Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Inverse Optimization”

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
NASA NTRSDate not supplied

Adding GPU Support to the Markov Chain Monte Carlo Code Catmip

In geophysics, we are confronted with many under-determined inverse problems. For example, all of our observations of earthquakes are made at the Earth’s surface. So, when we try to infer how slip during an earthquake evolves in space and time, we find that there are many potential slip histories that are consistent with our limited observations and our understanding of earthquake physics. One way to approach these problems is with Bayesian analysis which allows us to infer the ensemble of all potential slip models that satisfy the observations and our prior knowledge of earthquake physics. In Bayesian analysis, our prior knowledge is known as the prior probability density function or prior PDF, the fit to the data is known as the data likelihood, and the target PDF that satisfies both the prior PDF and data likelihood is known as the posterior PDF. However, simulating the posterior PDF typically requires using Markov Chain Monte Carlo (MCMC) to draw tens of billions of random realizations of earthquake slip models, which may not be computationally feasible. To make this and similar geophysical inversions computationally tractable, we developed the Cascading Adaptive Transitional Metropolis In Parallel (CATMIP) algorithm. CATMIP is an efficient parallel Markov Chain Monte Carlo (MCMC) sampler that is used for model fitting and uncertainty quantification in geophysics. Example use cases are earthquake rupture modeling, determining mineral composition on Mars, reconstructing the history of ocean salinity, and historical earthquake relocation. CATMIP employs many parallel instances of the Metropolis algorithm for sampling in a transitioning framework. Transitioning is a process in which a set of random samples at equilibrium with a known probability density function (PDF) are used as seeds for the Markov chains to sample successive target PDFs that incrementally move the distribution from the starting seeds to the final desired PDF that describes the relative plausibility of potential values for the model parameters. The algorithm is implemented as a Master-Worker model employing MPI for communication. The worker processes are loosely coupled with global parameters periodically optimized by the master process. This provides a very high amount of parallelism with little communication between updates. During the presentation we will discuss the history of the algorithm and elaborate the earthquake rupture modeling use case for the CATMIP package. Our first step toward GPU optimization was to optimize the code for the CPU. CPU profiling revealed that most of the compute time is spent in calls to level 2 BLAS routines and calls to GSL random number generators. We revised the algorithm to employ level 3 BLAS routines instead. In our presentation we will describe how this was accomplished. Adding GPU support to CATMIP consisted mostly of replacing the calls to GSL with calls to GPU vendor-provided library routines. A small number of loops were directly implemented in CUDA. In the presentation will provide implementation details. Finally, we will discuss methods for profiling and opportunities for further optimizing GPU execution. By creating a code with the flexibility to run on either a CPU or GPU architecture, CATMIP can be used on systems ranging from large CPU-based HPC environments to single servers with GPU acceleration and everything in between.

HECC↗
DOE OSTI2023

Multi-modal characterization methods of solid-electrolyte interphase in silicon-graphite composite electrodes

Composite silicon-graphite (Si-Gr) anodes can improve battery energy density, due to Si's high gravimetric capacity, while mitigating mechanical degradation of the anode and solid-electrolyte interphase (SEI) caused by Si volumetric expansion. Optimizing these anodes is challenging, in part due to difficulty characterizing the SEI structure and composition. In this work, we present multi-modal characterization of the SEI on composite Si-Gr anodes to relate SEI chemical composition and structure to functional properties. Discrepancies in elemental concentrations from X-ray photoelectron spectroscopy, Auger electron spectroscopy, and energy-dispersive X-ray spectroscopy (EDS) are attributed to varying information depth and lateral resolution of the individual probes. However, by combining quantitative composition information with spatially resolved element mapping from scanning transmission electron microscopy, EDS, and electron energy loss spectroscopy, a holistic picture of the SEI emerges. We observe the bilayer SEI structure and a direct correlation between elemental Li and F, suggesting that most Li in the SEI exists as lithium fluoride (LiF). Further, LiF concentration is directly proportional to the maximum SEI resistivity, as determined by scanning spreading resistance microscopy. Lastly, there is an inverse relationship between lithium carbonate and LiF concentration in the SEI, providing insight into the detailed chemistry of SEI formation and evolution.

25 ENERGY STORAGE↗
DOE OSTI2024

First experimental study of multiple orientation muon tomography, with image optimization in sparse data environments

Due to the high penetrating power of cosmic ray muons, they can be used to probe very thick and dense objects. As charged particles, they can be tracked by ionization detectors, determining the position and direction of the muons. With detectors on either side of an object, particle direction changes can be used to extract scattering information within an object. This can be used to produce a scattering intensity image within the object related to density and atomic number. Such imaging is typically performed with a single detector-object orientation, taking advantage of the more intense downward flux of muons, producing planar imaging with some depth-of-field information in the third dimension. Several simulation studies have been published with multi-orientation tomography, which can form a three-dimensional representation faster than a single orientation view. In this work we present the first experimental multiple orientation muon tomography study. Experimental muon-scatter based tomography was performed using a concrete filled steel drum with several different metal wedges inside, between detector planes. Data was collected from different detector-object orientations by rotating the steel drum. The data collected from each orientation were then combined using two different tomographic methods. Results showed that using a combination of multiple depth-of-field reconstructions, rather than a traditional inverse Radon transform approach used for CT, resulted in more useful images for sparser data. As cosmic ray muon flux imaging is rate limited, the imaging techniques were compared for sparse data. Using the combined depth-of-field reconstruction technique, fewer detector-object orientations were needed to reconstruct images that could be used to differentiate the metal wedge compositions.

Applied Physics (physics.app-ph)↗
DOE OSTI2023

Bayesian sequential optimal experimental design for nonlinear models using policy gradient reinforcement learning

We present a mathematical framework and computational methods for optimally designing a finite sequence of experiments. This sequential optimal experimental design (sOED) problem is formulated as a finite-horizon partially observable Markov decision process (POMDP) under a Bayesian setting and with information-theoretic utilities. The formulation is general and may accommodate continuous random variables, non-Gaussian posteriors, and nonlinear forward models. The sOED design policy incorporates elements of feedback and lookahead simultaneously, and we show it to generalize the commonly-used batch and greedy design strategies. We solve for the sOED policy using the policy gradient (PG) method from reinforcement learning, and provide a derivation for the PG expression in the sOED context. Adopting an actor-critic approach, the policy and value functions are parameterized using deep neural networks and improved via PG estimates produced from simulated episodes of designs and observations. The new PG-sOED algorithm is first validated on a linear-Gaussian benchmark, and then compared against other design baselines on a sensor movement problem for contaminant source inversion in a convection-diffusion field. As a result, we provide explanation for the policy behaviors using knowledge of the underlying physical process.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Securely Aggregated Coded Matrix Inversion

Coded computing is a method for mitigating straggling workers in a centralized computing network, by using erasure-coding techniques. Federated learning is a decentralized model for training data distributed across client devices. In this work we propose approximating the inverse of an aggregated data matrix, where the data is generated by clients; similar to the federated learning paradigm, while also being resilient to stragglers. To do so, we propose a coded computing method based on gradient coding. We modify this method so that the coordinator does not access the local data at any point; while the clients access the aggregated matrix in order to complete their tasks. Here, the network we consider is not centrally administrated, and the communications which take place are secure against potential eavesdroppers.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2014

Evaluation of the Lactate-to-N-Acetyl-aspartate Ratio Defined With Magnetic Resonance Spectroscopic Imaging Before Radiation Therapy as a New Predictive Marker of the Site of Relapse in Patients With Glioblastoma Multiforme

Purpose: Because lactate accumulation is considered a surrogate for hypoxia and tumor radiation resistance, we studied the spatial distribution of the lactate-to-N-acetyl-aspartate ratio (LNR) before radiation therapy (RT) with 3D proton magnetic resonance spectroscopic imaging (3D-{sup 1}H-MRSI) and assessed its impact on local tumor control in glioblastoma (GBM). Methods and Materials: Fourteen patients with newly diagnosed GBM included in a phase 2 chemoradiation therapy trial constituted our database. Magnetic resonance imaging (MRI) and MRSI data before RT were evaluated and correlated to MRI data at relapse. The optimal threshold for tumor-associated LNR was determined with receiver-operating-characteristic (ROC) curve analysis of the pre-RT LNR values and MRI characteristics of the tumor. This threshold was used to segment pre-RT normalized LNR maps. Two spatial analyses were performed: (1) a pre-RT volumetric comparison of abnormal LNR areas with regions of MRI-defined lesions and a choline (Cho)-to- N-acetyl-aspartate (NAA) ratio ≥2 (CNR2); and (2) a voxel-by-voxel spatial analysis of 4,186,185 voxels with the intention of evaluating whether pre-RT abnormal LNR areas were predictive of the site of local recurrence. Results: A LNR of ≥0.4 (LNR-0.4) discriminated between tumor-associated and normal LNR values with 88.8% sensitivity and 97.6% specificity. LNR-0.4 voxels were spatially different from those of MRI-defined lesions, representing 44% of contrast enhancement, 64% of central necrosis, and 26% of fluid-attenuated inversion recovery (FLAIR) abnormality volumes before RT. They extended beyond the overlap with CNR2 for most patients (median: 20 cm{sup 3}; range: 6-49 cm{sup 3}). LNR-0.4 voxels were significantly predictive of local recurrence, regarded as contrast enhancement at relapse: 71% of voxels with a LNR-0.4 before RT were contrast enhanced at relapse versus 10% of voxels with a normal LNR (P<.01). Conclusions: Pre-RT LNR-0.4 in GBM indicates tumor areas that are likely to relapse. Further investigations are needed to confirm lactate imaging as a tool to define additional biological target volumes for dose painting.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2025

Rhode Island Site - NREL ASSIST Thermodynamic Retrievals TROPoe v0.19 / Derived Data

This dataset contains daily files with thermodynamic profiles retrieved with the optimal estimation physical retrieval TROPoe v0.19 (Turner and Löhnert, 2014; Turner and Blumberg, 2019; Turner and Löhnert, 2021). The profiles are retrieved every 10 minutes from instantaneous observations from the NREL ASSIST-II (SN 12) infrared spectrometer. Observations are noise-filtered but not averaged in time to minimize errors due to non-uniform clouds. Additional input data in TROPoe are cloud base height from the collocated NREL upgraded Galion lidar (Newsom et al., 2019), and surface temperature, relative humidity, and pressure from the collocated PNNL met tower. The full pipeline to run the retrieval is available at https://github.com/StefanoWind/TROPoe_processor. In addition to these temporally resolved input data, TROPoe requires an a priori dataset (prior) that provides mean climatological estimates of thermodynamic profiles and specifies how temperature and humidity covary with height as an input (for details see, e.g., Djalalova et al., 2022). The prior is a key component of the retrieval and provides a constraint on the ill-posed inversion problem. A monthly prior was computed from operational radiosonde launches at Upton, NY.

17 WIND ENERGY↗
NASA NTRS2010

Fast Solution in Sparse LDA for Binary Classification

An algorithm that performs sparse linear discriminant analysis (Sparse-LDA) finds near-optimal solutions in far less time than the prior art when specialized to binary classification (of 2 classes). Sparse-LDA is a type of feature- or variable- selection problem with numerous applications in statistics, machine learning, computer vision, computational finance, operations research, and bio-informatics. Because of its combinatorial nature, feature- or variable-selection problems are NP-hard or computationally intractable in cases involving more than 30 variables or features. Therefore, one typically seeks approximate solutions by means of greedy search algorithms. The prior Sparse-LDA algorithm was a greedy algorithm that considered the best variable or feature to add/ delete to/ from its subsets in order to maximally discriminate between multiple classes of data. The present algorithm is designed for the special but prevalent case of 2-class or binary classification (e.g. 1 vs. 0, functioning vs. malfunctioning, or change versus no change). The present algorithm provides near-optimal solutions on large real-world datasets having hundreds or even thousands of variables or features (e.g. selecting the fewest wavelength bands in a hyperspectral sensor to do terrain classification) and does so in typical computation times of minutes as compared to days or weeks as taken by the prior art. Sparse LDA requires solving generalized eigenvalue problems for a large number of variable subsets (represented by the submatrices of the input within-class and between-class covariance matrices). In the general (fullrank) case, the amount of computation scales at least cubically with the number of variables and thus the size of the problems that can be solved is limited accordingly. However, in binary classification, the principal eigenvalues can be found using a special analytic formula, without resorting to costly iterative techniques. The present algorithm exploits this analytic form along with the inherent sequential nature of greedy search itself. Together this enables the use of highly-efficient partitioned-matrix-inverse techniques that result in large speedups of computation in both the forward-selection and backward-elimination stages of greedy algorithms in general.

Moghaddam, Baback↗
NASA NTRS2012

Assessment of Mars Atmospheric Temperature Retrievals from the Thermal Emission Spectrometer Radiances

Motivated by the needs of Mars data assimilation. particularly quantification of measurement errors and generation of averaging kernels. we have evaluated atmospheric temperature retrievals from Mars Global Surveyor (MGS) Thermal Emission Spectrometer (TES) radiances. Multiple sets of retrievals have been considered in this study; (1) retrievals available from the Planetary Data System (PDS), (2) retrievals based on variants of the retrieval algorithm used to generate the PDS retrievals, and (3) retrievals produced using the Mars 1-Dimensional Retrieval (M1R) algorithm based on the Optimal Spectral Sampling (OSS ) forward model. The retrieved temperature profiles are compared to the MGS Radio Science (RS) temperature profiles. For the samples tested, the M1R temperature profiles can be made to agree within 2 K with the RS temperature profiles, but only after tuning the prior and error statistics. Use of a global prior that does not take into account the seasonal dependence leads errors of up 6 K. In polar samples. errors relative to the RS temperature profiles are even larger. In these samples, the PDS temperature profiles also exhibit a poor fit with RS temperatures. This fit is worse than reported in previous studies, indicating that the lack of fit is due to a bias correction to TES radiances implemented after 2004. To explain the differences between the PDS and Ml R temperatures, the algorithms are compared directly, with the OSS forward model inserted into the PDS algorithm. Factors such as the filtering parameter, the use of linear versus nonlinear constrained inversion, and the choice of the forward model, are found to contribute heavily to the differences in the temperature profiles retrieved in the polar regions, resulting in uncertainties of up to 6 K. Even outside the poles, changes in the a priori statistics result in different profile shapes which all fit the radiances within the specified error. The importance of the a priori statistics prevents reliable global retrievals based a single a priori and strongly implies that a robust science analysis must instead rely on retrievals employing localized a priori information, for example from an ensemble based data assimilation system such as the Local Ensemble Transform Kalman Filter (LETKF).

Hoffman, Matthew J.↗
DOE OSTI2018

MULTI-PRED: A Software Module for Predictive Modeling of Coupled Multi-Physics Systems: User's Manual

This User’s Manual describes the code module MULTI-PRED, written in FORTRAN which implements the methodology for “predictive modeling of coupled multi-physics systems (PM-CMPS)” formulated by Cacuci (2014). This methodology fully takes into account the coupling terms between the systems but requires only the computational resources that would be needed to perform predictive modeling on each system separately. The PM-CMPS methodology uses the maximum entropy principle to construct an optimal approximation of the unknown a priori distribution based on a priori known mean values and uncertainties characterizing the experimental and computational parameters and results of interest responses, called for the multi-physics models under consideration. This “maximum entropy” a priori distribution is combined, using Bayes’ theorem, with the “likelihood” provided by the multi-physics simulation models to obtain a formal posterior distribution. Subsequently, the posterior distribution thus obtained is evaluated using the saddle-point method to obtain analytical expressions for the optimally predicted values for the multi-physics models parameters and responses along with corresponding reduced uncertainties. Noteworthy, the predictive modeling methodology for the coupled systems is constructed such that the systems can be considered sequentially rather than simultaneously, while preserving exactly the same results as if the systems were treated simultaneously. Consequently, very large coupled systems, which could perhaps exceed available computational resources if treated simultaneously, can be treated with the PM-CMPS methodology presented in this work sequentially and without any loss of generality or information, requiring just the resources that would be needed if the systems were treated sequentially. Three illustrative demonstration problems are also provided. The first problem presents the application of the PM-CMPS methodology to a simple particle diffusion problem which admits a closed-form analytical solution which facilitates a rapid understanding of this methodology and its predicted results. The second demonstration problem presents the application of the PM-CMPS methodology to the problem of inverse prediction, from detector responses in the presence of counting uncertainties, of the thickness of a homogeneous slab of material containing uniformly distributed gamma-emitting sources, for optically thin and thick slabs. This problem highlights the essential role played by the relative uncertainties (or, conversely, accuracies) of measured and computed responses. The third demonstration problem presents the application of the PM-CMPS methodology to the F-area cooling towers at the Savannah River National Lab. This problem demonstrates that the PM-CMPS methodology reduces the predicted response uncertainties not only at locations where measurements are available, but also at locations where measurements are not available.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1993

Modeling Spectralon's Bidirectional Reflectance for In-Flight Calibration of Earth-Orbiting Sensors

This paper describes the results of a model inversion which was conducted using measured Spectralon BRDF data at several illumination angles. We can describe the bidirectional reflectance of the Spectralon and in paticular its slight anisotrophy with a physically-based model. The root mean square difference between the model and the observations is currently of the order of the noise in the data (+/-1%). It is expected that meaningful corrections to account for the anisotrophy of the Spectralon panels will be achieved in the next few years and will improve the calibration of the MISR instrument.

EOS↗
DOE OSTI2021

A machine-learning inverse model framework for rapid forecasting and history matching in unconventional reservoirs

Model-based optimization for real-time forecasting in unconventional reser-voirs requires novel methods and work?ows since the strategies and work?ows used in conventional reservoirs are either inapplicable, or prohibitively expen-sive and time-consuming. Insu?cient site data and computational expense of high-?delity simulations mean that work?ows with high-?delity simulations are not ideal for usage in comprehensive uncertainty quanti?cation stud-ies that require 1000s of forward model runs. We present an alternative, novel work?ow for unconventional reservoirs, based on the interplay between reduced-order models and machine-learning. Our physics-informed machine-learning (PIML) work?ow addresses the challenges to real-time reservoir management in uncoventionals, namely lack of data (the time-frame for which the wells have been producing), and computational expense of high-?delity modeling. We use the machine-learning paradigm of transfer-learning to bind together fast but less accurate reduced-order models with slow, but accurate high-?delity models and circumvent the di?culties inherent in the current state-of-the-art for unconventionals. Such a PIML work?ow, grounded in physics, is a viable candidate for real-time history matching and production forecasting in a fractured shale gas reservoir. The signi?cance of our approach is that while it is developed for a particu-lar well and site in the Marcelus Shale gas reservoir of the Appalachian basin (MSEEL), it is not wedded to it. We expect the same work?ow can be ap-plied to other shale formations (e.g., Woodford, Barnett, Utica, EagleFord) should site-data become available, using the same set of machine-learning techniques from transfer learning. Some ?ne-tuning (or minimal retraining of the neural networks) will be required to transfer knowledge across shale gas sites/formations but it is a clearly superior alternative to developing a new machine-learning model altogether when considering a di?erent site.

Srinivasan, Shriram↗
DOE OSTI2013

Who Benefits From Adjuvant Radiation Therapy for Gastric Cancer? A Meta-Analysis

Purpose: Large randomized trials have demonstrated significant survival benefits with the use of adjuvant chemotherapy or chemoradiation therapy for gastric cancer. The importance of adjuvant radiation therapy (RT) remains unclear. We performed an up-to-date meta-analysis of randomized trials testing the use of RT for resectable gastric cancer. Methods and Materials: We searched MEDLINE, EMBASE, and the Cochrane Central Register of Controlled Trials for randomized trials testing adjuvant (including neoadjuvant) RT for resectable gastric cancer. Hazard ratios describing the impact of adjuvant RT on overall survival (OS) and disease-free survival (DFS) were extracted directly from the original studies or calculated from survival curves. Pooled estimates were obtained using the inverse variance method. Subgroup analyses were performed to determine whether the efficacy of RT varies with chemotherapy use, RT timing, geographic region, type of nodal dissection performed, or lymph node status. Results: Thirteen studies met all inclusion criteria and were used for this analysis. Adjuvant RT was associated with a significant improvement in both OS (HR = 0.78, 95% CI: 0.70-0.86, P<.001) and DFS (HR = 0.71, 95% CI: 0.63-0.80, P<.001). In the 5 studies that tested adjuvant chemoradiation therapy against adjuvant chemotherapy, similar effects were seen for OS (HR = 0.83, 95% CI: 0.67-1.03, P=.087) and DFS (HR = 0.77, 95% CI: 0.91-0.65, P=.002). Available data did not reveal any subgroup of patients that does not benefit from adjuvant RT. Conclusion: In randomized trials for resectable gastric cancer, adjuvant RT provides an approximately 20% improvement in both DFS and OS. Available data do not reveal a subgroup of patients that does not benefit from adjuvant RT. Further study is required to optimize the implementation of adjuvant RT for gastric cancer with regard to patient selection and integration with systemic therapy.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2015

Advancing Inverse Sensitivity/Uncertainty Methods for Nuclear Fuel Cycle Applications

The inverse sensitivity/uncertainty quantification (IS/UQ) method has recently been implemented in the Inverse Sensitivity/UnceRtainty Estimator (INSURE) module of the AMPX cross section processing system [M.E. Dunn and N.M. Greene, “AMPX-2000: A Cross-Section Processing System for Generating Nuclear Data for Criticality Safety Applications,” Trans. Am. Nucl. Soc. 86, 118–119 (2002)]. The IS/UQ method aims to quantify and prioritize the cross section measurements along with uncertainties needed to yield a given nuclear application(s) target response uncertainty, and doing this at a minimum cost. Since in some cases the extant uncertainties of the differential cross section data are already near the limits of the present-day state-of-the-art measurements, requiring significantly smaller uncertainties may be unrealistic. Therefore, we have incorporated integral benchmark experiments (IBEs) data into the IS/UQ method using the generalized linear least-squares method, and have implemented it in the INSURE module. We show how the IS/UQ method could be applied to systematic and statistical uncertainties in a self-consistent way and how it could be used to optimize uncertainties of IBEs and differential cross section data simultaneously. We itemize contributions to the cost of differential data measurements needed to define a realistic cost function.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2017

Kriging-Based Inverse Uncertainty Quantification of BISON Fission Gas Release Model

Nuclear reactor fuel performance analysis studies the thermo-mechanical behavior of fuel rods and verify their compliance with safety criteria under both normal operation and accidental conditions. One of the major concerns in fuel performance analysis is the behavior of the fission gases xenon and krypton in uranium dioxide fuel, which significantly affect the thermo-mechanical performance of the nuclear fuel rods employed in current LWRs. Previously, the modeling of fission gas release (FGR) in nuclear fuel performance codes have been relying on empirical models which are not applicable beyond their range of calibration. Recently a more efficient and flexible physics-based FGR and swelling model that can describe a wider range of reactor operation conditions have been implemented in fuel performance code BISON. In this work, we performed Bayesian calibration to inversely quantify the uncertainties associated with the 5 tuning parameters in BISON FGR model, based on the time-dependent FGR measurement data from Riso-AN3 benchmark. It takes around 1 hour and 40 minutes for BISON to simulate the Riso-AN3 benchmark with a moderate mesh size using 32 processors. As hundreds of thousands of BISON simulations are needed during Markov Chain Monte Carlo (MCMC) sampling, running BISON directly is not a viable solution. Therefore we propose to use Kriging-based metamodels to replace BISON during MCMC sampling. Surrogate model is an approximation of the input/output relation of a computer code/model. It is also called metamodel, response surface, emulator. Surrogate models usually take much less computational time than the original model (like BISON) while maintaining the input/output relation of the original model to a desirable accuracy. Constructing the metamodels normally requires small number of executions of the original model. Once validated, metamodels can be used to perform uncertainty and sensitivity analysis, Bayesian calibration, optimization, etc.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
NASA NTRS1991

On designing for quality

The problem of ensuring the required quality of products and/or technological processes often becomes more difficult due to the fact that there is not general theory of determining the optimal sets of value of the primary factors, i.e., of the output parameters of the parts and units comprising an object and ensuring the correspondence of the object's parameters to the quality requirements. This is the main reason for the amount of time taken to finish complex vital article. To create this theory, one has to overcome a number of difficulties and to solve the following tasks: the creation of reliable and stable mathematical models showing the influence of the primary factors on the output parameters; finding a new technique of assigning tolerances for primary factors with regard to economical, technological, and other criteria, the technique being based on the solution of the main problem; well reasoned assignment of nominal values for primary factors which serve as the basis for creating tolerances. Each of the above listed tasks is of independent importance. An attempt is made to give solutions for this problem. The above problem dealing with quality ensuring an mathematically formalized aspect is called the multiple inverse problem.

Vajingortin, L. D.↗
DOE OSTI2021

Model fusion with physics-guided machine learning: Projection-based reduced-order modeling

The unprecedented amount of data generated from experiments, field observations, and large-scale numerical simulations at a wide range of spatiotemporal scales has enabled the rapid advancement of data-driven and especially deep learning models in the field of fluid mechanics. Although these methods are proven successful for many applications, there is a grand challenge of improving their generalizability. This is particularly essential when data-driven models are employed within outer-loop applications like optimization. In this work, we put forth a physics-guided machine learning (PGML) framework that leverages the interpretable physics-based model with a deep learning model. Leveraging a concatenated neural network design from multi-modal data sources, the PGML framework is capable of enhancing the generalizability of data-driven models and effectively protects against or inform about the inaccurate predictions resulting from extrapolation. We apply the PGML framework as a novel model fusion approach combining the physics-based Galerkin projection model and long- to short-term memory (LSTM) network for parametric model order reduction of fluid flows. We demonstrate the improved generalizability of the PGML framework against a purely data-driven approach through the injection of physics features into intermediate LSTM layers. Our quantitative analysis shows that the overall model uncertainty can be reduced through the PGML approach, especially for test data coming from a distribution different than the training data. Moreover, we demonstrate that our approach can be used as an inverse diagnostic tool providing a confidence score associated with models and observations. The proposed framework also allows for multi-fidelity computing by making use of low-fidelity models in the online deployment of quantified data-driven models.

42 ENGINEERING↗