Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Bayesian approach”

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 1,369 records · Page 76
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Design optimization of an ethanol heavy-duty engine using design of experiments and bayesian optimization

Diesel-fueled engines still hold a large market share in the medium and heavy-duty transportation sector. However, the increase in fossil fuel prices and the strict emission regulations are leading engine manufacturers to seek cleaner alternatives without a compromise in performance. Alcohol-based fuels, such as ethanol, offer a promising alternative to diesel fuel in meeting regulatory demands. Ethanol provides cleaner combustion and lower levels of soot due to its chemical properties, in particular its lower level of carbon content. In addition, the stoichiometric operating conditions of alcohol fueled engines enable the mitigation of NOx emissions in aftertreatment stage. With the promise of retrofitting diesel engines to run on ethanol to reduce emissions, the thermal efficiency of these engines remains the primary optimization target. In order to find the optimal ethanol-fueled engine design that maximizes the thermal efficiency, a large design space needs to be investigated using engineering tools. In this study, previous research by the authors on optimizing the design of a single-cylinder ethanol-fueled engine was extended to explore the design space for a heavy-duty multi-cylinder engine configuration. A heavy-duty engine setup with multiple operating conditions at different engine speeds and loads were considered. A design optimization analysis was performed to identify the potential designs that maximize the indicated thermal efficiency in an ethanol-fueled compression ignition engine. First, a computational fluid dynamics (CFD) model of the engine was validated using experimental data for four drive cycle points. Using a design of experiments (DoE) approach and a parameterized piston bowl geometry, the model was then exercised to explore the relationship among geometric features of the piston bowl and spray targeting angle and indicated thermal efficiency across all tested operating conditions. After evaluating 165~candidate designs, a piston bowl geometry was identified that yielded an increase between 1.3 to 2.2 percentage points in indicated thermal efficiency for all tested conditions, while satisfying the operational design constraints for peak pressure and maximum pressure rise rate. The increased performance was attributed to enhanced mixing that led to the formation of a more homogeneous distribution of in-cylinder temperature and equivalence ratio, higher combustion temperatures, and shorter combustion duration. Finally, a Bayesian optimization (BOpt) analysis was employed to find the optimal piston bowl geometry with a fixed spray injector angle for one of the operating conditions. Using BOpt, a piston candidate was identified that resulted in a 1.9~percentage point increase in thermal efficiency from the baseline design, yet only required 65\% of the design samples investigated using the DoE approach.

Tekgul, Bulut↗
DOE OSTI2025

Accelerating Hamiltonian Monte Carlo for Bayesian inference in neural networks and neural operators

Hamiltonian Monte Carlo (HMC) is a powerful and accurate method to sample from the posterior distribution in Bayesian inference. However, HMC techniques are computationally demanding for Bayesian neural networks due to the high dimensionality of the network’s parameter space and the non-convexity of their posterior distributions. Therefore, various approximation techniques, such as variational inference (VI) or stochastic gradient MCMC, are often employed to infer the posterior distribution of the network parameters. Such approximations introduce inaccuracies in the inferred distributions, resulting in unreliable uncertainty estimates. In this work, we propose a hybrid approach that combines inexpensive VI and accurate HMC methods to efficiently and accurately quantify uncertainties in neural networks and neural operators. The proposed approach leverages an initial VI training on the full network. We examine the influence of individual parameters on the prediction uncertainty, which shows that a large proportion of the parameters do not contribute substantially to uncertainty in the network predictions. This information is then used to significantly reduce the dimension of the parameter space, and HMC is performed only for the subset of network parameters that strongly influence prediction uncertainties. This yields a framework for accelerating the full batch HMC for posterior inference in neural networks. We demonstrate the efficiency and accuracy of the proposed framework on deep neural networks and operator networks, showing that inference can be performed for large networks with tens to hundreds of thousands of parameters. Finally, we show that this method can effectively learn surrogates for complex physical systems by modeling the operator that maps from upstream conditions to wall-pressure data on a cone in hypersonic flow.

Bayesian inference↗
DOE OSTI2015

A Bayesian framework for adaptive selection, calibration, and validation of coarse-grained models of atomistic systems

A general adaptive modeling algorithm for selection and validation of coarse-grained models of atomistic systems is presented. A Bayesian framework is developed to address uncertainties in parameters, data, and model selection. Algorithms for computing output sensitivities to parameter variances, model evidence and posterior model plausibilities for given data, and for computing what are referred to as Occam Categories in reference to a rough measure of model simplicity, make up components of the overall approach. Computational results are provided for representative applications.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2022

Data augmentation for disruption prediction via robust surrogate models

The goal of this work is to generate large statistically representative data sets to train machine learning models for disruption prediction provided by data from few existing discharges. Such a comprehensive training database is important to achieve satisfying and reliable prediction results in artificial neural network classifiers. Here, we aim for a robust augmentation of the training database for multivariate time series data using Student t process regression. We apply Student t process regression in a state space formulation via Bayesian filtering to tackle challenges imposed by outliers and noise in the training data set and to reduce the computational complexity. Thus, the method can also be used if the time resolution is high. We use an uncorrelated model for each dimension and impose correlations afterwards via colouring transformations. We demonstrate the efficacy of our approach on plasma diagnostics data of three different disruption classes from the DIII-D tokamak. To evaluate if the distribution of the generated data is similar to the training data, we additionally perform statistical analyses using methods from time series analysis, descriptive statistics and classic machine learning clustering algorithms.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2012

Scaling cosmology with variable dark-energy equation of state

Interactions between dark matter and dark energy which result in a power-law behavior (with respect to the cosmic scale factor) of the ratio between the energy densities of the dark components (thus generalizing the ΛCDM model) have been considered as an attempt to alleviate the cosmic coincidence problem phenomenologically. We generalize this approach by allowing for a variable equation of state for the dark energy within the CPL-parametrization. Based on analytic solutions for the Hubble rate and using the Constitution and Union2 SNIa sets, we present a statistical analysis and classify different interacting and non-interacting models according to the Akaike (AIC) and the Bayesian (BIC) information criteria. We do not find noticeable evidence for an alleviation of the coincidence problem with the mentioned type of interaction.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2018

Quantitative Constraints on the Reionization History from the IGM Damping Wing Signature in Two Quasars at z > 7

During reionization, neutral hydrogen in the intergalactic medium (IGM) imprints a damping wing absorption feature on the spectrum of high-redshift quasars. A detection of this signature provides compelling evidence for a significantly neutral universe, and enables measurements of the hydrogen neutral fraction x HI (z) at that epoch. Obtaining reliable quantitative constraints from this technique, however, is challenging due to stochasticity induced by the patchy inside-out topology of reionization, degeneracies with quasar lifetime, and the unknown unabsorbed quasar spectrum close to rest-frame Lyα. Here, we combine a large-volume semi-numerical simulation of reionization topology with 1D radiative transfer through high-resolution hydrodynamical simulations of the high-redshift universe to construct models of quasar transmission spectra during reionization. Our state-of-the-art approach captures the distribution of damping wing strengths in biased quasar halos that should have reionized earlier, as well as the erosion of neutral gas in the quasar environment caused by its own ionizing radiation. Combining this detailed model with our new technique for predicting the quasar continuum and its associated uncertainty, we introduce a Bayesian statistical method to jointly constrain the neutral fraction of the universe and the quasar lifetime from individual quasar spectra. We apply this methodology to the spectra of the two quasars with the highest redshifts known, ULAS J1120+0641 and ULAS J1342+0928, and measure volume-averaged neutral fractions {x HI }(Z = 7.09) = 0.48 $+0.26\atop{-0.26}$and {x HI }(Z = 7.54) = 0.60 $+0.20\atop{-0.23}$ (posterior medians and 68% credible intervals) when marginalized over quasar lifetimes of 10 3 ≤ t q ≤ 10 8 yr.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2022

Data augmentation for disruption prediction via robust surrogate models

The goal of this work is to generate large statistically representative datasets to train machine learning models for disruption prediction provided by data from few existing discharges. Such a comprehensive training database is important to achieve satisfying and reliable prediction results in artificial neural network classifiers. Here, we aim for a robust augmentation of the training database for multivariate time series data using Student-t process regression. We apply Student-t process regression in a state space formulation via Bayesian filtering to tackle challenges imposed by outliers and noise in the training data set and to reduce the computational complexity. Thus, the method can also be used if the time resolution is high. We use an uncorrelated model for each dimension and impose correlations afterwards via coloring transformations. We demonstrate the efficacy of our approach on plasma diagnostics data of three different disruption classes from the DIII-D tokamak. To evaluate if the distribution of the generated data is similar to the training data, we additionally perform statistical analyses using methods from time series analysis, descriptive statistics, and classic machine learning clustering algorithms.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1986

Application of physical parameter identification to finite element models

A time domain technique for matching response predictions of a structural dynamic model to test measurements is developed. Significance is attached to prior estimates of physical model parameters and to experimental data. The Bayesian estimation procedure allows confidence levels in predicted physical and modal parameters to be obtained. Structural optimization procedures are employed to minimize an error functional with physical model parameters describing the finite element model as design variables. The number of complete FEM analyses are reduced using approximation concepts, including the recently developed convoluted Taylor series approach. The error function is represented in closed form by converting free decay test data to a time series model using Prony' method. The technique is demonstrated on simulated response of a simple truss structure.

Bronowicki, Allen J.↗
DOE OSTI2018

Calibration of Multi-Parameter Models of Avascular Tumor Growth Using Time Resolved Microscopy Data

Abstract Two of the central challenges of using mathematical models for predicting the spatiotemporal development of tumors is the lack of appropriate data to calibrate the parameters of the model, and quantitative characterization of the uncertainties in both the experimental data and the modeling process itself. We present a sequence of experiments, with increasing complexity, designed to systematically calibrate the rates of apoptosis, proliferation, and necrosis, as well as mobility, within a phase-field tumor growth model. The in vitro experiments characterize the proliferation and death of human liver carcinoma cells under different initial cell concentrations, nutrient availabilities, and treatment conditions. A Bayesian framework is employed to quantify the uncertainties in model parameters. The average difference between the calibration and the data, across all time points is between 11.54% and 14.04% for the apoptosis experiments, 7.33% and 23.30% for the proliferation experiments, and 8.12% and 31.55% for the necrosis experiments. The results indicate the proposed experiment-computational approach is generalizable and appropriate for step-by-step calibration of multi-parameter models, yielding accurate estimations of model parameters related to rates of proliferation, apoptosis, and necrosis.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2025

Parallelizing autotuning for HPC applications: Unveiling the potential of the speculation strategy in Bayesian optimization

In the exascale computing era, tuning High-Performance Computing (HPC) applications has become a significant computational challenge. Although Bayesian optimization (BO) has emerged as a promising tool for HPC performance tuning, the BO workflow is inherently sequential (i.e., one function evaluation at a time) and cannot leverage the huge amount of parallel resources present in modern supercomputers, resulting in a considerable underutilization of their computational capabilities. This paper explores the trade-off between search quality and parallelism in BO, investigating a diverse set of methods. Building upon both previous approaches from the literature and novel methodologies introduced in this work, our study provides a deep analysis to accelerate BO performance tuning. By examining a set of synthetic functions and practical HPC applications, our exploration analyzes the interaction among various BO methods for parallelization, the quantity of parallel resources, the runtime distribution of target HPC applications, and the costs associated with different search orchestration mechanisms that have been overlooked in previous studies. Compared to sequential BO, our novel methodology achieves comparable quality while demonstrating robust scalability in search time as the amount of parallel resources increases; it also outperforms a state-of-the-art tuner, which supports parallelization, achieving up to 3.67x faster search time. We provide high-value insights for practitioners seeking to leverage the power of parallel computing for efficient HPC application tuning. Additionally, to further assist researchers in accelerating the performance tuning of their HPC applications, we provide an extension of an existing open-source tuning framework that incorporates our methods.

Bayesian optimization↗
DOE OSTI2025

Compactly‐Supported Nonstationary Kernels for Computing Exact Gaussian Processes on Big Data

The Gaussian process (GP) is a widely used method for analyzing large-scale data sets, including spatio-temporal measurements of nonlinear processes that are now commonplace in the environmental sciences. Traditional implementations of GPs involve stationary kernels (also termed covariance functions) that limit their flexibility, and exact methods for inference that prevent application to data sets with more than about 10,000 points. Modern approaches to address stationarity assumptions generally fail to accommodate large data sets, while all attempts to address scalability focus on approximating the Gaussian likelihood, which can involve subjectivity and lead to inaccuracies. In this work, we explicitly derive an alternative kernel that can discover and encode both sparsity and nonstationarity. We embed the kernel within a fully Bayesian GP model and leverage high-performance computing resources to enable the analysis of massive data sets. We demonstrate the favorable performance of our novel kernel relative to existing exact and approximate GP methods across a variety of synthetic data examples. Furthermore, we conduct space–time prediction based on more than 1 million measurements of daily maximum temperature and verify that our results outperform state-of-the-art methods in the Earth sciences. More broadly, having access to exact GPs that use ultra-scalable, sparsity-discovering, nonstationary kernels allows GP methods to truly compete with a wide variety of machine learning methods.

Gaussian processes↗
DOE OSTI2012

ATMOSPHERIC RETRIEVAL FOR SUPER-EARTHS: UNIQUELY CONSTRAINING THE ATMOSPHERIC COMPOSITION WITH TRANSMISSION SPECTROSCOPY

We present a retrieval method based on Bayesian analysis to infer the atmospheric compositions and surface or cloud-top pressures from transmission spectra of exoplanets with general compositions. In this study, we identify what can unambiguously be determined about the atmospheres of exoplanets from their transmission spectra by applying the retrieval method to synthetic observations of the super-Earth GJ 1214b. Our approach to inferring constraints on atmospheric parameters is to compute their joint and marginal posterior probability distributions using the Markov Chain Monte Carlo technique in a parallel tempering scheme. A new atmospheric parameterization is introduced that is applicable to general atmospheres in which the main constituent is not known a priori and clouds may be present. Our main finding is that a unique constraint of the mixing ratios of the absorbers and two spectrally inactive gases (such as N{sub 2} and primordial H{sub 2}+ He) is possible if the observations are sufficient to quantify both (1) the broadband transit depths in at least one absorption feature for each absorber and (2) the slope and strength of the molecular Rayleigh scattering signature. A second finding is that the surface pressure or cloud-top pressure can be quantified if a surface or cloud deck is present at low optical depth. A third finding is that the mean molecular mass can be constrained by measuring either the Rayleigh scattering slope or the shapes of the absorption features, thus enabling one to distinguish between cloudy hydrogen-rich atmospheres and high mean molecular mass atmospheres. We conclude, however, that without the signature of molecular Rayleigh scattering-even with robustly detected infrared absorption features (>10{sigma})-there is no reliable way to tell from the transmission spectrum whether the absorber is a main constituent of the atmosphere or just a minor species with a mixing ratio of X{sub abs} < 0.1%. The retrieval method leads us to a conceptual picture of which details in transmission spectra are essential for unique characterizations of well-mixed exoplanet atmospheres.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2019

Reexamining the proton-radius problem using constrained Gaussian processes

The “proton radius puzzle” refers to an 8-year-old problem that highlights major inconsistencies in the extraction of the charge radius of the proton from muonic Lamb-shift experiments as compared against experiments using elastic electron scattering. For the latter approach, the determination of the charge radius involves an extrapolation of the experimental form factor to zero momentum transfer. To estimate the proton radius, a novel and powerful nonparametric method based on a constrained Gaussian process is introduced. Furthermore, the constrained Gaussian process models the electric form factor as a function of the momentum transfer. Within a Bayesian paradigm, we develop a model flexible enough to fit the data without any parametric assumptions on the form factor. The Bayesian estimation is guided by imposing only two physical constraints on the form factor: (a) its value at zero momentum transfer (normalization) and (b) its overall shape, assumed to be a monotonically decreasing function of the momentum transfer. Variants of these assumptions are explored to assess their impact.By adopting both constraints and incorporating the whole range of experimental data available we extracted a charge radius of r p = 0.845 ± 0.001 fm, consistent with the muonic experiment. Nevertheless, we show that within our model the extracted radius depends on both the assumed constraints and the range of experimental data used to fit the Gaussian process. For example, if only low-momentum-transfer data are used, relaxing the normalization constraint provides a value compatible with the larger electronic value. We have presented a novel technique to estimate the proton radius from electron-scattering data based on a constrained Gaussian process. We demonstrated that the impact of imposing sensible physical constraints on the form factor is substantial. Also critical is the range of the experimental data used in the extrapolation. We are hopeful that as the technique gets refined, together with the anticipated new results from the PRad experiment, we will get closer to a resolution of the puzzle.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
NASA NTRS2018

A Survey of Methods for Computing Best Estimates of Endoatmospheric and Exoatmospheric Trajectories

Beginning with the mathematical prediction of planetary orbits in the early seventeenth century up through the most recent developments in sensor fusion methods, many techniques have emerged that can be employed on the problem of endo and exoatmospheric trajectory estimation. Although early methods were ad hoc, the twentieth century saw the emergence of many systematic approaches to estimation theory that produced a wealth of useful techniques. The broad genesis of estimation theory has resulted in an equally broad array of mathematical principles, methods and vocabulary. Among the fundamental ideas and methods that are briefly touched on are batch and sequential processing, smoothing, estimation, and prediction, sensor fusion, sensor fusion architectures, data association, Bayesian and non Bayesian filtering, the family of Kalman filters, models of the dynamics of the phases of a rocket's flight, and asynchronous, delayed, and asequent data. Along the way, a few trajectory estimation issues are addressed and much of the vocabulary is defined.

Bernard, William P.↗
DOE OSTI2024

Measuring the robustness of predictive probability for early stopping in two-group comparisons

We report physical experiments are often expensive and time-consuming. Test engineers must certify the compatibility of aircraft and their weapon systems before they can be deployed in the field, but the testing required is time consuming, expensive, and resource limited. Adopting Bayesian adaptive designs is a promising way to borrow from the successes seen in the clinical trials domain. The use of predictive probability (PP) to stop testing early and make faster decisions is particularly appealing given the aforementioned constraints. Given the high-consequence nature of the tests performed in the national security space, a strong understanding of new methods is required before being deployed. Although PP has been thoroughly studied for binary data, there is less work with continuous data, where many reliability studies are interested in certifying the specification limits of components. A simulation study evaluating the robustness of this approach indicates early stopping based on PP is reasonably robust to minor assumption violations, especially when only a few interim analyses are conducted. The simulation study also compares PP to conditional power, showing its relative strengths and weaknesses. A post-hoc analysis exploring whether release requirements of a weapon system from an aircraft are within specification with desired reliability resulted in stopping the experiment early and saving 33% of the experimental runs.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Global stochastic optimization of stellarator coil configurations

In the construction of a stellarator, the manufacturing and assembling of the coil system is a dominant cost. These coils need to satisfy strict engineering tolerances, and if those are not met the project could be cancelled as in the case of the National Compact Stellarator Experiment (NCSX) project. Therefore, our goal is to find coil configurations that increase construction tolerances without compromising the performance of the magnetic field. In this paper, we develop a gradient-based stochastic optimization model which seeks robust stellarator coil configurations in high dimensions. In particular, we design a two-step method: first, we perform an approximate global search by a sample efficient trust-region Bayesian optimization; second, we refine the minima found in step one with a stochastic local optimizer. To this end, we introduce two stochastic local optimizers: BFGS applied to the sample average approximation; and Adam, equipped with a control variate for variance reduction. Numerical simulations performed on a W7-X-like coil configuration demonstrate that our global optimization approach finds a variety of promising local solutions at less than 0.1% of the cost of previous work, which considered solely local stochastic optimization.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2020

Accelerating quantum optics experiments with statistical learning

Quantum optics experiments, involving the measurement of low-probability photon events, are known to be extremely time-consuming. We present a methodology for accelerating such experiments using physically motivated ansatzes together with simple statistical learning techniques such as Bayesian maximum a posteriori estimation based on few-shot data. We show that it is possible to reconstruct time-dependent data using a small number of detected photons, allowing for fast estimates in under a minute and providing a one-to-two order of magnitude speed-up in data acquisition time. We test our approach using real experimental data to retrieve the second order intensity correlation function, G (2) ($τ$), as a function of time delay τ between detector counts, for thermal light as well as anti-bunched light emitted by a quantum dot driven by periodic laser pulses. The proposed methodology has a wide range of applicability and has the potential to impact the scientific discovery process across a multitude of domains.

36 MATERIALS SCIENCE↗
DOE OSTI2013

An Approach for Assessing the Signature Quality of Various Chemical Assays when Predicting the Culture Media Used to Grow Microorganisms

We demonstrate an approach for assessing the quality of a signature system designed to predict the culture medium used to grow a microorganism. The system was comprised of four chemical assays designed to identify various ingredients that could be used to produce the culture medium. The analytical measurements resulting from any combination of these four assays can be used in a Bayesian network to predict the probabilities that the microorganism was grown using one of eleven culture media. We evaluated combinations of the signature system by removing one or more of the assays from the Bayes network. We measured and compared the quality of the various Bayes nets in terms of fidelity, cost, risk, and utility, a method we refer to as Signature Quality Metrics

Holmes, Aimee E.↗