Search NASA⌕ Search

SEARCH · Search NASA

Results for “Bayesian statistics”

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 955 records · Page 53

Measurement of the electron antineutrino mass in tritium beta decay in the Troitsk nu-mass experiment

The results obtained in the Troitsk nu-mass experiment by measuring the electron-antineutrino mass in tritium beta decay are presented. The facility used consists of a gaseous windowless tritium source and an electrostatic electron spectrometer involving an adiabatic magnetic collimation. Runs in which measurement conditions were reliably established were thoroughly selected in analyzing data obtained from 1994 to 2004. All known systematic effects were taken into account. For the square of the electron-antineutrino mass, the treatment of measured spectra yielded the following result: m{sub {nu}}{sup 2} = -0.67 {+-} 1.89{sub stat.} {+-} 1.68{sub syst.} eV{sup 2}. The use of the Bayesian method and the Feldman-Cousins unified approach made it possible to obtain the following upper limits on the mass: m{sub {nu}} < 2.12 eV (at a 95% C.L.; Bayesian method) and m{sub {nu}} < 2.05 eV (at a 95% C.L., Feldman-Cousins method). At the same time, an estimation of the sensitivity limit without allowance for negative values of the square of the mass leads to m{sub {nu}} < 2.2 eV (at a 95% C.L.). Measured spectra were analyzed for the possible existence of an additional structure (step) in the electron spectrum near the boundary energy. The conclusion drawn from this analysis was that, within the existing statistical errors, there are no reasons for introducing such a feature.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Modeling Forest Biomass and Growth: Coupling Long-Term Inventory and Lidar Data

Combining spatially-explicit long-term forest inventory and remotely sensed information from Light Detection and Ranging (LiDAR) datasets through statistical models can be a powerful tool for predicting and mapping above-ground biomass (AGB) at a range of geographic scales. We present and examine a novel modeling approach to improve prediction of AGB and estimate AGB growth using LiDAR data. The proposed model accommodates temporal misalignment between field measurements and remotely sensed data-a problem pervasive in such settings-by including multiple time-indexed measurements at plot locations to estimate AGB growth. We pursue a Bayesian modeling framework that allows for appropriately complex parameter associations and uncertainty propagation through to prediction. Specifically, we identify a space-varying coefficients model to predict and map AGB and its associated growth simultaneously. The proposed model is assessed using LiDAR data acquired from NASA Goddard's LiDAR, Hyper-spectral & Thermal imager and field inventory data from the Penobscot Experimental Forest in Bradley, Maine. The proposed model outperformed the time-invariant counterpart models in predictive performance as indicated by a substantial reduction in root mean squared error. The proposed model adequately accounts for temporal misalignment through the estimation of forest AGB growth and accommodates residual spatial dependence. Results from this analysis suggest that future AGB models informed using remotely sensed data, such as LiDAR, may be improved by adapting traditional modeling frameworks to account for temporal misalignment and spatial dependence using random effects.

Babcock, Chad↗

Online Multi-Modal Learning and Adaptive Information Trajectory Planning for Autonomous Exploration

In robotic information gathering missions, scientists are typically interested in understanding variables which require proxy measurements from specialized sensor suites to estimate. However, energy and time constraints limit how often these sensors can be used in a mission. Robots are also equipped with cheaper to use navigation sensors such as cameras. In this paper, we explore a challenging planning problem in which a robot is required to learn about a scientific variable of interest in an initially unknown environment by planning informative paths and deciding when and where to use its sensors. To tackle this we present two innovations: a Bayesian generative model framework to automatically learn correlations between expensive science sensors and cheaper to use navigation sensors online, and a sampling based approach to plan for multiple sensors while handling long horizons and budget constraints. Our approach does not grow in complexity with data and is anytime making it highly applicable to field robotics. We tested our approach extensively in simulation and validated it with real data collected during the 2014 Mojave Volatiles Prospector Mission. Our planning algorithm performs statistically significantly better than myopic approaches and at least as well as a coverage-based algorithm in an initially unknown environment while having added advantages of being able to exploit prior knowledge and handle other intricacies of the real world without further algorithmic modifications.

learning↗

Determining the neutrino mass hierarchy with cosmology

The combination of current large-scale structure and cosmic microwave background anisotropies data can place strong constraints on the sum of the neutrino masses. Here we show that future cosmic shear experiments, in combination with cosmic microwave background constraints, can provide the statistical accuracy required to answer questions about differences in the mass of individual neutrino species. Allowing for the possibility that masses are nondegenerate we combine Fisher matrix forecasts for a weak lensing survey like Euclid with those for the forthcoming Planck experiment. Under the assumption that neutrino mass splitting is described by a normal hierarchy we find that the combination Planck and Euclid will possibly reach enough sensitivity to put a constraint on the mass of a single species. Using a Bayesian evidence calculation we find that such future experiments could provide strong evidence for either a normal or an inverted neutrino hierarchy. Finally we show that if a particular neutrino hierarchy is assumed then this could bias cosmological parameter constraints, for example, the dark energy equation of state parameter, by > or approx. 1{sigma}, and the sum of masses by 2.3{sigma}. We finally discuss the impact of uncertainties on the theoretical modeling of nonlinearities. The results presented in this analysis are obtained under an approximation to the nonlinear power spectrum. This significant source of uncertainty needs to be addressed in future work.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

The footprints of visual attention in the Posner cueing paradigm revealed by classification images

In the Posner cueing paradigm, observers' performance in detecting a target is typically better in trials in which the target is present at the cued location than in trials in which the target appears at the uncued location. This effect can be explained in terms of a Bayesian observer where visual attention simply weights the information differently at the cued (attended) and uncued (unattended) locations without a change in the quality of processing at each location. Alternatively, it could also be explained in terms of visual attention changing the shape of the perceptual filter at the cued location. In this study, we use the classification image technique to compare the human perceptual filters at the cued and uncued locations in a contrast discrimination task. We did not find statistically significant differences between the shapes of the inferred perceptual filters across the two locations, nor did the observed differences account for the measured cueing effects in human observers. Instead, we found a difference in the magnitude of the classification images, supporting the idea that visual attention changes the weighting of information at the cued and uncued location, but does not change the quality of processing at each individual location.

Non-NASA Center↗

THE HUNT FOR EXOMOONS WITH KEPLER (HEK). V. A SURVEY OF 41 PLANETARY CANDIDATES FOR EXOMOONS

We present a survey of 41 Kepler Objects of Interest (KOIs) for exomoons using Bayesian photodynamics, more than tripling the number of KOIs surveyed with this technique. We find no compelling evidence for exomoons although 13 KOIs yield spurious detections driven by instrumental artifacts, stellar activity, and/or perturbations from unseen bodies. Regarding the latter, we find seven KOIs exhibiting >5 σ evidence of transit timing variations, including the “mega-Earth” Kepler-10c, likely indicating an additional planet in that system. We exploit the moderately large sample of 57 unique KOIs surveyed to date to infer several useful statistics. For example, although there is a diverse range in sensitivities, we find that we are sensitive to Pluto–Charon mass-ratio systems for ≃40% of KOIs studied and Earth–Moon mass-ratios for 1 in 8 cases. In terms of absolute mass, our limits probe down to 1.7 Ganymede masses, with a sensitivity to Earth-mass moons for 1 in 3 cases studied and to the smallest moons capable of sustaining an Earth-like atmosphere (0.3 M{sub ⨁}) for 1 in 4. Despite the lack of positive detections to date, we caution against drawing conclusions yet, since our most interesting objects remain under analysis. Finally, we point out that had we searched for the photometric transit signals of exomoons alone, rather than using photodynamics, we estimate that 1 in 4 KOIs would have erroneously been concluded to harbor exomoons due to residual time correlated noise in the Kepler data, posing a serious problem for alternative methods.

79 ASTRONOMY AND ASTROPHYSICS↗

Quantifying bias due to non-Gaussian foregrounds in an optimal reconstruction of CMB lensing and temperature power spectra

We estimate the magnitude of the bias due to non-Gaussian extragalactic foregrounds on the optimal reconstruction of the cosmic microwave background (CMB) lensing potential and temperature power spectra. The reconstruction is performed using a Bayesian inference method known as the marginal unbiased score expansion (MUSE). We apply MUSE to a minimum variance combination of multifrequency maps drawn from the Agora publicly available simulations of the lensed CMB and correlated extragalactic foreground emission. Taking noise levels appropriate to the SPT-3G D1 release, we find non-Gaussian foregrounds may bias the MUSE reconstruction of the lensing potential amplitude at the level of (0.7 ± 0.3)σ when using modes up to ℓ max = 3500. We do not detect a statistically significant bias, finding a value of (-0.4 ± 0.3)σ, when restricted to lower angular multipoles, ℓ max = 3000. This work is a first step toward understanding the impact of extragalactic foregrounds on optimal reconstructions of CMB temperature and lensing potential power spectra.

Statistical sampling techniques↗

Characteristics of infrasound signals from North Korean underground nuclear explosions on 2016 January 6 and September 9

This study investigates infrasound signals from North Korean underground nuclear explosions on 2016 January 6 at 01:30:01 UTC (UNEJ16) and on 2016 September 9 at 00:30:01 UTC (UNES16). Infrasound observations from eight, seismo-acoustic arrays at distances of 300–600 km from the explosions and aligned approximately east–west across South Korea are used in the analysis. Progressive multichannel correlation was used followed by analyst review to identify the infrasound arrivals with estimates of azimuth, phase velocity, celerity, F-statistic, signal-to-noise ratio, correlation and signal duration. These observations are compared with model predictions of travel path, ray turning height and relative amplitudes, using Ground-to-Space atmospheric specifications at the time of each explosion. Stratospheric conditions at the time of the UNEJ16 are favourable to the Korean arrays when all arrays detected infrasound signals, while the few detections for UNES16 are consistent with unfavourable stratospheric winds and reduced amplitude thermospheric arrivals. Infrasound locations of the explosions were estimated using the Bayesian Infrasonic Source Location method (BISL), with wind-corrected backazimuth estimates based on the ray tracing calculations. Adding wind corrections to BISL resulted in infrasound locations up to 50 per cent closer to the seismic epicentre over uncorrected locations. The UNES16 infrasound location estimate using atmospheric corrections had larger differences (~67 km difference from seismic location) than the UNEJ16 location (~11 km difference from seismic location) as a result of the reduced number of observations and poor azimuth coverage for UNES16. Based on stratospheric phases recorded at six arrays and the empirical yield-scaling relations, wind corrected amplitudes from UNEJ16 lead to equivalent infrasound source energy estimates between 0.9 and 16.1 tons of TNT detonated in the atmosphere with an average of 6.4 tons and a standard deviation of 4.6 tons.

58 GEOSCIENCES↗

Cosmological Evidence Modelling: a new simulation-based approach to constrain cosmology on non-linear scales

Extracting accurate cosmological information from galaxy–galaxy and galaxy–matter correlation functions on non-linear scales (${\lesssim } 10 \, h^{-1}{\rm {Mpc}}$) requires cosmological simulations. Additionally, one has to marginalize over several nuisance parameters of the galaxy–halo connection. However, the computational cost of such simulations prohibits naive implementations of stochastic posterior sampling methods like Markov chain Monte Carlo (MCMC) that would require of order $\mathcal {O}(10^6)$ samples in cosmological parameter space. Several groups have proposed surrogate models as a solution: a so-called emulator is trained to reproduce observables for a limited number of realizations in parameter space. Afterwards, this emulator is used as a surrogate model in an MCMC analysis. In this work, we demonstrate a different method called Cosmological Evidence Modelling (CEM). First, for each simulation, we calculate the Bayesian evidence marginalized over the galaxy–halo connection by repeatedly populating the simulation with galaxies. We show that this Bayesian evidence is directly related to the posterior probability of cosmological parameters. Finally, we build a physically motivated model for how the evidence depends on cosmological parameters as sampled by the simulations. We demonstrate the feasibility of CEM by using simulations from the Aemulus simulation suite and forecasting cosmological constraints from BOSS CMASS measurements of redshift-space distortions. Our analysis includes exploration of how galaxy assembly bias affects cosmological inference. Overall, CEM has several potential advantages over the more common approach of emulating summary statistics, including the ability to easily marginalize over highly complex models of the galaxy–halo connection and greater accuracy, thereby reducing the number of simulations required.

79 ASTRONOMY AND ASTROPHYSICS↗

Designing an Optimal Sensor Network via Minimizing Information Loss

Optimal experimental design is a classic topic in statistics, with many well-studied problems, applications, and solutions. The design problem we study is the placement of sensors to monitor spatiotemporal processes, explicitly accounting for the temporal dimension in our modeling and optimization. We observe that recent advancements in computational sciences often yield large datasets based on physics-based simulations, which are rarely leveraged in experimental design. We introduce a novel model-based sensor placement criterion, along with a highly-efficient optimization algorithm, which integrates physics-based simulations and Bayesian experimental design principles to identify sensor networks that “minimize information loss” from simulated data. Our technique relies on sparse variational inference and (separable) Gauss-Markov priors, and thus may adapt many techniques from Bayesian experimental design. We validate our method through a case study monitoring air temperature in Phoenix, Arizona, using state-of-the-art physics-based simulations. Our results show our framework to be superior to random or quasi-random sampling, particularly with a limited number of sensors. We conclude by discussing practical considerations and implications of our framework, including more complex modeling tools and real-world deployments.

54 ENVIRONMENTAL SCIENCES↗

Model-Based Localizatiion in a Shallow Ocean Environment: A Sequential Bayesian/Optimization Approach

The shallow ocean is a dynamically changing medium leading to nonstationary statistical behavior when subjected to temperature, wind, surface variations, bottom interactions, noise and extraneous disturbances as well as other conditions that render it a uniquely challenging environment — especially from a signal processing perspective. Processors must account for and adapt to such instantaneous changes in order to be effective. Thus, a processor is required to “adapt” to these environmental variations while simultaneously providing meaningful estimates that are necessary for such applications as detection, localization, inversion and enhancement. In this paper, we develop a parametrically adaptive, sequential Bayesian processor capable of jointly estimating both modal functions and environmental parameters to provide enhanced estimates for a focused optimizer capable locating a target in the noisy shallow ocean environment.

42 ENGINEERING↗

Magnetic Inflation and Stellar Mass. II. On the Radii of Single, Rapidly Rotating, Fully Convective M-Dwarf Stars

Main-sequence, fully convective M dwarfs in eclipsing binaries are observed to be larger than stellar evolutionary models predict by as much as 10%–15%. A proposed explanation for this discrepancy involves effects from strong magnetic fields, induced by rapid rotation via the dynamo process. Although, a handful of single, slowly rotating M dwarfs with radius measurements from interferometry also appear to be larger than models predict, suggesting that rotation or binarity specifically may not be the sole cause of the discrepancy. We test whether single, rapidly rotating, fully convective stars are also larger than expected by measuring their Rsin i distribution. We combine photometric rotation periods from the literature with rotational broadening (vsin i) measurements reported in this work for a sample of 88 rapidly rotating M dwarf stars. Using a Bayesian framework, we find that stellar evolutionary models underestimate the radii by 10%--15%{sub −2.5}{sup +3}, but that at higher masses (0.18 < M < 0.4 M {sub Sun}), the discrepancy is only about 6% and comparable to results from interferometry and eclipsing binaries. At the lowest masses (0.08 < M < 0.18 M {sub Sun}), we find that the discrepancy between observations and theory is 13%–18%, and we argue that the discrepancy is unlikely to be due to effects from age. Furthermore, we find no statistically significant radius discrepancy between our sample and the handful of M dwarfs with interferometric radii. We conclude that neither rotation nor binarity are responsible for the inflated radii of fully convective M dwarfs, and that all fully convective M dwarfs are larger than models predict.

79 ASTRONOMY AND ASTROPHYSICS↗

Profiles of Gamma-Ray Bursts and Their Component Pulses

One physically informative regularity of their otherwise heterogeneous ensemble, is that many Gamma-Ray Bursts consist of well defined pulses. To objectively quantify the temporal structure of BATSE bursts, we have developed an automatic modeling procedure that separates overlapping pulses and determines the energy-dependence of the pulse-shape parameters. No binning of photon arrival times is needed, so when applied to time-tagged events (TTE) the procedure captures variability information down to the shortest time scales present in the raw data. Maximizing the Bayesian likelihood function Pr(data/model) yields estimates of the model parameters, including the number of pulses present, and allows intercomparison of models of different forms. As with any nonlinear optimization, good initial guesses are crucial to avoid convergence to undesirable local minima. We find excellent initial pulse decompositions by wavelet-denoising a cumulative distribution of the raw photon arrival data; differentiation then gives a time profile mostly free of the systematic effects of degraded resolution (as in ordinary Fourier smoothing) and binning. We present statistical information on pulse rise-time, decay-time, peakedness, and amplitudes, plus their energy dependences - both within a single burst and for a large ensemble of bursts.

Scargle, Jeff D.↗

Bayesian Networks for Social Modeling

This paper describes a body of work developed over the past five years. The work addresses the use of Bayesian network (BN) models for representing and predicting social/organizational behaviors. The topics covered include model construction, validation, and use. These topics show the bulk of the lifetime of such model, beginning with construction, moving to validation and other aspects of model ‘critiquing’, and finally demonstrating how the modeling approach might be used to inform policy analysis. To conclude, we discuss limitations of using BN for this activity and suggest remedies to address those limitations. The primary benefits of using a well-developed computational, mathematical, and statistical modeling structure, such as BN, are 1) there are significant computational, theoretical and capability bases on which to build 2) ability to empirically critique the model, and potentially evaluate competing models for a social/behavioral phenomena.

99 GENERAL AND MISCELLANEOUS↗

KaRMMa – kappa reconstruction for mass mapping

ABSTRACT We present KaRMMa, a novel method for performing mass map reconstruction from weak-lensing surveys. We employ a fully Bayesian approach with a physically motivated lognormal prior to sample from the posterior distribution of convergence maps. We test KaRMMa on a suite of dark matter N-body simulations with simulated DES Y1-like shear observations. We show that KaRMMa outperforms the basic Kaiser–Squires mass map reconstruction in two key ways: (1) our best map point estimate has lower residuals compared to Kaiser–Squires; and (2) unlike the Kaiser–Squires reconstruction, the posterior distribution of KaRMMa maps is nearly unbiased in all summary statistics we considered, namely: one-point and two-point functions, and peak/void counts. In particular, KaRMMa successfully captures the non-Gaussian nature of the distribution of κ values in the simulated maps. We further demonstrate that the KaRMMa posteriors correctly characterize the uncertainty in all summary statistics we considered.

79 ASTRONOMY AND ASTROPHYSICS↗

Large-Scale Uncertainty and Error Analysis for Time-dependent Fluid/Structure Interactions in Wind Turbine Applications

The following is the final report covering the entire period of this aforementioned grant, June 1, 2011 - May 31, 2013 for the portion of the effort corresponding to Stanford University (SU). SU has partnered with Sandia National Laboratories (PI: Mike S. Eldred) and Purdue University (PI: Dongbin Xiu) to complete this research project and this final report includes those contributions made by the members of the team at Stanford. Dr. Eldred is continuing his contributions to this project under a no-cost extension and his contributions to the overall effort will be detailed at a later time (once his effort has concluded) on a separate project submitted by Sandia National Laboratories. At Stanford, the team is made up of Profs. Alonso, Iaccarino, and Duraisamy, post-doctoral researcher Vinod Lakshminarayan, and graduate student Santiago Padron. At Sandia National Laboratories, the team includes Michael Eldred, Matt Barone, John Jakeman, and Stefan Domino, and at Purdue University, we have Prof. Dongbin Xiu as our main collaborator. The overall objective of this project was to develop a novel, comprehensive methodology for uncertainty quantification by combining stochastic expansions (nonintrusive polynomial chaos and stochastic collocation), the adjoint approach, and fusion with experimental data to account for aleatory and epistemic uncertainties from random variable, random field, and model form sources. The expected outcomes of this activity were detailed in the proposal and are repeated here to set the stage for the results that we have generated during the time period of execution of this project: 1. The rigorous determination of an error budget comprising numerical errors in physical space and statistical errors in stochastic space and its use for optimal allocation of resources; 2. A considerable increase in efficiency when performing uncertainty quantification with a large number of uncertain variables in complex non-linear multi-physics problems; 3. A solution to the long-time integration problem of spectral chaos approaches; 4. A rigorous methodology to account for aleatory and epistemic uncertainties, to emphasize the most important variables via dimension reduction and dimension-adaptive refinement, and to support fusion with experimental data using Bayesian inference; 5. The application of novel methodologies to time-dependent reliability studies in wind turbine applications including a number of efforts relating to the uncertainty quantification in vertical-axis wind turbine applications. In this report, we summarize all accomplishments in the project (during the time period specified) focusing on advances in UQ algorithms and deployment efforts to the wind turbine application area. Detailed publications in each of these areas have also been completed and are available from the respective conference proceedings and journals as detailed in a later section.

97 MATHEMATICS AND COMPUTING↗

Multi-Mechanism Flood Hazard Assessment: Critical Review of Current Practice and Approaches

This report documents the initial findings from the Nuclear Regulatory commission (NRC)-sponsored research project Methods for Estimating Joint Probabilities of Coincident and Correlated Flooding Mechanisms for Nuclear Power Plant Flood Hazard Assessments.1 This research project is a part of NRC’s Probabilistic Flood Hazard Assessment (PFHA) Research Program and will aid the development of guidance on the use of PFHA methods to evaluate infrastructure safety for existing and proposed US nuclear power plants (NPPs). More specifically, this project intends to provide technical background for the development of flood hazard curves for multi-mechanism floods (MMFs). MMFs are flood events caused by more than one flooding mechanism (e.g., flood events due to the simultaneous occurrence of precipitation-induced river flooding and storm surge). Project activities include three main tasks: Task 1—Survey of current concepts and methods in assessing MMF hazards; Task 2—Critical assessment of selected methods and approaches for quantifying probabilistic MMF hazard risk; Task 3—Development of example case studies to illustrate best practices for quantifying probabilistic MMF hazard risk The initial findings from Tasks 1 and 2 are documented in this report. Task 1 comprised a survey of approaches and methods that have been applied to understand and assess flood hazards due to MMFs. Task 2 involved a critical review of the selected approaches and methods. To that end, the scope of this report includes documentation of (1) a reconnaissance-level survey of the current state of concepts and practice for MMF hazard assessment; (2) a generalized MMF assessment framework to address the distinctions among various types of flood-forcing phenomena, flood mechanisms (grouped into three mechanism types), and flood severity metrics; (3) a wide-ranging survey of approaches and methods that have been applied to various flooding phenomena and settings; and (4) a critical assessment of MMF hazard assessment methods. Studies were identified involving MMFs related to coastal flooding mechanisms, fluvial (rivers/streams) flooding mechanisms, and associated combinations of coastal and fluvial flooding mechanisms. Studies were also identified that address MMFs involving coastal and fluvial flooding mechanisms as well as coastal flooding mechanisms along with extreme precipitation (without specific attribution to fluvial or pluvial mechanisms). The studies identified for review in this report included assessments at varying spatial scales (from local to global) with differing geographic regions of focus using both observed and synthetic data. The majority of studies identified and reviewed were site-specific assessments focusing on relatively short return periods. Studies considered a range of flood severity metrics, made differing assumptions regarding the occurrence of extrema, and used multiple statistical techniques; the use of copulas for the development of joint distributions was a particularly popular analysis technique. The literature review highlighted the differences among existing studies relative to terminology used, means of presenting results, framework and techniques employed, and level of sophistication regarding the number and types of variables considered. Despite the significant diversity in existing studies, the review identified several promising techniques that will be considered in future work under this project, including the development of joint distributions for MMFs using copula and Bayesian-motivated approaches.

98 NUCLEAR DISARMAMENT, SAFEGUARDS, AND PHYSICAL P↗

Lithium-Ion Battery Diagnostics Using Electrochemical Impedance via Machine-Learning

Diagnosing battery states such as health, state-of-charge, or temperature is crucial for ensuring the safety and reliability of electrochemical energy storage systems. While some states, such as temperature, may be measured using cheap sensors, accurate diagnosis of battery health metrics usually requires time-consuming performance measurements, making them infeasible for use in real-world operation. These health metrics can be measured during lab-testing and then estimated on-line using predictive life models or via state observer algorithms such as Kalman filters, but these predictive methods should be supplemented by actual measurement of battery health whenever possible to ensure reliability. Rapid measurement of battery health may be done by various types of fast diagnostic techniques such as electrochemical impedance spectroscopy (EIS), which can be performed in only a few minutes and require only a fraction of the energy and power needed for a full charge and discharge measurement. But there is a substantial challenge for estimating battery health using EIS data, as EIS is sensitive to cell temperature, state-of-charge, current, and resting time in addition to health. Thus, utilizing EIS data to predict battery capacity requires correcting for all these additional variables, a task that is extremely difficult to handle analytically. This talk utilizes machine-learning methods to estimate the effectiveness of battery capacity prediction from EIS data, leveraging a data set of hundreds of EIS measurements recorded at varying temperature and state-of-charge throughout a 500-day aging study of 32 commercial, large-format NMC-Graphite lithium-ion batteries. Using EIS as input to machine-learning models is complicated by the nonlinear response of impedance to battery health, temperature, and state-of-charge, as well as the collinearity between the impedance response at neighboring frequencies, which can easily lead to overfit models. To train robust models, features from EIS data need to be extracted from the data or some subset of critical frequencies selected. Many approaches for extracting and selecting features from EIS data from electrochemical analysis and machine-learning fields were identified for analysis: using the entire raw spectra; selection of one, two, or many frequencies from the entire spectra; selecting interesting points from the EIS measurement using domain knowledge; fitting EIS with an equivalent-circuit model; calculating statistics on the raw impedance values; and reducing the dimensionality of the data using unsupervised linear (principal component analysis) and non-linear (uniform manifold approximation and projection) methods. These approaches were rigorously compared using a machine-learning pipeline approach, training linear, Gaussian process, and random forest regression models and quantifying performance using cross-validation as well as a held-out test set. An artificial neural network model trained on the raw spectra was also tested. Promising pipelines were fine-tuned via Bayesian hyperparameter optimization using cross-validation loss and training with class-specific weights to counter data set imbalance. The most reliable method for utilizing impedance in this work was the selection of two optimal frequencies through an exhaustive search, resulting in about 2% mean absolute error on test data for both Gaussian process and random forest model architectures. Interrogation of a variety of models reveals critical frequencies of 100 Hz and 103 Hz for this data set, though the optimal set of frequencies is not necessarily intuitive, i.e., the best performing models are not simply those that use impedance at frequencies that have the highest correlation to the relative discharge capacity. The best performing model is an ensemble model, which is able to predict battery capacity with 1.9% mean absolute error for unseen cells using impedance recorded at a variety of temperatures and states-of-charge.

battery↗