Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Model calibration”

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 415 records · Page 23
All sourcesNASA NTRSDOE OSTI
NASA NTRS2001

Hyper-X Mach 10 Engine Flowpath Development: Fifth Entry Test Conditions and Methodology

A series of Hyper-X Mach 10 flowpath ground tests are underway to obtain engine performance and operation data and to confirm and refine the flowpath design methods. The model used is a full-scale height, partial-width replica of the Hyper-X Research Vehicle propulsive flowpath with truncated forebody and aftbody. This is the fifth test entry for this model in the NASA-HYPULSE facility at GASL. For this entry the facility nozzle and model forebody were modified to better simulate the engine inflow conditions at the target flight conditions. The forebody was modified to be a wide flat plate with no flow fences, the facility nozzle Mach number was increased, and the model was positioned to be tested in a semi-direct-connect arrangement. This paper presents a review of the test conditions, model calibrations, and a description of steady flow confirmation. The test series included runs using hydrogen fuel, and a silane-in-hydrogen fuel mixture. Other test parameters included the model mounting angle (relative to the tunnel flow), and the test gas oxygen fraction to account for the presence of [NO] in the test gas at the M10 conditions.

Bakos, R. J.↗
NASA NTRS2004

Modeling of High Speed Reacting Flows: Established Practices and Future Challenges

Computational fluid dynamics (CFD) has proven to be an invaluable tool for the design and analysis of high- speed propulsion devices. Massively parallel computing, together with the maturation of robust CFD codes, has made it possible to perform simulations of complete engine flowpaths. Steady-state Reynolds-Averaged Navier-Stokes simulations are now routinely used in the scramjet engine development cycle to determine optimal fuel injector arrangements, investigate trends noted during testing, and extract various measures of engine efficiency. Unfortunately, the turbulence and combustion models used in these codes have not changed significantly over the past decade. Hence, the CFD practitioner must often rely heavily on existing measurements (at similar flow conditions) to calibrate model coefficients on a case- by-case basis. This paper provides an overview of the modeled equations typically employed by commercial- quality CFD codes for high-speed combustion applications. Careful attention is given to the approximations employed for each of the unclosed terms in the averaged equation set. The salient features (and shortcomings) of common models used to close these terms are covered in detail, and several academic efforts aimed at addressing these shortcomings are discussed.

Baurle, R. A.↗
NASA NTRS1991

Development of a calibrated software reliability model for flight and supporting ground software for avionic systems

The object of this project was to develop and calibrate quantitative models for predicting the quality of software. Reliable flight and supporting ground software is a highly important factor in the successful operation of the space shuttle program. The models used in the present study consisted of SMERFS (Statistical Modeling and Estimation of Reliability Functions for Software). There are ten models in SMERFS. For a first run, the results obtained in modeling the cumulative number of failures versus execution time showed fairly good results for our data. Plots of cumulative software failures versus calendar weeks were made and the model results were compared with the historical data on the same graph. If the model agrees with actual historical behavior for a set of data then there is confidence in future predictions for this data. Considering the quality of the data, the models have given some significant results, even at this early stage. With better care in data collection, data analysis, recording of the fixing of failures and CPU execution times, the models should prove extremely helpful in making predictions regarding the future pattern of failures, including an estimate of the number of errors remaining in the software and the additional testing time required for the software quality to reach acceptable levels. It appears that there is no one 'best' model for all cases. It is for this reason that the aim of this project was to test several models. One of the recommendations resulting from this study is that great care must be taken in the collection of data. When using a model, the data should satisfy the model assumptions.

Lawrence, Stella↗
NASA NTRS2022

Development of a “Nature Run” for Observing System Simulation Experiments (OSSEs) for Snow Mission Development

Snow is a fundamental component of global and regional water budgets, particularly in mountainous areas and regions downstream that rely on snowmelt for water resources. Land surface models (LSMs) are commonly used to develop spatially distributed estimates of snow water equivalent (SWE) and runoff. However, LSMs are limited by uncertainties in model physics and parameters, among other factors. In this study, we describe the use of model calibration tools to improve snow simulations within the Noah-MP LSM as the first step in an observing system simulation experiment (OSSE). Noah-MP is calibrated against the University of Arizona (UA) SWE product over a western Colorado domain. With spatially varying calibrated parameters, we run calibrated and default Noah-MP simulations for water years 2010–20. By evaluating both simulations against the UA dataset, we show that calibration decreases domain averaged temporal RMSE and bias for snow depth from 0.15 to 0.13 m and from −0.036 to −0.0023 m, respectively, and improves the timing of snow ablation. Increased snow simulation performance also improves estimates of model-simulated runoff in four of six study basins, though only one has statistically significant improvement. Spatially distributed Noah-MP snow parameters perform better than default uniform values. We demonstrate that calibrating variables related to snow albedo calculations and rain–snow partitioning, among other processes, is a necessary step for creating a nature run that reasonably approximates true snow conditions for the OSSEs. Additionally, the inclusion of a snowfall scaling term can address biases in precipitation from meteorological forcing datasets, further improving the utility of LSMs for generating reliable spatiotemporal estimates of snow.

Melissa L. Wrzesien↗
DOE OSTI2022

Data-model files associated with the manuscript "The Effects of Spatial and Temporal Resolution of Gridded Meteorological Forcing on Watershed Hydrological Responses" (Shuai et al., 2022 HESS)

This data package contains the model inputs and outputs used in "The Effects of Spatial and Temporal Resolution of Gridded Meteorological Forcing on Watershed Hydrological Responses" (Shuai et al., 2022 HESS). The data.zip file contains the data used to drive the model simulations. The model.zip file contains the XML input file for ATS. The notebook.zip file contains the Jupyter notebooks for pre- and post- processing model results. The figures.zip file contains the raw figures associated with the manuscript.Meteorological forcing plays a critical role in accurately simulating the watershed hydrological cycle. With the advancement of high-performance computing and the development of integrated watershed models, simulating the watershed hydrological cycle at high temporal (hourly to daily) and spatial resolution (10s of meters) has become efficient and computationally affordable. These hyperresolution watershed models require high resolution of meteorological forcing as model input to ensure the fidelity and accuracy of simulated responses. In this study, we utilized the Advanced Terrestrial Simulator (ATS), an integrated watershed model, to simulate surface and subsurface flow and land surface processes using unstructured meshes at the Coal Creek Watershed near Crested Butte (Colorado). We compared simulated watershed hydrologic responses including streamflow, and distributed variables such as evapotranspiration, snow water equivalent (SWE), and groundwater table driven by three publicly available, gridded meteorological forcings (GMFs) -- Daily Surface Weather and Climatological Summaries (Daymet), Parameter-elevation Regressions on Independent Slopes Model (PRISM), and North American Land Data Assimilation System (NLDAS). By comparing various spatial resolutions (ranging from 400 m to 4 km) of PRISM, the simulated streamflow only becomes marginally worse when spatial resolution of meteorological forcing is coarsened to 4 km (or 30% of the watershed area). However, the 4 km resolution has much worse performance than finer resolution in spatially distributed variables such as SWE. Using temporally disaggregated PRISM, we compared models forced by different temporal resolutions (hourly to daily), sub-daily resolution preserves the dynamic watershed responses (e.g., diurnal fluctuation of streamflow) that are absent in results forced by daily resolution. Conversely, the simulated streamflow shows better performance using daily resolution compared to that using sub-daily resolution. Our findings suggest that the choice of GMF and its spatiotemporal resolution depends on the quantity of interest and its spatial and temporal scale, which may have important implications on model calibration and watershed management decisions.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Gravitational waves from global cosmic strings and cosmic archaeology

Global cosmic strings are predicted in many motivated extensions to the Standard Model of particle physics, with close connections to axion dark matter physics. Recent studies suggest that, although subdominant relative to Goldstone emission, gravitational wave (GW) signals from global strings can be detectable with current and planned GW detectors such as LIGO, LISA, DECIGO/BBO, ET/CE and AEDGE/AION, as well as pulsar timing arrays such as PPTA, NANOGrav and SKA. This work is an extensive, updated study on GWs from a global cosmic string network, taking into account of the most recent developments related to the subject. The main analysis is based on the analytical Velocity-dependent One-Scale (VOS) model calibrated with recent simulation results, which provides a generic protocol for such calculations with details given. We also demonstrate how the GW signal can be influenced with variations to the baseline model: this includes considering the uncertainties of model parameters and the potential deviation from the conventional VOS model prediction (i.e. the scaling behavior) as suggested by some of the recent simulation results. Furthermore, we investigated in detail the effect of a non-standard cosmology (e.g. early matter domination or kination) or new particle species on the GW signals from global strings. We demonstrate that the frequency spectrum of GW background from global cosmic strings can be used to probe the cosmic history prior to the Big Bang nucleosynthesis (BBN) (i.e. the primordial dark age) up to a temperature of T ~ 10 8 GeV.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2021

Assessment of a tomography-informed polyhedral discrete element modelling approach for complex-shaped granular woody biomass in stress consolidation

The design of handling equipment for granular biomass primarily requires experiments. Discrete element models (DEM) can provide designers with detailed insight into the behaviour of granular materials. However, granular biomass comprising complex-shaped particles is difficult to model with DEM. A tomography-informed DEM approach and an exhaustive assessment of the ability of the approach to predict the bulk behaviour of milled pines using experimental data is presented. Nano-CT scan was conducted to obtain 3D particle surface geometries as the basis for particle shape approximation by a polyhedral model and a sphero-polyhedral model. These models were applied in the simulation of a compressibility test. Our parametric study showed that particle Young's modulus and restitution coefficient are the two main properties influencing the simulated bulk behaviour of the DEM particles and that the level of approximating DEM particle shape for real particles is critical for accurately replicating the bulk behaviour of milled pines. The polyhedral model demonstrated better suitability than its sphero-polyhedral counterpart for modelling pine particles. The polyhedral model calibrated in a compressibility test was then applied in the simulations of a friction test without additional parameter tuning. Interlocking was dominant in the shear of bulk pine particles. Remarkably, the polyhedral model predicted the frictional behaviour of the pine particles when compared to the experimental results. The limitations of the model, as well as possible ways for enhancement, are discussed. This work provided novel insights into the suitability of complex-shaped DEM models for granular woody biomass.

Biomass flow↗
DOE OSTI2021

A critical look at the prediction of the temperature field around a laser-induced melt pool on metallic substrates

The study of microstructure evolution in additive manufacturing of metals would be aided by knowing the thermal history. Since temperature measurements beneath the surface are difficult, estimates are obtained from computational thermo-mechanical models calibrated against traces left in the sample revealed after etching, such as the trace of the melt pool boundary. Here we examine the question of how reliable thermal histories computed from a model that reproduces the melt pool trace are. To this end, we perform experiments in which one of two different laser beams moves with constant velocity and power over a substrate of 17-4PH SS or Ti-6Al-4V, with low enough power to avoid generating a keyhole. We find that thermal histories appear to be reliably computed provided that (a) the power density distribution of the laser beam over the substrate is well characterized, and (b) convective heat transport effects are accounted for. Poor control of the laser beam leads to potentially multiple three-dimensional melt pool shapes compatible with the melt pool trace, and therefore to multiple potential thermal histories. Ignoring convective effects leads to results that are inconsistent with experiments, even for the mild melt pools here.

42 ENGINEERING↗
DOE OSTI2024

Comparison of Results between the Legacy and Refined RELAP5-3D Models of the High Temperature Test Facility in Exercises 1 and 2 of the HTTF Benchmark

Work conducted in FY23 identified that RELAP5-3D was capable of reproducing trends in HTTF data during experiment PG-27 but was incapable of reproducing measured values. The primary cause of this discrepancy between RELAP5-3D results and experimental data was hypothesized to be a distortion in power density that was introduced by the radial nodalization of the model. We further hypothesized that a new model would provide better results when compared to the experiments PG-27 and PG-29. Work this FY developed a new model that is better capable of capturing local heat generation rates and contains a representation of each 1/6 azimuthal sector of the core. We used this model to develop a new set of solutions to Exercises 1 and 2 of Problems 2 and 3 in the benchmark. In this report, we present the first comprehensive comparison of the results between the two models. We see that in Exercise 1A, which is common between problems 2 and 3, the results are similar, though the results from the new model show greater detail than those from the legacy model. In Problem 2 Exercise 1B and Problem 3 Exercise 1B, we see that heat removal is slower in the new model than the legacy model. Problem 3 Exercise 1C shows temperatures that are lower in most places in the new model than the legacy model, but the area with active heat generation has higher block temperatures in the new model than the legacy model. Problem 3 Exercise 1D further shows that long-term heat removal is lower in the new model. Problem 2 Exercise 1C demonstrated that the new model observes higher temperatures in the core regions than the legacy model, justifying the need to preserve the power density in HTTF. The validation of PG-27 and PG-29 also demonstrated the improved temperature agreement in the core regions, particularly with a calibrated model that implements an effective thermal conductivity for the core material. Overall, PG-27 models show reasonable to excellent agreement for steady-state temperatures and minimal to reasonable agreement for transients. PG-29 models showed minimal to insufficient agreement with the data.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2025

Physics-informed machine learning for building performance simulation-A review of a nascent field

Building performance simulation (BPS) is critical for understanding building dynamics and behavior, analyzing the performance of the built environment, optimizing energy efficiency, improving demand flexibility, and enhancing building resilience. However, conducting BPS is not trivial. Traditional BPS relies on accurate building energy models, which are primarily physics-based and heavily dependent on detailed building information, expert knowledge, and case-by-case model calibrations, significantly limiting their scalability. With the development of sensing technology and the increased availability of data, there is growing attention and interest in data-driven BPS. However, purely data-driven models often suffer from limited generalization ability and a lack of physical consistency, resulting in poor performance in real-world applications. To address these limitations, recent studies have begun integrating physics priors into data-driven models, a methodology known as physics-informed machine learning (PIML). PIML is an emerging field where its definitions, methodologies, evaluation criteria, application scenarios, and future directions remain open. To bridge those gaps, this study systematically reviews the state-of-the-art PIML for BPS, offering a comprehensive definition of PIML and comparing it to traditional BPS approaches regarding data requirements, modeling effort, performance, and computational cost. We also summarize the commonly used methodologies, validation approaches, application domains, available data sources, open-source packages, and testbeds. In addition, this study provides a general guideline for selecting appropriate PIML models based on BPS applications. Finally, this study identifies key challenges and outlines future research directions, providing a solid foundation and valuable insights to advance R&D of PIML in BPS.

Jiang, Zixin↗
DOE OSTI2020

The impact of composition choices on solar evolution: age, helio- and asteroseismology, and neutrinos

ABSTRACT The Sun is the most studied and well-known star, and as such, solar fundamental parameters are often used to bridge gaps in the knowledge of other stars, when these are required for modelling. However, the two most powerful and precise independent methodologies currently available to infer the internal solar structure are in disagreement. We aim to show the potential impact of composition choices in the overall evolution of a star, using the Sun as example. To this effect, we create two Standard Solar Models and a comparison model using different combinations of metallicity and relative element abundances and compare evolutionary, helioseismic, and neutrino-related properties for each. We report differences in age for models calibrated to the same point on the HR diagram, in red giant branch, of more than 1 Gyr, and found that the current precision level of asteroseismic measurements is enough to differentiate these models, which would exhibit differences in period spacing of 1.30–2.58 per cent. Additionally, we show that the measurement of neutrino fluxes from the carbon–nitrogen–oxygen cycle with a precision of around 17 per cent, which could be achieved by the next generation of solar neutrino experiments, could help resolve the stellar abundance problem.

Capelo, Diogo↗
DOE OSTI2026

Quantification of trace iodine using laser-induced breakdown spectroscopy for real-time monitoring of nuclear off-gas streams

This study evaluated the potential of laser-induced breakdown spectroscopy (LIBS) for real-time monitoring of trace gas-phase iodine, which is an element of high significance in nuclear applications due to its long radioactive half-life (as iodine-129), volatility, and biological impact. In anticipation of iodine evolving into off-gas systems in molten salt reactor and nuclear fuel recycling applications, this research aimed to assess LIBS performance in flowing argon and helium matrices; optimize measurement parameters using a multichannel spectrometer; and perform calibrations to assess predictive capabilities and limits of detection (LODs). Experimental results successfully measured gas-phase iodine in flowing argon and helium; however, trace iodine was not detected in air. Optimal delay times were determined to be 10 µs for argon and 1 µs for helium, which are consistent with the expected shorter plasma lifetime in helium relative to argon. An emission line survey was provided with the 206.16, 804.37, 902.24, and 905.83 nm peaks, which were identified as the strongest emission peaks. Calibration models were successfully built in both helium and argon, achieving LODs down to 3 ppm in helium and 5 ppm in argon. The iodine emission at 905.83 nm emerged as the most robust for calibration and was subsequently applied to a time series dataset in argon. The predictive trace confirmed the feasibility of employing LIBS for continuous, online quantification of trace iodine in flowing gas systems.

Andrews, Hunter B. [Oak Ridge National Laboratory ↗
DOE OSTI2026

Pu(IV) quantification via visible–near-infrared absorption spectroscopy: tackling interferences using D-optimal design and partial least squares

Here, this study presents a novel analytical approach for quantifying Pu(IV) in glove box environments using fiber-optic-based visible–near-infrared absorption spectroscopy in combination with partial least squares regression (PLSR) and design of experiments. The method addresses significant challenges posed by overlapping spectral features arising from Nd(III), which is a common fission product impurity, and the speciation variability of Pu(IV) nitrato complexes in HNO 3 concentrations ranging from 2.5 to 11 M. A curated training set consisting of data from 20 samples was developed via D-optimal design to enable robust PLSR model calibration for Pu(IV) using the near-infrared band near 1050 nm. The training set was acquired from samples in cuvettes with a 1-cm path length and was used to build the PLSR model. The robustness of the model was validated with data collected using a dip probe with a 1-cm path length and varying Pu(IV) concentrations. The strong performance of the model indicates good model transfer from cuvette to dip probe and highlights the potential for in situ measurements and online monitoring of reactions in a crystallization reactor vessel. The results demonstrate that this combined spectroscopic and chemometric approach can accurately and simultaneously quantify Pu(IV) and HNO 3 , thereby offering a promising tool for real-time monitoring in process environments.

Actinide↗
DOE OSTI2025

Increasing Mosquito Abundance Under Global Warming

Mosquitoes are a key virus vector that poses significant health threats globally, affecting 700 million individuals and causing 1 million deaths annually. Accurately predicting mosquito abundance and dispersion remains a challenge. Complex interactions between mosquito dynamics and various environmental factors, notably hydrology, contribute to this challenge. Existing models typically focus on precipitation and temperature and often overlook further impacts of hydrological variables within mosquito modeling. In this study, we developed an artificial intelligence‐based model for mosquito dynamics, explicitly accounting for different hydrological variables, such as precipitation, soil moisture and streamflow. Using Toronto, Canada, as a case study, we identified causal relationships between changes in mosquito populations, hydrological factors, vegetation (e.g., leaf area index), and climate variables (e.g., daylight length, precipitation, and temperature). We embedded these relationships into a Long Short‐Term Memory (LSTM) Neural Network Model capable of accurately detecting mosquito dynamics across annual, seasonal, and monthly time scales. The LSTM is able to explain, on average, approximately 40% of the variance in the observed mosquito abundance data. Using the calibrated model, we predicted that the summer season mosquito abundance would increase by ∼16% and ∼19% under an intermediate greenhouse emission scenario, Shared Socioeconomic Pathway (SSP) 2–4.5, and a high greenhouse emission scenario, SSP5‐8.5, respectively. We expect that this model can serve as a valuable tool and inform science‐based decisions affecting mosquito dynamics and public health. It can also build a foundation for future risk analysis at the regional and larger scales.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS1994

A Robotic Lighting System for Solar Illumination Simulation

This report details our development fo a computer controlled spotlight with four actuated degrees-of-freedom for pan, tilt, linear movement and beam focusing. We review the mechanical structure, kinematic model, calibration data, and control algorithm.

space station freedom ssf solar illumination orbit↗
NASA NTRS2021

The connection between forest degradation and urban energy demand in sub-Saharan Africa: a characterization based on high-resolution remote sensing data

Charcoal is a key energy source for urban households in sub-Saharan Africa and charcoal production is the main cause of forest degradation across the region. We used multitemporal high-resolution remote sensing optical imagery to quantify the extent and intensity of forest degradation associated with charcoal production and its impact on forest carbon stocks for the main supplying area of an African capital. This analysis documents the advance of forest degradation and quantifies its aboveground biomass removals over a seven-year period, registering that, between 2013 and 2016, the average annual area under charcoal production was 103 sq.km and the annual aboveground biomass removals reached 1081 000 (SD = 2461) Mg. Kiln densities in the study area rose to 2 kilns-ha, with an average of 90.7 Mg ha^(−1) of extracted aboveground biomass. Charcoal production was responsible for the degradation of 55.5% of the mopane woodlands in the study area between 2013 and 2019. We estimated post-disturbance recovery times using an ecosystem model calibrated for the study area. The simulations showed that recovery times could require up to 150 years for current aboveground biomass extraction rates. The results of the remote sensing analysis and the simulations of the ecosystem model corroborate the unsustainability of the present patterns of charcoal production. The detailed characterization of the spatial and temporal patterns of charcoal production was combined with household survey information to quantify the impact of the urban energy demand of the Maputo urban area on forest carbon stocks. The analysis shows that Maputo charcoal demand was responsible for the annual degradation of up to 175.3 sq.km and that the contribution of the study area to this demand fluctuated between 75% and 33% over the study period. The extent, advance pace and distance from urban centers documented in this study support the idea that forest degradation from charcoal production cannot merely be considered a peri-urban process. The intensity of the aboveground biomass (AGB) removals and its contribution to forest carbon stocks changes is significant at the national and regional levels.

F Sedano↗
DOE OSTI2022

Comparing and contrasting In-Vial and full-scale systems for sparging volatile analytes

Here, in-vial sparging was demonstrated as an effective, practical alternative to a full-scale sparging system for supporting the analysis of volatile constituents. Using elemental mercury (Hg 0 ) and toluene as representative purgeable analytes, the mass removal for various sparge configurations was measured and a reduced order model was developed and validated. In the primary experiments, Hg 0 in the sparge gas was trapped on activated carbon or gold, thermally desorbed, and quantified using atomic absorption or atomic fluorescence spectroscopy. Toluene experiments using the same in-vial sparge apparatus and sparge parameters were performed to demonstrate the applicability of the reduced order model to a broad range of compounds. Toluene removal was tracked by measuring the remaining toluene in sparged aliquots using Ultraviolet-visible (UV–Vis) spectroscopy. For the sparging, flow rates varied from 25 to 75 mL/min for periods from 0 to 30 min. Sparge performance, mass removal as a function of time, and sparge gas volume were measured for both in-vial and full-scale systems. A model based on dimensionless Henry's Law coefficient, normalized sparge gas volume, and fractional extent of equilibrium matched the experimental data for both compounds and provides a practical tool for future applications. For the conditions tested in this study, the calibrated model indicated that the sparge gas in the in-vial system reached approximately 33% of its equilibrium value before exiting the water surface, while a full-scale system reached approximately 100%. The tests validated the quality, reproducibility, and predictability of sparging performance for both full scale and in-vial sparge systems. Related factors such as waste generation, worker risk, and labor were also assessed. Full scale sparge systems provide the advantage of lower detection levels due to larger sample volume, while the in-vial sparge systems provide advantages for most other factors; including automatability, reducing secondary wastes, lessening the need to clean and check the sparge apparatus, and lowering labor and costs. The data and associated reduced order model support continued development and deployment of in-vial sparge platforms as a practical option for analysis of purgeable analytes such as volatile organic compounds and volatile metals/organometallics.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

Robust scalable initialization for Bayesian variational inference with multi-modal Laplace approximations

Predictive modeling typically relies on Bayesian model calibration to provide uncertainty quantification. Variational inference utilizing fully independent (“mean-field”) Gaussian distributions are often used as approximate probability density functions. This simplification is attractive since the number of variational parameters grows only linearly with the number of unknown model parameters. However, the resulting diagonal covariance structure and unimodal behavior can be too restrictive to provide useful approximations of intractable Bayesian posteriors that exhibit highly non-Gaussian behavior, including multimodality. High-fidelity surrogate posteriors for these problems can be obtained by considering the family of Gaussian mixtures. Gaussian mixtures are capable of capturing multiple modes and approximating any distribution to an arbitrary degree of accuracy, while maintaining some analytical tractability. Unfortunately, variational inference using Gaussian mixtures with full-covariance structures suffers from a quadratic growth in variational parameters with the number of model parameters. The existence of multiple local minima due to strong nonconvex trends in the loss functions often associated with variational inference present additional complications, These challenges motivate the need for robust initialization procedures to improve the performance and computational scalability of variational inference with mixture models. In this work, we propose a method for constructing an initial Gaussian mixture model approximation that can be used to warm-start the iterative solvers for variational inference. The procedure begins with a global optimization stage in model parameter space. In this step, local gradient-based optimization, globalized through multistart, is used to determine a set of local maxima, which we take to approximate the mixture component centers. Around each mode, a local Gaussian approximation is constructed via the Laplace approximation. Finally, the mixture weights are determined through constrained least squares regression. The robustness and scalability of the proposed methodology is demonstrated through application to an ensemble of synthetic tests using high-dimensional, multimodal probability density functions. Here, the practical aspects of the approach are demonstrated with inversion problems in structural dynamics.

97 MATHEMATICS AND COMPUTING↗