SEARCH · Engineering Papers
Results for “Bayesian modeling”
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.
A Data-Driven Bayesian Crack Nucleation Model for Fatigue in Nickel-based Superalloys
Explore the source record for details and available documents.
Integrated Software Health Management for Aircraft GN and C
Modern aircraft rely heavily on dependable operation of many safety-critical software components. Despite careful design, verification and validation (V&V), on-board software can fail with disastrous consequences if it encounters problematic software/hardware interaction or must operate in an unexpected environment. We are using a Bayesian approach to monitor the software and its behavior during operation and provide up-to-date information about the health of the software and its components. The powerful reasoning mechanism provided by our model-based Bayesian approach makes reliable diagnosis of the root causes possible and minimizes the number of false alarms. Compilation of the Bayesian model into compact arithmetic circuits makes SWHM feasible even on platforms with limited CPU power. We show initial results of SWHM on a small simulator of an embedded aircraft software system, where software and sensor faults can be injected.
Bayesian Rules of Thumb: Robust Uncertainty Quantification in Early Project Cost Estimation
Systems engineers often make use of cost Rules ofThumb in order to estimate cost during early phases of projectformulation. These Rules of Thumb typically take the form ofa sequence of percentages over which a total cost is allocatedacross NASA WBS elements. Rules of Thumb can then be usedto extrapolate cost from one or more known WBS elements tothe remaining unknown WBS elements, assisting early projectformulation architecture studies (such as those in JPL’s Team Xand A Team).A number of issues can arise when generating and using costRules of Thumb. For example, many records of project costsconsist of incomplete data. Typical methods of dealing withincomplete cost allocation data include (a) ignoring missionswith incomplete data, or (b) taking averages of the non-zero percentagesacross missions, but both of these methods can result inbiased estimates if the existence of incomplete data correlateswith total mission cost or any particular WBS element. Anothercommon example is cost reported in one or more incorrect WBSelements. This is especially prevalent in smaller missions whereit is more common for engineers to perform tasks that fall underthe purview of multiple WBS elements.Furthermore, a Rule of Thumb estimate is typically reported asa point estimate; there is no reported uncertainty around thepercentages used to generate an allocation. Even in the rarecase in which confidence intervals around mean percentages areprovided, there may be positive or negative correlations betweenWBS elements which can skew estimates.Here we attempt to address these problems by formulatingprobabilistic Rules of Thumb in which a distribution of allocationschemes, rather than a single allocation scheme, is generated.We use a bootstrap imputation method to simultaneouslyaccount for uncertainty in the missing data while using allavailable information contained in the dataset. The imputeddatasets are then input into a multivariate Bayesian modelwhich accounts for correlations between WBS elements andproperly accounts for uncertainty in the final Rule of Thumbpercentages and predictions. We describe the mathematicalmodel and provides snippets of R code utilizing the brms(Bayesian Regression Models using Stan) package. To illustratethis model, we generate a Bayesian Level 2 WBS Cost Rule ofThumb for MIDEX (Medium-Class Explorers) missions withdata extracted from NASA’s CADRe. We then compare thismethod’s performance with the classical Rule of Thumb method.
Application of a Bayesian Framework for Plasticity Model Selection
Interpretable Machine Learning (IML) has performed well when tasked with deriving constitutive material models. However, IML has been shown to prefer models that overfit noise in data, which tends to lead to bloat and a decrease in interpretability. Due to these issues, the ability of IML to reliably derive models that fit the data and are both interpretable and generalizable is limited. A method developed recently has shown promise to improve upon traditional IML by using a Bayesian fitness definition for the evolution of free-form models with non-deterministic parameters. This framework was developed for genetic-programming-based symbolic regression(GPSR) and involves model parameter estimation using Sequential Monte Carlo sampling (SMC).The method has demonstrated a reduction in bloat when dealing with noisy data in comparison to conventional GPSR. The results of this framework applied to stress-strain data for copper show models that more effectively predict the experimental data better than was previously shown with GPSR.
A Bayesian analysis of two probability models describing thunderstorm activity at Cape Kennedy, Florida
A Bayesian analysis of the two discrete probability models, the negative binomial and the modified negative binomial distributions, which have been used to describe thunderstorm activity at Cape Kennedy, Florida, is presented. The Bayesian approach with beta prior distributions is compared to the classical approach which uses a moment method of estimation or a maximum-likelihood method. The accuracy and simplicity of the Bayesian method is demonstrated.
Interpolating Fields of Carbon Monoxide Data Using a Hybrid Statistical-Physical Model
Atmospheric Carbon Monoxide (CO) is a pollutant gas of which the US congress has mandated regular monitoring, and satellite sensors can be used to retrieve regional concentrations of CO over several vertical layers. However, CO at cloudy locations cannot be observed and have to be estimated from the observed data set, resulting in an interpolation problem. The current state-of-the-art solution is to combine prior information, computed by a deterministic physical model, with observations. However, the deterministic model may introduce uncertainties that do not derive from the data. While sharing certain features with the physical model, this paper presents a Bayesian hierarchical model for interpolating CO on a 3-dimensional spatial grid, across time. To our knowledge such a model has not been considered before. The model is applied to a hypothetical air-quality monitoring scenario, and is compared to existing interpolation methods. The results provide motivation for the use of the statistical model for regional to local applications.
The effect of modeling dose uncertainty on low-boom community noise dose-response curves
In logistic dose-response modeling, failing to account for uncertainty in estimated doses can cause an artificial flattening or attenuation of the slope of the summary curve. In Lee et al. [J. Acoust. Soc. Am. 147(4), pp. 2222-2234 (2020)], data from two NASA low-amplitude sonic boom community noise survey tests were modeled using a Bayesian multilevel logistic regression (MLR) statistical model that assumed there was no uncertainty in the noise dose estimates. However, in these community tests, the noise dose uncertainty was estimated by Page et al. [NASA/CR-2014-218180 and NASA/CR-2020-220589/Volume I] using a leave-one-out method. In the current work, a term was added to extend the Bayesian MLR model to account for the estimated noise dose uncertainty quantified in the Page et al. analyses. This uncertainty term was included in two ways, either as classical or as Berkson uncertainty, and yield similar results. When the uncertainty is accounted for in the Bayesian MLR model, the dose-response curves become 5-10% steeper, but the difference in the noise dose that elicits a 5% highly annoyed response is small (less than 1 dB). This result is encouraging for future X-59 community tests whose survey area will be sparsely populated with noise monitors.
Developing Deep Learning Models for System Remaining Useful Life Predictions: Application to Aircraft Engines
Prognostics and health management (PHM) is an important part of ensuring reliable operations of complex safety- critical systems. System-level remaining useful life (RUL) estimation is a much more complex problem than making estimations at the component level, and system-level RUL methodologies remain sparse in the literature. Model-based approaches have traditionally worked in the past for components such as capacitors, MOSFETs, batteries, or hard-drives (to name a few examples), but developing high fidelity dynamics models of cyber physical systems that can be used to study the effects of multiple degrading components in the system remains a challenging task. Some initial work on model-based System RUL predictions was demonstrated in Khorasgani, et al [1], but, to generalize the system-level prognostics problem, we have to resort to pure data driven and hybrid approaches. In this work, we propose an end-to-end data- driven framework for developing deep learning models to predict remaining useful life of cyber physical systems operating under unknown faulty conditions. The raw data is organized with a data schema that improves the model development process and down stream data analysis tasks. Due to the unknown faulty conditions, the raw sensor data is transformed into signals that expose the underlying degradation processes, which are then used for model development. Bayesian Optimization is used to tune the model parameters prior to training and validation. We show that this approach results in accurate predictions within 3 cycles to end of life (EOL). We demonstrate the effectiveness of our approach by applying it to the N-CMAPSS turbofan engine dataset recently released by NASA, which includes high fidelity degradation modeling, real world operating conditions, and a large set of fault operating modes.
Assimilating microwave cloudy observations into NASA GEOS model using a novel Bayesian Monte Carlo technique
Despite the importance of clouds and their influence on atmospheric water and energy balance, Numerical Weather Prediction (NWP) centers systematically exclude cloud information from the assimilation process and only assimilate clear-sky radiances (Janiskov´a et al. 2012). In order to ensure that only clear sky radiances are assimilated, strict cloud detection thresholds are applied before radiances are fed into data assimilation (DA) systems. This process not only excludes a large portion of satellite radiances, but causes loss of information in the regions that are of high interest to meteorologists and are most challenging for weather forecasts (Errico et al. 2007; Haddad et al. 2015). Although, in recent years there has been great advances in the operational weather forecasting, the prediction of tropical cyclones (TC), especially the intensity of TCs, remains challenging. According to Aksoy et al. (2013), in addition to the model deficiencies, another important factor that contributes to this challenge includes lack of observations in the peripheral environment (rain- bands) of TCs mainly because of the selective assimilation of existing observations. Satellite observations provide more than 90 % of the input data for the initialization of NWP models but more than 75 % of satellite observations are discarded due to the cloud contamination as well as land, snow, and ice emissivity issues (Bauer et al. 2010).
Territory and Population Attributes Affect Florida Scrub-Jay Fecundity in Fire Adapted Ecosystems
Fecundity, the number of young produced by a breeding pair during a breeding season, is a primary component in evolutionary and ecological theory and applications. Fecundity can be influenced by many environmental factors and requires long-term study due to the range of variation in ecosystem dynamics. Fecundity data often include a large proportion of zeros when many pairs fail to produce any young during a breeding season due to nest failure or when all young die independently after fledging. We conducted color banding and monthly censuses of Florida scrub-jays (Aphelocoma coerulescens) across 31 years, 15 populations, and 761 territories along central Florida’s Atlantic coast. We quantified how fecundity (juveniles/pair-year) was influenced by habitat quality, presence/absence of nonbreeders, population density, breeder experience, and rainfall, with a zero-inflated Bayesian hierarchical model including both a Bernoulli (e.g., brood success) and a Poisson (counts of young) submodel, and random effects for year, population, and territory. The results identified the importance of increasing “strong” quality habitat, which was a mid-successional state related to fire frequency and extent, because strong territories, and the proportion of strong territories in the overall population, influenced fecundity of breeding pairs. Populations subject to supplementary feeding also had greater fecundity. Territory size, population density, breeder experience, and rainfall surprisingly had no or small effects. Different mechanisms appeared to cause annual variation in fecundity, as estimates of random effects were not correlated between the success and count submodels. The increased fecundity for pairs with nonbreeders, compared to pairs without, identified empirical research needed to understand how the proportion of low-quality habitats influences population recovery and sustainability, because dispersal into low-quality habitats can drain nonbreeders from strong territories and decrease overall fecundity. We also describe how long term study resulted in reversals in our understanding because of complications involving habitat quality, sociobiology, and population density.
Assimilation of Microwave Cloudy Radiances into NASA GEOS Model Using a Novel Bayesian Monte Carlo Technique
No abstract available
Monte Carlo Bayesian Inference on a Statistical Model of Sub-gridcolumn Moisture Variability Using High-resolution Cloud Observations : Sensitivity Tests and Results - Part II
Part I presented a Monte Carlo Bayesian method for constraining a complex statistical model of GCM sub-gridcolumn moisture variability using high-resolution MODIS cloud data, thereby permitting large-scale model parameter estimation and cloud data assimilation. This part performs some basic testing of this new approach, verifying that it does indeed significantly reduce mean and standard deviation biases with respect to the assimilated MODIS cloud optical depth, brightness temperature and cloud top pressure, and that it also improves the simulated rotational-Ramman scattering cloud optical centroid pressure (OCP) against independent (non-assimilated) retrievals from the OMI instrument. Of particular interest, the Monte Carlo method does show skill in the especially difficult case where the background state is clear but cloudy observations exist. In traditional linearized data assimilation methods, a subsaturated background cannot produce clouds via any infinitesimal equilibrium perturbation, but the Monte Carlo approach allows finite jumps into regions of non-zero cloud probability. In the example provided, the method is able to restore marine stratocumulus near the Californian coast where the background state has a clear swath. This paper also examines a number of algorithmic and physical sensitivities of the new method and provides guidance for its cost-effective implementation. One obvious difficulty for the method, and other cloud data assimilation methods as well, is the lack of information content in the cloud observables on cloud vertical structure, beyond cloud top pressure and optical thickness, thus necessitating strong dependence on the background vertical moisture structure. It is found that a simple flow-dependent correlation modification due to Riishojgaard (1998) provides some help in this respect, by better honoring inversion structures in the background state.
Development of a Bayesian Belief Network Runway Incursion Model
In a previous paper, a statistical analysis of runway incursion (RI) events was conducted to ascertain their relevance to the top ten Technical Challenges (TC) of the National Aeronautics and Space Administration (NASA) Aviation Safety Program (AvSP). The study revealed connections to perhaps several of the AvSP top ten TC. That data also identified several primary causes and contributing factors for RI events that served as the basis for developing a system-level Bayesian Belief Network (BBN) model for RI events. The system-level BBN model will allow NASA to generically model the causes of RI events and to assess the effectiveness of technology products being developed under NASA funding. These products are intended to reduce the frequency of RI events in particular, and to improve runway safety in general. The development, structure and assessment of that BBN for RI events by a Subject Matter Expert panel are documented in this paper.
Operations for Learning with Graphical Models
This paper is a multidisciplinary review of empirical, statistical learning from a graphical model perspective. Well-known examples of graphical models include Bayesian net- works, directed graphs representing a Markov chain, and undirected networks representing a Markov field. These graphical models are extended to model data analysis and empirical learning using the notation of plates. Graphical operations for simplifying and manipulating a problem are provided including decomposition, differentiation, and the manipulation of probability models from the exponential family. These operations adapt existing techniques from statistics and automatic differentiation to graphs. Two standard algorithm schemes for learning are reviewed in a graphical framework: Gibbs sampling and the expectation maximization algorithm. Some algorithms are developed in this graphical framework including a generalized version of linear regression, techniques for feed-forward networks, and learning Gaussian and discrete Bayesian networks from data. The paper concludes by sketching some implications for data analysis and summarizing some popular algorithms that fall within the framework presented. The main original contributions here are the decomposition techniques and the demonstration that graphical models provide a framework for understanding and developing complex learning algorithms.
Trending in Probability of Collision Measurements via a Bayesian Zero-Inflated Beta Mixed Model
We investigate the performance of a generalized linear mixed model in predicting the Probabilities of Collision (Pc) for conjunction events. Specifically, we apply this model to the log(sub 10) transformation of these probabilities and argue that this transformation yields values that can be considered bounded in practice. Additionally, this bounded random variable, after scaling, is zero-inflated. Consequently, we model these values using the zero-inflated Beta distribution, and utilize the Bayesian paradigm and the mixed model framework to borrow information from past and current events. This provides a natural way to model the data and provides a basis for answering questions of interest, such as what is the likelihood of observing a probability of collision equal to the effective value of zero on a subsequent observation.
Multilevel Logistic Regression with Random Slope for Community Annoyance Survey Data
This paper documents recent dose-response modeling work at NASA in anticipation of follow-on work by a contactor. Specifically, this paper compares the results of a Bayesian MLR model with a fixed slope to one with a random slope using WSPR and QSF18 data. Previously reported dose-response modeling efforts of WSPR and QSF18 data have used a MLR model with a fixed slope term. A random slope may more accurately depict the dose-response relationship of individuals in the efforts to produce a population summary dose-response curve. Results of a fixed versus random slope model with WSPR and QSF18 data indicate minimal difference between the modeling methods. The simpler fixed slope model is preferable for these data, but these results do not preclude consideration of a random slope term in modeling efforts of future X-59 community test data.
Predicting Software Suitability Using a Bayesian Belief Network
The ability to reliably predict the end quality of software under development presents a significant advantage for a development team. It provides an opportunity to address high risk components earlier in the development life cycle, when their impact is minimized. This research proposes a model that captures the evolution of the quality of a software product, and provides reliable forecasts of the end quality of the software being developed in terms of product suitability. Development team skill, software process maturity, and software problem complexity are hypothesized as driving factors of software product quality. The cause-effect relationships between these factors and the elements of software suitability are modeled using Bayesian Belief Networks, a machine learning method. This research presents a Bayesian Network for software quality, and the techniques used to quantify the factors that influence and represent software quality. The developed model is found to be effective in predicting the end product quality of small-scale software development efforts.