Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Constrained regression”

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 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Sparse regression for plasma physics

Many scientific problems can be formulated as sparse regression, i.e., regression onto a set of parameters when there is a desire or expectation that some of the parameters are exactly zero or do not substantially contribute. This includes many problems in signal and image processing, system identification, optimization, and parameter estimation methods such as Gaussian process regression. Sparsity facilitates exploring high-dimensional spaces while finding parsimonious and interpretable solutions. In the present work, we illustrate some of the important ways in which sparse regression appears in plasma physics and point out recent contributions and remaining challenges to solving these problems in this field. Further, a brief review is provided for the optimization problem and the state-of-the-art solvers, especially for constrained and high-dimensional sparse regression.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

Integrating adaptive learning with post hoc model explanation and symbolic regression to build interpretable surrogate models

Abstract We develop a materials informatics workflow to build an interpretable surrogate model for micromagnetic simulations. Our goal is to predict the energy barrier of a moving isolated skyrmion in rare-earth-free $$\hbox {Mn}_4$$ Mn 4 N. Our approach integrates adaptive learning with post hoc model explanation and symbolic regression methods. We discuss an unexplored acquisition function (information condensing active learning) within the adaptive learning loop and compare it with the known standard deviation function for efficient navigation of the search space. Model-agnostic post hoc explanation techniques then uncover trends learned by the trained model, which we then leverage to constrain the expressions used for symbolic regression. Graphical abstract

Biswas, Ankita↗
NASA NTRS2007

Salience Assignment for Multiple-Instance Regression

We present a Multiple-Instance Learning (MIL) algorithm for determining the salience of each item in each bag with respect to the bag's real-valued label. We use an alternating-projections constrained optimization approach to simultaneously learn a regression model and estimate all salience values. We evaluate this algorithm on a significant real-world problem, crop yield modeling, and demonstrate that it provides more extensive, intuitive, and stable salience models than Primary-Instance Regression, which selects a single relevant item from each bag.

regression↗
DOE OSTI2020

A Survey of Constrained Gaussian Process: Approaches and Implementation Challenges

Gaussian process regression is a popular Bayesian framework for surrogate modeling of expensive data sources. As part of a larger effort in scientific machine learning, many recent works have incorporated physical constraints or other a priori information within Gaussian process regression to supplement limited data and regularize the behavior of the model. We provide an overview and survey of several classes of Gaussian process constraints, including positivity or bound constraints, monotonicity and convexity constraints, differential equation constraints provided by linear PDEs, and boundary condition constraints. We compare the strategies behind each approach as well as the differences in implementation, concluding with a discussion of the computational challenges introduced by constraints.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2007

Aeroelastic Model Structure Computation for Envelope Expansion

Structure detection is a procedure for selecting a subset of candidate terms, from a full model description, that best describes the observed output. This is a necessary procedure to compute an efficient system description which may afford greater insight into the functionality of the system or a simpler controller design. Structure computation as a tool for black-box modeling may be of critical importance in the development of robust, parsimonious models for the flight-test community. Moreover, this approach may lead to efficient strategies for rapid envelope expansion that may save significant development time and costs. In this study, a least absolute shrinkage and selection operator (LASSO) technique is investigated for computing efficient model descriptions of non-linear aeroelastic systems. The LASSO minimises the residual sum of squares with the addition of an l(Sub 1) penalty term on the parameter vector of the traditional l(sub 2) minimisation problem. Its use for structure detection is a natural extension of this constrained minimisation approach to pseudo-linear regression problems which produces some model parameters that are exactly zero and, therefore, yields a parsimonious system description. Applicability of this technique for model structure computation for the F/A-18 (McDonnell Douglas, now The Boeing Company, Chicago, Illinois) Active Aeroelastic Wing project using flight test data is shown for several flight conditions (Mach numbers) by identifying a parsimonious system description with a high percent fit for cross-validated data.

Kukreja, Sunil L.↗
DOE OSTI2024

Local primordial non-Gaussianity from the large-scale clustering of photometric DESI luminous red galaxies

ABSTRACT We use angular clustering of luminous red galaxies from the Dark Energy Spectroscopic Instrument (DESI) imaging surveys to constrain the local primordial non-Gaussianity parameter fNL. Our sample comprises over 12 million targets, covering 14 000 deg2 of the sky, with redshifts in the range 0.2 < z < 1.35. We identify Galactic extinction, survey depth, and astronomical seeing as the primary sources of systematic error, and employ linear regression and artificial neural networks to alleviate non-cosmological excess clustering on large scales. Our methods are tested against simulations with and without fNL and systematics, showing superior performance of the neural network treatment. The neural network with a set of nine imaging property maps passes our systematic null test criteria, and is chosen as the fiducial treatment. Assuming the universality relation, we find $f_{\rm NL} = 34^{+24(+50)}_{-44(-73)}$ at 68 per cent (95 per cent) confidence. We apply a series of robustness tests (e.g. cuts on imaging, declination, or scales used) that show consistency in the obtained constraints. We study how the regression method biases the measured angular power spectrum and degrades the fNL constraining power. The use of the nine maps more than doubles the uncertainty compared to using only the three primary maps in the regression. Our results thus motivate the development of more efficient methods that avoid overcorrection, protect large-scale clustering information, and preserve constraining power. Additionally, our results encourage further studies of fNL with DESI spectroscopic samples, where the inclusion of 3D clustering modes should help separate imaging systematics and lessen the degradation in the fNL uncertainty.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Implications of a Pervasive Climate Model Bias for Low‐Cloud Feedback

Abstract How low clouds respond to warming constitutes a key uncertainty for climate projections. Here we observationally constrain low‐cloud feedback through a controlling factor analysis based on ridge regression. We find a moderately positive global low‐cloud feedback (0.45 W , 90% range 0.18–0.72 W ), about twice the mean value (0.22 W ) of 16 models from the Coupled Model Intercomparison Project. We link this discrepancy to a pervasive model mean‐state bias: models underestimate the low‐cloud response to warming because (a) they systematically underestimate present‐day tropical marine low‐cloud amount, and (b) the low‐cloud sensitivity to warming is proportional to this present‐day low‐cloud amount. Our results hence highlight the importance of reducing model biases in both the mean state of clouds and their sensitivity to environmental factors for accurate climate change projections.

58 GEOSCIENCES↗
DOE OSTI2024

Reinterpreting fundamental plane correlations with machine learning

ABSTRACT This work explores the relationships between galaxy sizes and related observable galaxy properties in a large volume cosmological hydrodynamical simulation. The objectives of this work are to develop a better understanding of both the correlations between galaxy properties and the influence of environment on galaxy physics in order to build an improved model for the galaxy sizes, building off of the fundamental plane. With an accurate intrinsic galaxy size predictor, the residuals in the observed galaxy sizes can potentially be used for multiple cosmological applications, including making measurements of galaxy velocities in spectroscopic samples, estimating the rate of cosmic expansion, and constraining the uncertainties in the photometric redshifts of galaxies. Using projection pursuit regression, the model accurately predicts intrinsic galaxy sizes and have residuals which have limited correlation with galaxy properties. The model decreases the spatial correlation of galaxy size residuals by a factor of ∼5 at small scales compared to the baseline correlation when the mean size is used as a predictor.

Astronomy & Astrophysics↗
DOE OSTI2026

Leveraging Inequality-Constrained Data for Enhanced Liquidus Temperature Prediction in Nuclear Waste Glass Melts

Inequality-constrained data are frequently discarded in engineering, leading to significant information loss in data-scarce domains like glass characterization in nuclear waste vitrification. This paper presents a nonparametric censored-data regression framework based on an l1-norm optimization criterion that leverages slack variables to integrate left-, right-, and interval-constrained observations into training without distributional assumptions. Validated on synthetic data and a Physics-Informed Neural Network (PINN) for predicting liquidus temperature (TL), the method improved R2 from 0.60 to 0.89 and reduced Mean Absolute Error (MAE) by 48% (51.46 to 26.89?rC) on deterministic values. The traditional models failed to satisfy any inequality constraints while the proposed l1-norm PINN satisfies 81.25% of the constraints. The proposed framework effectively extracts actionable information from previously unusable data to enhance predictive accuracy, reduce epistemic uncertainty, and ensure physical consistency in complex industrial applications.

Garcia-Morado, Erick↗
DOE OSTI2020

Clumped isotope constraints on changes in latest Pleistocene hydroclimate in the northwestern Great Basin: Lake Surprise, California

Abstract During the Last Glacial Maximum (LGM) and subsequent deglaciation, the Great Basin in the southwestern United States was covered by numerous extensive closed-basin lakes, in stark contrast with the predominately arid climate observed today. This transition from lakes in the Late Pleistocene to modern aridity implies large changes in the regional water balance. Whether these changes were driven by increased precipitation rates due to changes in atmospheric dynamics, decreased evaporation rates resulting from temperature depression and summer insolation changes, or some combination of the two remains uncertain. The factors contributing to these large-scale changes in hydroclimate are critical to resolve, given that this region is poised to undergo future anthropogenic-forced climate changes with large uncertainties in model simulations for the 21st century. Furthermore, there are ambiguous constraints on the magnitude and even the sign of changes in key hydroclimate variables between the Last Glacial Maximum and the present day in both proxy reconstructions and climate model analyses of the region. Here we report thermodynamically derived estimates of changes in temperature, precipitation, and evaporation rates, as well as the isotopic composition of lake water, using clumped isotope data from an ancient lake in the northwestern Great Basin, Lake Surprise (California). Compared to modern climate, mean annual air temperature at Lake Surprise was 4.7 °C lower during the Last Glacial Maximum, with decreased evaporation rates and similar precipitation rates to modern. During the mid-deglacial period, the growth of Lake Surprise implied that the lake hydrologic budget briefly departed from steady state. Our reconstructions indicate that this growth took place rapidly, while the subsequent lake regression took place over several thousand years. Using models for precipitation and evaporation constrained from clumped isotope results, we determine that the disappearance of Lake Surprise coincided with a moderate increase in lake temperature, along with increasing evaporation rates outpacing increasing precipitation rates. Concomitant analysis of proxy data and climate model simulations for the Last Glacial Maximum are used to provide a robust means to understand past climate change, and by extension, predict how current hydroclimates may respond to expected future climate forcings. We suggest that an expansion of this analysis to more basins across a larger spatial scale could provide valuable insight into proposed climate forcings, and aid in climate model process depiction. Ultimately, our analysis highlights the importance of temperature-driven evaporation as a mechanism for lake growth and retreat in this region.

Geology↗
NASA NTRS1996

System Synthesis in Preliminary Aircraft Design using Statistical Methods

This paper documents an approach to conceptual and preliminary aircraft design in which system synthesis is achieved using statistical methods, specifically design of experiments (DOE) and response surface methodology (RSM). These methods are employed in order to more efficiently search the design space for optimum configurations. In particular, a methodology incorporating three uses of these techniques is presented. First, response surface equations are formed which represent aerodynamic analyses, in the form of regression polynomials, which are more sophisticated than generally available in early design stages. Next, a regression equation for an overall evaluation criterion is constructed for the purpose of constrained optimization at the system level. This optimization, though achieved in a innovative way, is still traditional in that it is a point design solution. The methodology put forward here remedies this by introducing uncertainty into the problem, resulting a solutions which are probabilistic in nature. DOE/RSM is used for the third time in this setting. The process is demonstrated through a detailed aero-propulsion optimization of a high speed civil transport. Fundamental goals of the methodology, then, are to introduce higher fidelity disciplinary analyses to the conceptual aircraft synthesis and provide a roadmap for transitioning from point solutions to probabalistic designs (and eventually robust ones).

DeLaurentis, Daniel↗
NASA NTRS1996

System Synthesis in Preliminary Aircraft Design Using Statistical Methods

This paper documents an approach to conceptual and early preliminary aircraft design in which system synthesis is achieved using statistical methods, specifically Design of Experiments (DOE) and Response Surface Methodology (RSM). These methods are employed in order to more efficiently search the design space for optimum configurations. In particular, a methodology incorporating three uses of these techniques is presented. First, response surface equations are formed which represent aerodynamic analyses, in the form of regression polynomials, which are more sophisticated than generally available in early design stages. Next, a regression equation for an Overall Evaluation Criterion is constructed for the purpose of constrained optimization at the system level. This optimization, though achieved in an innovative way, is still traditional in that it is a point design solution. The methodology put forward here remedies this by introducing uncertainty into the problem, resulting in solutions which are probabilistic in nature. DOE/RSM is used for the third time in this setting. The process is demonstrated through a detailed aero-propulsion optimization of a High Speed Civil Transport. Fundamental goals of the methodology, then, are to introduce higher fidelity disciplinary analyses to the conceptual aircraft synthesis and provide a roadmap for transitioning from point solutions to probabilistic designs (and eventually robust ones).

DeLaurentis, Daniel↗
NASA NTRS2011

Development of Head Injury Assessment Reference Values Based on NASA Injury Modeling

NASA is developing a new capsule-based, crewed vehicle that will land in the ocean, and the space agency desires to reduce the risk of injury from impact during these landings. Because landing impact occurs for each flight and the crew might need to perform egress tasks, current injury assessment reference values (IARV) were deemed insufficient. Because NASCAR occupant restraint systems are more effective than the systems used to determine the current IARVs and are similar to NASA s proposed restraint system, an analysis of NASCAR impacts was performed to develop new IARVs that may be more relevant to NASA s context of vehicle landing operations. Head IARVs associated with race car impacts were investigated by completing a detailed analysis of all of the 2002-2008 NASCAR impact data. Specific inclusion and exclusion criteria were used to select 4071 impacts from the 4015 recorder files provided (each file could contain multiple impact events). Of the 4071 accepted impacts, 274 were selected for numerical simulation using a custom NASCAR restraint system and Humanetics Hybrid-III 50th percentile numerical dummy model in LS-DYNA. Injury had occurred in 32 of the 274 selected impacts, and 27 of those injuries involved the head. A majority of the head injuries were mild concussions with or without brief loss of consciousness. The 242 non-injury impacts were randomly selected and representative of the range of crash dynamics present in the total set of 4071 impacts. Head dynamics data (head translational acceleration, translational change in velocity, rotational acceleration, rotational velocity, HIC-15, HIC-36, and the Head 3ms clip) were filtered according to SAE J211 specifications and then transformed to a log scale. The probability of head injury was estimated using a separate logistic regression analysis for each log-transformed predictor candidate. Using the log transformation constrains the estimated probability of injury to become negligible as IARVs approach zero. For the parameters head translational acceleration, head translational velocity change, head rotational acceleration, HIC-15, and HIC-36, conservative values (in the lower 95% confidence interval) that gave rise to a 5% risk of any injury occurring were estimated as 40.0 G, 7.9 m/s, 2200 rad/s2, 98.4, and 77.4 respectively. Because NASA is interested in the consequence of any particular injury on the ability of the crew to perform egress tasks, the head injuries that occurred in the NASCAR dataset were classified according to a NASA-developed scale (Classes I - III) for operationally relevant injuries, which classifies injuries on the basis of their operational significance. Additional analysis of the data was performed to determine the probability of each injury class occurring, and this was estimated using an ordered probit model. For head translational acceleration, head translational velocity change, head rotational acceleration, head rotational velocity, HIC-36, and head 3ms clip, conservative values of IARVs that produced a 5% risk of Class II injury were estimated as 50.7 G, 9.5 m/s, 2863 rad/s2, 11.0 rad/s, 30.3, and 46.4 G respectively. The results indicate that head IARVs developed from the NASCAR dataset may be useful to protect crews during landing impact.

Somers, Jeffrey T.↗
NASA NTRS2013

Statistical Issues in Galaxy Cluster Cosmology

The number and growth of massive galaxy clusters are sensitive probes of cosmological structure formation. Surveys at various wavelengths can detect clusters to high redshift, but the fact that cluster mass is not directly observable complicates matters, requiring us to simultaneously constrain scaling relations of observable signals with mass. The problem can be cast as one of regression, in which the data set is truncated, the (cosmology-dependent) underlying population must be modeled, and strong, complex correlations between measurements often exist. Simulations of cosmological structure formation provide a robust prediction for the number of clusters in the Universe as a function of mass and redshift (the mass function), but they cannot reliably predict the observables used to detect clusters in sky surveys (e.g. X-ray luminosity). Consequently, observers must constrain observable-mass scaling relations using additional data, and use the scaling relation model in conjunction with the mass function to predict the number of clusters as a function of redshift and luminosity.

Galaxy↗
NASA NTRS1982

Sensitivity of constrained linear inversions to the selection of the Lagrange multiplier

The influence of the choice of the Lagrange multiplier on constrained linear inversions is explored, with reference made to applications in inferring the columnar aerosol size distributions from spectral aerosol optical depth measurements. A range of the Lagrange multiplier is examined to find all positive solutions for the solution vector, which represents modifying factors to the assumed form of the size distribution. An iterative method is devised to constrain the calculations to consideration of only positive quantities and a requirement that the regression fit to data be consistent with measurement errors. The determination of the variances and covariances is formulated and applied to existing data sets for optical depth. Variances in the solution are found to be large for particle radii when the information content of the data is small.

King, M. D.↗
DOE OSTI2025

Jensen–Shannon divergence based novel loss functions for Bayesian neural networks

Bayesian neural networks (BNNs) are state-of-the-art machine learning methods that can naturally regularize and systematically quantify uncertainties using their stochastic parameters. Kullback–Leibler (KL) divergence-based variational inference used in BNNs suffer from unstable optimization and challenges in approximating light-tailed posteriors due to the unbounded nature of the KL divergence. To resolve these issues, we formulate a novel loss function for BNNs based on a new modification to the generalized Jensen–Shannon (JS) divergence, which is bounded. In addition, we propose a Geometric JS divergence-based loss, which is computationally efficient since it can be evaluated analytically. We found that the JS divergence-based variational inference is intractable, and hence employed a constrained optimization framework to formulate these losses. Our theoretical analysis and empirical experiments on multiple regression and classification data sets suggest that the proposed losses perform better than the KL divergence-based loss, especially when the data sets are noisy or biased. Specifically, there are approximately 5% and 8% improvements in accuracy for a noise-added CIFAR-10 dataset and a regression dataset, respectively. There is about 13% reduction in false negative predictions of a biased histopathology dataset. Additionally, we quantify and compare the uncertainty metrics for the regression and classification tasks.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Reconstructing the Universe: Testing the Mutual Consistency of the Pantheon and SDSS/eBOSS BAO Data Sets with Gaussian Processes

We test the mutual consistency between the baryon acoustic oscillation measurements from the eBOSS SDSS final release and the Pantheon supernova compilation in a model-independent fashion using Gaussian process regression. We also test their joint consistency with the ΛCDM model in a model-independent fashion. We also use Gaussian process regression to reconstruct the expansion history that is preferred by these two data sets. While this methodology finds no significant preference for model flexibility beyond ΛCDM, we are able to generate a number of reconstructed expansion histories that fit the data better than the best-fit ΛCDM model. These example expansion histories may point the way toward modifications to ΛCDM. We also constrain the parameters Ω{sub k} and H {sub 0} r {sub d} both with ΛCDM and with Gaussian process regression. We find that H {sub 0} r {sub d} = 10,030 ± 130 km s{sup −1} and Ω{sub k} = 0.05 ± 0.10 for ΛCDM and that H {sub 0} r {sub d} = 10,040 ± 140 km s{sup −1} and Ω{sub k} = 0.02 ± 0.20 for the Gaussian process case.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗