristive device when exposed to ionizing radiation. Finally, we discuss how this approach enables the design of robust and energy-efficient neuromorphic systems that perform well, even in resource-constrained scenarios with extreme environments such as edge processing.

context modulation↗
NASA NTRS1977

Development of a parameter optimization technique for the design of automatic control systems

Parameter optimization techniques for the design of linear automatic control systems that are applicable to both continuous and digital systems are described. The model performance index is used as the optimization criterion because of the physical insight that can be attached to it. The design emphasis is to start with the simplest system configuration that experience indicates would be practical. Design parameters are specified, and a digital computer program is used to select that set of parameter values which minimizes the performance index. The resulting design is examined, and complexity, through the use of more complex information processing or more feedback paths, is added only if performance fails to meet operational specifications. System performance specifications are assumed to be such that the desired step function time response of the system can be inferred.

Whitaker, P. H.↗
NASA NTRS2007

Comparison of CALIPSO-Like, LaRC, and MODIS Retrievals of Ice Cloud Properties over SIRTA in France and Florida during CRYSTAL-FACE

This study compares cirrus particle effective radius retrieved by a CALIPSO-like method with two similar methods using MODIS, MODI Airborne Simulator (MAS), and GOES imagery. The CALIPSO-like method uses lidar measurements coupled with the split-window technique that uses the infrared spectral information contained at the 8.65-micrometer, 11.15-micrometer and 12.05-micrometer bands to infer the microphysical properties of cirrus clouds. The two other methods, sing passive remote sensing at visible and infrared wavelengths, are the operational MODIS cloud products (referred to by its archival product identifier MOD06 for MODIS Terra) and MODIS retrievals performed by the CERES team at LaRC (Langley Research Center) in support of CERES algorithms; the two algorithms will be referred to as MOD06- and LaRC-method, respectively. The three techniques are compared at two different latitudes: (i) the mid-latitude ice clouds study uses 18 days of observations at the Palaiseau ground-based site in France (SIRTA: Site Instrumental de Recherche par Teledetection Atmospherique) including a ground-based 532 nm lidar and the Moderate Resolution Imaging Spectrometer (MODIS) overpasses on the Terra Platform, (ii) the tropical ice clouds study uses 14 different flight legs of observations collected in Florida, during the intensive field experiment CRYSTAL-FACE (Cirrus Regional Study of Tropical Anvils and cirrus Layers-Florida Area Cirrus Experiment), including the airborne Cloud Physics Lidar (CPL) and the MAS. The comparison of the three methods gives consistent results for the particle effective radius and the optical thickness, but discrepancies in cloud detection and altitudes. The study confirms the value of an active remote-sensing method (CALIPSO-like) for the study of sub-visible ice clouds, in both mid-latitudes and tropics. Nevertheless, this method is not reliable in optically very thick tropical ice clouds.

Chiriaco, M.↗
NASA NTRS2014

The Impacts of Microphysics and Planetary Boundary Layer Physics on Model Simulations of U.S. Deep South Summer Convection

Convection-allowing numerical weather simula- tions have often been shown to produce convective storms that have significant sensitivity to choices of model physical parameterizations. Among the most important of these sensitivities are those related to cloud microphysics, but planetary boundary layer parameterizations also have a significant impact on the evolution of the convection. Aspects of the simulated convection that display sensitivity to these physics schemes include updraft size and intensity, simulated radar reflectivity, timing and placement of storm initi- ation and decay, total storm rainfall, and other storm features derived from storm structure and hydrometeor fields, such as predicted lightning flash rates. In addition to the basic parameters listed above, the simulated storms may also exhibit sensitivity to im- posed initial conditions, such as the fields of soil temper- ature and moisture, vegetation cover and health, and sea and lake water surface temperatures. Some of these sensitivities may rival those of the basic physics sensi- tivities mentioned earlier. These sensitivities have the potential to disrupt the accuracy of short-term forecast simulations of convective storms, and thereby pose sig- nificant difficulties for weather forecasters. To make a systematic study of the quantitative impacts of each of these sensitivities, a matrix of simulations has been performed using all combinations of eight separate microphysics schemes, three boundary layer schemes, and two sets of initial conditions. The first version of initial conditions consists of the default data from large-scale operational model fields, while the second features specialized higher- resolution soil conditions, vegetation conditions and water surface temperatures derived from datasets created at NASA's Short-term Prediction and Operational Research Tran- sition (SPoRT) Center at the National Space Science and Technology Center (NSSTC) in Huntsville, AL. Simulations as outlined above, each 48 in number, were conducted for five midsummer weakly sheared coastal convective events each at two sites, Mobile, AL (MOB) and Houston, TX (HGX). Of special interest to operational forecasters at MOB and HGX were accuracy of timing and placement of convective storm initiation, reflectivity magnitudes and coverage, rainfall and inferred lightning threat.

McCaul, Eugene W., Jr.↗