Search NASA⌕ Search

SEARCH · Search NASA

Results for “network flow equations”

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 289 records · Page 16

Comparison of Optimal Thermodynamic Models of the Tricarboxylic Acid Cycle from Heterotrophs, Cyanobacteria, and Green Sulfur Bacteria

We have applied a new stochastic simulation approach to predict the metabolite levels, energy flow, and material flux in the different oxidative TCA cycles found in E. coli and Synechococcus sp. PCC 7002, and in the reductive TCA cycle typical of chemolithoautotrophs and phototrophic green sulfur bacteria such as Chlorobaculum tepidum. The simulation approach is based on equations of state and employs an assumption similar to that used in transition state theory. The ability to evaluate the thermodynamics of metabolic pathways allows one to understand the relationship between coupling of energy and material gradients in the environment and the selforganization of stable biological systems, and it is shown that each cycle operates in the direction expected due to its environmental niche. The simulations predict changes in metabolite levels and flux in response to changes in cofactor concentrations that would be hard to predict without an elaborate model based on the law of mass action. In fact, we show that a thermodynamically unfavorable reaction can still have flux in the forward direction when it is part of a reaction network. The ability to predict metabolite levels, energy flow and material flux should be significant for understanding the dynamics of natural systems and for understanding principles for engineering organisms for production of specialty chemicals, such as biofuels.

Thomas, Dennis G.↗

Operation Optimization Using Reinforcement Learning with Integrated Artificial Reasoning Framework

In large and complex systems, operational decision-making requires a systematic analysis with a vast amount of data from both process parameters and component status monitoring. In this paper, we present an integrated artificial reasoning approach for system state transition models that can help operational decision-making with explainable and traceable reasoning. The integrated artificial reasoning framework is a physics-based approach of defining the system structure in a Bayesian network, so we leveraged it in a Markov decision process (MDP) for finding optimal operational solutions. In our proposed framework, the MDP is implemented on a dynamic Bayesian network (DBN), which represents causalities in a system. The multilevel flow modeling was utilized in order to extract these causalities in a more efficient and objective manner. Since multilevel flow modeling is based on the fundamental energy and mass conservation laws, the target system is decomposed into several mass, energy, and information structures, which serve as the basis for a DBN. The MDP consists of the processes of finding a solution for the Bellman equation, which can be derived from the conditional probability equations of the constructed DBN. System operators can capture stochastic system dynamics as multiple subsystem state transitions based on their physical relations and uncertainties coming from the component degradation process or random failures. We analyzed a simplified example system to illustrate finding an optimal operational policy with this approach.

Kim, Junyung↗

Deep Learning Approaches to Surrogates for Solving the Diffusion Equation for Mechanistic Real-World Simulations

In many mechanistic medical, biological, physical, and engineered spatiotemporal dynamic models the numerical solution of partial differential equations (PDEs), especially for diffusion, fluid flow and mechanical relaxation, can make simulations impractically slow. Biological models of tissues and organs often require the simultaneous calculation of the spatial variation of concentration of dozens of diffusing chemical species. One clinical example where rapid calculation of a diffusing field is of use is the estimation of oxygen gradients in the retina, based on imaging of the retinal vasculature, to guide surgical interventions in diabetic retinopathy. Furthermore, the ability to predict blood perfusion and oxygenation may one day guide clinical interventions in diverse settings, i.e., from stent placement in treating heart disease to BOLD fMRI interpretation in evaluating cognitive function (Xie et al., 2019; Lee et al., 2020). Since the quasi-steady-state solutions required for fast-diffusing chemical species like oxygen are particularly computationally costly, we consider the use of a neural network to provide an approximate solution to the steady-state diffusion equation. Machine learning surrogates, neural networks trained to provide approximate solutions to such complicated numerical problems, can often provide speed-ups of several orders of magnitude compared to direct calculation. Surrogates of PDEs could enable use of larger and more detailed models than are possible with direct calculation and can make including such simulations in real-time or near-real time workflows practical. Creating a surrogate requires running the direct calculation tens of thousands of times to generate training data and then training the neural network, both of which are computationally expensive. Often the practical applications of such models require thousands to millions of replica simulations, for example for parameter identification and uncertainty quantification, each of which gains speed from surrogate use and rapidly recovers the up-front costs of surrogate generation. We use a Convolutional Neural Network to approximate the stationary solution to the diffusion equation in the case of two equal-diameter, circular, constant-value sources located at random positions in a two-dimensional square domain with absorbing boundary conditions. Such a configuration caricatures the chemical concentration field of a fast-diffusing species like oxygen in a tissue with two parallel blood vessels in a cross section perpendicular to the two blood vessels. To improve convergence during training, we apply a training approach that uses roll-back to reject stochastic changes to the network that increase the loss function. The trained neural network approximation is about 1000 times faster than the direct calculation for individual replicas. Because different applications will have different criteria for acceptable approximation accuracy, we discuss a variety of loss functions and accuracy estimators that can help select the best network for a particular application. We briefly discuss some of the issues we encountered with overfitting, mismapping of the field values and the geometrical conditions that lead to large absolute and relative errors in the approximate solution.

60 APPLIED LIFE SCIENCES↗

In Search of Data-Driven Improvements to RANS Models Applied to Separated Flows

The goal of this work is to improve the capability of Reynolds-averaged Navier-Stokes turbulence models for separated flows using data-driven enhancements. The resulting model should be “universal” in the sense that it can be used by anyone and applied to as many flows as possible without concern for unusual or detrimental behavior. At worst, the data-driven corrections should not degrade the accuracy of the baseline model (in this case the Spalart-Allmaras one-equation model), while preserving the Galilean invariance and similar theoretical qualities of the original model. In the literature, most current data-driven improvements to turbulence models are only applicable to very similar types of cases as those used to train the model for a specific class of flows. In this work, the impact of using a wide array of cases in the machine-learning training is described. Unwanted behaviors from trained neural networks are examined, and possible mitigation strategies are proposed. However, to date, consistent and broadly applicable data-driven improvements for separated flows have not been achieved.

turbulence modeling↗

When ancient numerical demons meet physics-informed machine learning: adjoint-based gradients for implicit differentiable modeling

Recent advances in differentiable modeling, a genre of physics-informed machine learning that trains neural networks (NNs) together with process-based equations, have shown promise in enhancing hydrological models' accuracy, interpretability, and knowledge-discovery potential. Current differentiable models are efficient for NN-based parameter regionalization, but the simple explicit numerical schemes paired with sequential calculations (operator splitting) can incur numerical errors whose impacts on models' representation power and learned parameters are not clear. Implicit schemes, however, cannot rely on automatic differentiation to calculate gradients due to potential issues of gradient vanishing and memory demand. Here we propose a “discretize-then-optimize” adjoint method to enable differentiable implicit numerical schemes for the first time for large-scale hydrological modeling. The adjoint model demonstrates comprehensively improved performance, with Kling–Gupta efficiency coefficients, peak-flow and low-flow metrics, and evapotranspiration that moderately surpass the already-competitive explicit model. Therefore, the previous sequential-calculation approach had a detrimental impact on the model's ability to represent hydrological dynamics. Furthermore, with a structural update that describes capillary rise, the adjoint model can better describe baseflow in arid regions and also produce low flows that outperform even pure machine learning methods such as long short-term memory networks. The adjoint model rectified some parameter distortions but did not alter spatial parameter distributions, demonstrating the robustness of regionalized parameterization. Despite higher computational expenses and modest improvements, the adjoint model's success removes the barrier for complex implicit schemes to enrich differentiable modeling in hydrology.

58 GEOSCIENCES↗

Multi-Body Analysis of the 1/5 Scale Wind Tunnel Model of the V-22 Tiltrotor

The paper presents a multi-body analysis of the 1/5 scale wind tunnel model of the V-22 tiltrotor, the Wing and Rotor Aeroelastic Testing System (WRATS), currently tested at NASA Langley Research Center. An original multi-body formulation has been developed at the Dipartimento di Ingegneria Aerospaziale of the Politecnico di Milano, Italy. It is based on the direct writing of the equilibrium equations of independent rigid bodies, connected by kinematic constraints that result in the addition of algebraic constraint equations, and by dynamic constraints, that directly contribute to the equilibrium equations. The formulation has been extended to the simultaneous solution of interdisciplinary problems by modeling electric and hydraulic networks, for aeroservoelastic problems. The code has been tailored to the modeling of rotorcrafts while preserving a complete generality. A family of aerodynamic elements has been introduced to model high aspect aerodynamic surfaces, based on the strip theory, with quasi-steady aerodynamic coefficients, compressibility, post-stall interpolation of experimental data, dynamic stall modeling, and radial flow drag. Different models for the induced velocity of the rotor can be used, from uniform velocity to dynamic in flow. A complete dynamic and aeroelastic analysis of the model of the V-22 tiltrotor has been performed, to assess the validity of the formulation and to exploit the unique features of multi-body analysis with respect to conventional comprehensive rotorcraft codes; These are the ability to model the exact kinematics of mechanical systems, and the possibility to simulate unusual maneuvers and unusual flight conditions, that are particular to the tiltrotor, e.g. the conversion maneuver. A complete modal validation of the analytical model has been performed, to assess the ability to reproduce the correct dynamics of the system with a relatively coarse beam model of the semispan wing, pylon and rotor. Particular care has been used to model the kinematics of the gimbal joint, that characterizes the rotor hub, and of the control system, consisting in the entire swashplate mechanism. The kinematics of the fixed and the rotating plates have been modeled, with variable length control links used to input the controls, the rotating flexible links, the pitch horns and the pitch bearings. The investigations took advantage of concurring wind tunnel test runs, that were performed in August 1998, and allowed the acquisition of data specific to the multi-body analysis.

Ghiringhelli, G. L.↗

Structure Learning and Statistical Estimation in Distribution Networks - Part II

Limited placement of real-time monitoring devices in the distribution grid, recent trends notwithstanding, has prevented the easy implementation of demand-response and other smart grid applications. Part I of this paper discusses the problem of learning the operational structure of the grid from nodal voltage measurements. In this work (Part II), the learning of the operational radial structure is coupled with the problem of estimating nodal consumption statistics and inferring the line parameters in the grid. Based on a Linear-Coupled(LC) approximation of AC power flows equations, polynomial time algorithms are designed to identify the structure and estimate nodal load characteristics and/or line parameters in the grid using the available nodal voltage measurements. Then the structure learning algorithm is extended to cases with missing data, where available observations are limited to a fraction of the grid nodes. The efficacy of the presented algorithms are demonstrated through simulations on several distribution test cases.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Coarse-grained description of cosmic structure from Szekeres models

We show that the full dynamical freedom of the well known Szekeres models allows for the description of elaborated 3-dimensional networks of cold dark matter structures (over-densities and/or density voids) undergoing ''pancake'' collapse. By reducing Einstein's field equations to a set of evolution equations, which themselves reduce in the linear limit to evolution equations for linear perturbations, we determine the dynamics of such structures, with the spatial comoving location of each structure uniquely specified by standard early Universe initial conditions. By means of a representative example we examine in detail the density contrast, the Hubble flow and peculiar velocities of structures that evolved, from linear initial data at the last scattering surface, to fully non-linear 10–20 Mpc scale configurations today. To motivate further research, we provide a qualitative discussion on the connection of Szekeres models with linear perturbations and the pancake collapse of the Zeldovich approximation. This type of structure modelling provides a coarse grained—but fully relativistic non-linear and non-perturbative —description of evolving large scale cosmic structures before their virialisation, and as such it has an enormous potential for applications in cosmological research.

79 ASTRONOMY AND ASTROPHYSICS↗

Distribution System State Estimation in the Presence of High Solar Penetration

In recent years, the low-to-medium voltage distribution networks are experiencing rising levels of distributed energy resources, including renewable generation, along with improved sensing, communication and automation infrastructure. As such, state estimation methods for distribution systems are becoming increasingly relevant as a means to enable better control strategies that can both leverage the benefits and mitigate the risks associated with high penetration of variable and uncertain distributed generation resources. Primary challenges of this problem include modeling complexities (nonlinear non-convex power-flow equations), limited availability of sensor measurements, and high penetration of uncertain renewable generation. This paper formulates the distribution system state estimation as a nonlinear weighted least squares problem, based on sensor measurements as well as forecasts data (both load and generation). We investigate the sensitivity of state estimator accuracy to (load/generation) forecast uncertainties, sensor accuracy, and sensor coverage levels.

Distribution system state estimation, nonlinear op↗

CAVE: A computer code for two-dimensional transient heating analysis of conceptual thermal protection systems for hypersonic vehicles

A digital computer code CAVE (Conduction Analysis Via Eigenvalues), which finds application in the analysis of two dimensional transient heating of hypersonic vehicles is described. The CAVE is written in FORTRAN 4 and is operational on both IBM 360-67 and CDC 6600 computers. The method of solution is a hybrid analytical numerical technique that is inherently stable permitting large time steps even with the best of conductors having the finest of mesh size. The aerodynamic heating boundary conditions are calculated by the code based on the input flight trajectory or can optionally be calculated external to the code and then entered as input data. The code computes the network conduction and convection links, as well as capacitance values, given basic geometrical and mesh sizes, for four generations (leading edges, cooled panels, X-24C structure and slabs). Input and output formats are presented and explained. Sample problems are included. A brief summary of the hybrid analytical-numerical technique, which utilizes eigenvalues (thermal frequencies) and eigenvectors (thermal mode vectors) is given along with aerodynamic heating equations that have been incorporated in the code and flow charts.

Rathjen, K. A.↗

Multilevel Algorithm for Atmospheric Data Assimilation

A multiscale algorithm for the problem of optimal statistical interpolation of observed data has been developed. This problem includes the calculation of the vector of the 'analyzed' (best estimated) atmosphere flow field w(sup a) by the formula: w(sup a) = w(sup f) + P(sup f) H(sup T) y, where the quantity y is defined by the equation (H P(sup f) H(sup T) + R)y = w(sup o) - H w(sup f), using the given model forecast first guess w(sup f) and the vector of observations w(sup o); H is an interpolation operator from the regular grid to the observation network, P(sup f) is the forecast error covariance matrix, and R is the observation error covariance matrix. At this initial stage the case of univariate analysis of single level radiosonde height data is considered. The matrix R is assumed to be diagonal, and the matrix P(sup f) is assumed to be given by the formula P(sub ij)(sup f) = sigma(sub i)(sup f) mu(sub ij) sigma(sub j)(sub f), where mu(sub ij) is a smooth, decreasing function of the distance between the i-th and the j-th points. In this paper we describe a multiscale iterative process based on a multiresolution, simultaneous displacement technique and a localized variational calculation of iteration parameters.

Brandt, Achi↗

Objective Lightning Forecasting at Kennedy Space Center/Cape Canaveral Air Force Station using Cloud-to-Ground Lightning Surveillance System Data

The 45th Weather Squadron (45 WS) forecasters at Cape Canaveral Air Force Station (CCAFS) in Florida include a probability of thunderstorm occurrence in their daily morning briefings. This information is used by personnel involved in determining the possibility of violating Launch Commit Criteria, evaluating Flight Rules for the Space Shuttle, and daily planning for ground operation activities on Kennedy Space Center (KSC)/CCAFS. Much of the current lightning probability forecast is based on a subjective analysis of model and observational data. The forecasters requested that a lightning probability forecast tool based on statistical analysis of historical warm-season (May - September) data be developed in order to increase the objectivity of the daily thunderstorm probability forecast. The tool is a set of statistical lightning forecast equations that provide a lightning occurrence probability for the day by 1100 UTC (0700 EDT) during the warm season. This study used 15 years (1989-2003) of warm season data to develop the objective forecast equations. The local CCAFS 1000 UTC sounding was used to calculate stability parameters for equation predictors. The Cloud-to-Ground Lightning Surveillance System (CGLSS) data were used to determine lightning occurrence for each day. The CGLSS data have been found to be more reliable indicators of lightning in the area than surface observations through local informal analyses. This work was based on the results from two earlier research projects. Everitt (1999) used surface observations and rawinsonde data to develop logistic regression equations that forecast the daily thunderstorm probability at CCAFS. The Everitt (1999) equations showed an improvement in skill over the Neumann-Pfeffer thunderstorm index (Neumann 1971), which uses multiple linear regression, and also persistence and climatology forecasts. Lericos et al. (2002) developed lightning distributions over the Florida peninsula based on specific flow regimes. The flow regimes were inferred from the average wind direction in the 1000-700 mb layer at Miami (MIA), Tampa (TBW), and Jacksonville (JAX), Florida, and the lightning data were from the National Lightning Detection Network. The results suggested that the daily flow regime may be an important predictor of lightning occurrence on KSC/CCAFS.

Lambert, Winifred↗

Predicting departure from nucleate boiling with advanced data- and physics-driven approaches

The subcooled and low-quality saturated flow boiling critical heat flux (CHF) corresponding to the departure from nucleate boiling (DNB) is one of the major limiting factors in the design and operation of pressurized water reactors (PWRs). Due to lack of agreement on its physical mechanisms, development of an accurate DNB-type CHF model has been elusive. The various steady-state (SS) predictive tools that have been proposed can be grouped into two categories: (1) data-driven methods—namely best-fit empirical correlations and look-up tables that result in relatively good agreement with specific experimental datasets but often fail to extend beyond their ranges of validity, and (2) the physics-based mechanistic models that rely on reasonable yet limited understanding of the underlying physics and apply constitutive relations to close the conservation equations. On one hand, with recent advances in computational capabilities and optimization techniques, machine learning (ML)-based methods constitute an alternative and a more advanced, data-driven approach. With enough data, they can be particularly useful in engineering fields where the physical phenomena are complex and challenging to model. On the other hand, a state-of-the-art physics-driven approach would not only provide more accurate prediction of SS CHF, but it would also help better understand and model transient applications. The data-driven regression models use deep feed-forward neural networks (NNs) and random forests to cross-validate with 1,865 CHF test cases, covering a wide range of flow conditions and channel geometries. The best-estimate ML-based predictors compare favorably with the widely used look-up table for annulus and plate, and sensitivity analysis has confirmed their effectiveness. The key advantage of ML-based methods is their online extensibility of applicability domain. The proposed physics-driven approach combines key assumptions and parameters in the relatively well-accepted theories of liquid sublayer dryout and near-wall bubble crowding. A more realistic understanding of local mechanisms has been modeled. The new model has been optimized and validated against 1,439 tube data, and it shows considerably improved performance when compared to a recent mechanistic model, demonstrating unbiased close agreement with measurements over a wide range of operating conditions. Future work will extend the improved physics-driven model to non-tube geometry applications and will couple the model with ML via a hybrid approach. The mechanistic SS work will also serve to improve understanding and to model transient CHF scenarios.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Adversarial uncertainty quantification in physics-informed neural networks

Here, we introduce a deep learning framework for quantifying and propagating uncertainty in systems governed by non-linear differential equations using physics-informed neural networks. Specifically, we employ latent variable models to construct probabilistic representations for the system states, and put forth an adversarial inference procedure for training them on data, while constraining their predictions to satisfy given physical laws expressed by partial differential equations. Such physics-informed constraints provide a regularization mechanism for effectively training deep generative models as surrogates of physical systems in which the cost of data acquisition is high, and training data-sets are typically small. This provides a flexible framework for characterizing uncertainty in the outputs of physical systems due to randomness in their inputs or noise in their observations that entirely bypasses the need for repeatedly sampling expensive experiments or numerical simulators. Moreover, we demonstrate the effectiveness of our approach through a series of examples involving uncertainty propagation in non-linear conservation laws, and the discovery of constitutive laws for flow through porous media directly from noisy data.

97 MATHEMATICS AND COMPUTING↗

Code Demonstration: Approximation of nearly-periodic symplectic maps via structure-preserving neural networks

In this notebook, we use the symplectic gyroceptron architecture from [Duruisseaux et al., 2022] to learn a surrogate map for the nearly-periodic symplectic flow map associated to a nearly-periodic Hamiltonian system composed of two nonlinearly coupled oscillators, where one of them oscillates significantly faster than the other: $\Big\lbrace$ $ {\dot{q}_1 = p_1 \hspace{5mm} \dot{p}_1 = -q_1 - \epsilon \partial_{q1}U(q_1 , q_2 )} \atop {\dot{q}_2 = \epsilon p_2 \hspace{4mm} \dot{p}_2 = -\epsilon q_2 - \epsilon \partial_{q2}U(q_1 , q_2 )}$. These equations of motion are the Hamilton's equations associated to the Hamiltonian $H_{\epsilon}(q_1,q_2,p_1,p_2)=\frac{1}{2}(q^2_1+p^2_1)+\frac{1}{2}\epsilon(q^2_2+p^2_2)+\epsilon$$U(q_1,q_2)$.

97 MATHEMATICS AND COMPUTING↗

Assessing the Optimality of LinDist3Flow for Optimal Tap Selection of Step Voltage Regulators in Unbalanced Distribution Networks: Preprint

The adoption of distributed energy resources such as photovoltaics (PVs) has increased dramatically during the previous decade. The increased penetration of PVs into distribution networks (DNs) can cause voltage fluctuations that have to be mitigated. One of the key utility assets employed to this end are step-voltage regulators (SVRs). It is desirable to include tap selection of SVRs in optimal power flow (OPF) routines, a task that turns out to be challenging because the resultant OPF problem is nonconvex with added complexities stemming from accurate SVR modeling. While several convex relaxations based on semi-definite programming (SDP) have been presented in the literature for optimal tap selection, SDP based schemes do not scale well and are challenging to implement in large-scale planning or operational frameworks. This paper deals with the optimal tap selection (OPTS) problem for wye-connected SVRs using linear approximations of power flow equations. Specifically, the LinDist3Flow model is adopted and the effective SVR ratio is assumed to be continuous–enabling the formulation of a problem called LinDist3Flow-OPTS, which amounts to a linear program. The scalability and optimality gap of LinDist3Flow-OPTS are evaluated with respect to existing SDP-based and nonlinear programming techniques for optimal tap selection in three standard feeders, namely, the IEEE 13-bus, 123-bus, and 8500-node DNs. For all DNs considered, LinDist3Flow-OPTS achieves an optimality gap of approximately 1% or less while significantly lowering the computational burden.

linear approximations↗

Generalized Fluid System Simulation Program, Version 6.0

The Generalized Fluid System Simulation Program (GFSSP) is a general purpose computer program for analyzing steady state and time-dependent flow rates, pressures, temperatures, and concentrations in a complex flow network. The program is capable of modeling real fluids with phase changes, compressibility, mixture thermodynamics, conjugate heat transfer between solid and fluid, fluid transients, pumps, compressors, and external body forces such as gravity and centrifugal. The thermofluid system to be analyzed is discretized into nodes, branches, and conductors. The scalar properties such as pressure, temperature, and concentrations are calculated at nodes. Mass flow rates and heat transfer rates are computed in branches and conductors. The graphical user interface allows users to build their models using the 'point, drag, and click' method; the users can also run their models and post-process the results in the same environment. Two thermodynamic property programs (GASP/WASP and GASPAK) provide required thermodynamic and thermophysical properties for 36 fluids: helium, methane, neon, nitrogen, carbon monoxide, oxygen, argon, carbon dioxide, fluorine, hydrogen, parahydrogen, water, kerosene (RP-1), isobutene, butane, deuterium, ethane, ethylene, hydrogen sulfide, krypton, propane, xenon, R-11, R-12, R-22, R-32, R-123, R-124, R-125, R-134A, R-152A, nitrogen trifluoride, ammonia, hydrogen peroxide, and air. The program also provides the options of using any incompressible fluid with constant density and viscosity or ideal gas. The users can also supply property tables for fluids that are not in the library. Twenty-four different resistance/source options are provided for modeling momentum sources or sinks in the branches. These options include pipe flow, flow through a restriction, noncircular duct, pipe flow with entrance and/or exit losses, thin sharp orifice, thick orifice, square edge reduction, square edge expansion, rotating annular duct, rotating radial duct, labyrinth seal, parallel plates, common fittings and valves, pump characteristics, pump power, valve with a given loss coefficient, Joule-Thompson device, control valve, heat exchanger core, parallel tube, and compressible orifice. The program has the provision of including additional resistance options through User Subroutines. GFSSP employs a finite volume formulation of mass, momentum, and energy conservation equations in conjunction with the thermodynamic equations of state for real fluids as well as energy conservation equations for the solid. The system of equations describing the fluid network is solved by a hybrid numerical method that is a combination of the Newton-Raphson and successive substitution methods. The application and verification of the code has been demonstrated through 30 example problems.

Majumdar, A. K.↗

Extended Galerkin Neural Network Approximation of Singular Variational Problems with Error Control

We present extended Galerkin neural networks, a variational framework for approximating general boundary value problems (BVPs) with error control. The main contributions of this work are (1) a rigorous theory guiding the construction of new weighted least squares variational formulations suitable for use in neural network approximation of general BVPs, and (2) an “extended” feedforward network architecture which incorporates and is even capable of learning singular solution structures, thus greatly improving approximability of singular solutions. Furthermore, numerical results are presented for several problems, including steady Stokes flow around reentrant corners and in convex corners with Moffatt eddies in order to demonstrate efficacy of the method.

a posteriori error estimate↗