Search NASA⌕ Search

SEARCH · Search NASA

Results for “gradient flow dynamics”

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 19 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Plateau Phenomenon in Gradient Descent Training of RELU Networks: Explanation, Quantification, and Avoidance

The ability of neural networks to provide ‘best in class’ approximation across a wide range of applications is well-documented. Nevertheless, the powerful expressivity of neural networks comes to naught if one is unable to effectively train (choose) the parameters defining the network. In general, neural networks are trained by gradient descent type optimization methods,a stochastic variant thereof. In practice, such methods result in the loss function decreases rapidly at the beginning of training but then, after a relatively small number of steps, significantly slow down. The loss may even appear to stagnate over the period of a large number of epochs, only to then suddenly start to decrease fast again for no apparent reason. This so-called plateau phenomenon manifests itself in many learning tasks. The present work aims to identify and quantify the root causes of plateau phenomenon.analysis is carried out in the setting of univariate ReLU networks. No assumptions are made on the number of neurons relative to the number of training data, and our results hold for both the lazy and adaptive regimes. Here, the main findings are: plateaux correspond to periods during which activation patterns remain constant, where activation pattern refers to the number of data points that activate a given neuron; quantification of convergence of the gradient flow dynamics; and, characterization stationary points in terms solutions of local least squares regression lines over subsets of the training data. Based on these conclusions, we propose a new iterative training method, the Active Neuron Least Squares (ANLS), characterised by the explicit adjustment of the activation pattern at each step, which is designed to enable a quick exit from a plateau. Illustrative numerical examples are included throughout.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1994

Modification of the MML turbulence model for adverse pressure gradient flows

Computational fluid dynamics is being used increasingly to predict flows for aerospace propulsion applications, yet there is still a need for an easy to use, computationally inexpensive turbulence model capable of accurately predicting a wide range of turbulent flows. The Baldwin-Lomax model is the most widely used algebraic model, even though it has known difficulties calculating flows with strong adverse pressure gradients and large regions of separation. The modified mixing length model (MML) was developed specifically to handle the separation which occurs on airfoils and has given significantly better results than the Baldwin-Lomax model. The success of these calculations warrants further evaluation and development of MML. The objective of this work was to evaluate the performance of MML for zero and adverse pressure gradient flows, and modify it as needed. The Proteus Navier-Stokes code was used for this study and all results were compared with experimental data and with calculations made using the Baldwin-Lomax algebraic model, which is currently available in Proteus. The MML model was first evaluated for zero pressure gradient flow over a flat plate, then modified to produce the proper boundary layer growth. Additional modifications, based on experimental data for three adverse pressure gradient flows, were also implemented. The adapted model, called MMLPG (modified mixing length model for pressure gradient flows), was then evaluated for a typical propulsion flow problem, flow through a transonic diffuser. Three cases were examined: flow with no shock, a weak shock and a strong shock. The results of these calculations indicate that the objectives of this study have been met. Overall, MMLPG is capable of accurately predicting the adverse pressure gradient flows examined in this study, giving generally better agreement with experimental data than the Baldwin-Lomax model.

Conley, Julianne M.↗
NASA NTRS1996

A dynamic response model for pressure sensors in continuum and high Knudsen number flows with large temperature gradients

This paper develops a dynamic model for pressure sensors in continuum and rarefied flows with longitudinal temperature gradients. The model was developed from the unsteady Navier-Stokes momentum, energy, and continuity equations and was linearized using small perturbations. The energy equation was decoupled from momentum and continuity assuming a polytropic flow process. Rarefied flow conditions were accounted for using a slip flow boundary condition at the tubing wall. The equations were radially averaged and solved assuming gas properties remain constant along a small tubing element. This fundamental solution was used as a building block for arbitrary geometries where fluid properties may also vary longitudinally in the tube. The problem was solved recursively starting at the transducer and working upstream in the tube. Dynamic frequency response tests were performed for continuum flow conditions in the presence of temperature gradients. These tests validated the recursive formulation of the model. Model steady-state behavior was analyzed using the final value theorem. Tests were performed for rarefied flow conditions and compared to the model steady-state response to evaluate the regime of applicability. Model comparisons were excellent for Knudsen numbers up to 0.6. Beyond this point, molecular affects caused model analyses to become inaccurate.

Whitmore, Stephen A.↗
DOE OSTI2021

Deep Kronecker neural networks: A general framework for neural networks with adaptive activation functions

Here we propose a new type of neural networks, Kronecker neural networks (KNNs), that form a general framework for neural networks with adaptive activation functions. KNNs employ the Kronecker product, which provides an efficient way of constructing a very wide network while keeping the number of parameters low. Our theoretical analysis reveals that under suitable conditions, KNNs induce a faster decay of the loss than that by the feed-forward networks. This is also empirically verified through a set of computational examples. Furthermore, under certain technical assumptions, we establish global convergence of gradient descent for KNNs. As a specific case, we propose the Rowdy activation function that is designed to get rid of any saturation region by injecting sinusoidal fluctuations, which include trainable parameters. The proposed Rowdy activation function can be employed in any neural network architecture like feed-forward neural networks, Recurrent neural networks, Convolutional neural networks etc. The effectiveness of KNNs with Rowdy activation is demonstrated through various computational experiments including function approximation using feed-forward neural networks, solution inference of partial differential equations using the physics-informed neural networks, and standard deep learning benchmark problems using convolutional and fully-connected neural networks.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Effects of an External Magnetic Field on Keyhole Mode Laser Melting of 316 Stainless Steel

Keyhole-mode laser melting is an efficient method for joining or cutting large, thick components, but controlling keyhole depth and fluctuations has remained challenging. Applying an external magnetic field can control melt pool flows and indirectly influence keyhole morphology and dynamics. The induced Lorentz force, comprising Seebeck and damping components, plays a crucial role in the melt pool dynamics, depending on temperature gradient, flow rates, and magnetic field orientation and magnitude. This research investigates the effects of an external magnetic field on keyhole behavior during laser spot melting of 316 stainless steel using synchronized high-speed synchrotron X-ray and thermal imaging. Findings revealed that a longitudinal magnetic field (120 mT) increased keyhole depth but exacerbated lateral fluctuations, resulted in a 20% increase in the melt pool temperature gradient and a 27% decrease in cooling rate. Conversely, a transverse magnetic field (760 mT) reduced keyhole depth and improved porosity formation. The findings suggest that a decrease in keyhole depth correlates with a decrease in fluctuations, and vice versa. These insights enhance understanding of external magnetic fields’ impact on laser melting, with implications for improving part quality.

Alamdari, Aslan Bafahm [The Ohio State Univ., Colu↗
NASA NTRSDate not supplied

Updates to Implicit Edge-Based Gradient Methods

In this paper, we report updates to the implicit edge-based gradient methods originally introduced in [H. Nishikawa, AIAA Paper 2020-3048, 2020]. First, we clarify the relationship between gradient accuracy and truncation error and show that the quadratic method involves a free parameter. Then, we provide a complete description and a simplified matrix form of the implicit gradient systems including a consistent boundary treatment, and derive a set of parameters for achieving fourth-order gradient accuracy on regular tetrahedral grids. A stability analysis is performed for a relaxation scheme used to solve the implicit gradient systems, and the result serves as a guide for choosing parameters. Numerical results are shown for accuracy verification and also for realistic inviscid flow problems in three dimensions, including flows with shock waves.

Weighted Least-Squares↗
NASA NTRSDate not supplied

Towards a Third-Order Accurate, Second-Derivative-Free, Shock-Capturing Finite-Volume Method for Hypersonic Flows on Tetrahedral Grids

In this paper, we report progress in the development of a third-order accurate, second-derivative-free, shock-capturing finite-volume solver for three-dimensional unstructured grids. The method is economical in the sense that the computation and storage of second derivatives are not required for third-order accuracy. It is based on point-valued numerical solutions stored at cells, gradients computed and stored at nodes, and an efficient projected-derivative formula that eliminates the need for second derivatives in a quadratic solution interpolation. The projected-derivative formula is also used to eliminate second derivatives from a high-order flux quadrature formula, so that it can be implemented conveniently in the form of a numerical flux at a face center plus a correction term. Similarly, a high-order source quadrature formula can also be implemented in the form of a cell-center point evaluation plus a similar correction term. These features make it relatively straightforward to extend an existing second-order finite-volume code to third-order. This paper reports progress of implementing the method in the NASA VULCAN-CFD code and discusses the implementation of a high-order accurate limiter for shock capturing.

Weighted Least-Squares↗
NASA NTRSDate not supplied

F-ANG+: A 3-D Augmented-Stencil Face-Averaged Nodal-Gradient Cell-Centered Finite-Volume Method for Hypersonic Flows

We describe the extension of a 2-D simplified face-averaged nodal-gradient (F-ANG) method to 3-D and demonstrate that the 3-D simplified F-ANG method is accomplished by augmenting the nodecentered gradient least squares stencil. This augmented stencil F-ANG method is shown to result in advection and diffusion schemes that are stable for hexahedral, prismatic, pyramidal and tetrahedral cells without having to resort to cell-averaged nodal gradients. In addition, we describe the modifications to the augmented stencil required to support the use of wall function boundary conditions. Finally we describe a consistent, face-stencil based multi-dimensional limiter procedure (MLP), and show it to be fully consistent and compatible with the linearity-preserving unstructured- MUSCL (LP-U-MUSCL) scheme for all values of kappa. These methods and schema are implemented in the cell-centered finite-volume code VULCAN-CFD, which is then used to investigate whether the robustness improvements demonstrated in 2-D carry over to 3-D by computing hypersonic flows using mixed-element grids as well as highly adapted tetrahedral grids.

Weighted Least-Squares↗
DOE OSTI2025

The impact of argon addition on hydrogen superpermeation through palladium alloy metal foil pumps during direct internal recycling

Metal foil pumps (MFPs) are a leading technology for the direct internal recycling (DIR) of hydrogen isotopes from the plasma exhaust of fusion devices. MFPs rely on the concept of superpermeation, where plasma-generated atomic hydrogen absorbs into the metal foil, rapidly diffuses, and desorbs downstream. To date, studies of superpermeation have predominantly employed pure hydrogen or in some cases trace levels of impurities. In practice the plasma exhaust may contain significant levels of plasma enhancement gases such as argon, an inert gas with metastable states that can enhance the plasma. In this work, we systematically study the impact of Ar addition on the performance of PdCu and PdAg MFPs at low temperature. Performance was strongly dependent on the DIR fraction. At negligible DIR levels Ar addition did not significantly improve the flux over dilution effects. However, under appreciable DIR operation the flux was enhanced up to 90 % relative to pure H 2 , with the optimal concentration range being 5–10 % Ar exiting the system. Beyond 15 % addition plasma enhancement benefits were offset by dilution. Performance correlated with the atomic H emission, and benefits were more pronounced for PdAg than PdCu. Operation at significant DIR levels dramatically alters the flow dynamics resulting in concentration gradients near the MFP, creating plasma conditions that promote H 2 dissociation.

70 - PLASMA PHYSICS AND FUSION TECHNOLOGY↗
NASA NTRS2005

Consistent Large-Eddy Simulation of a Temporal Mixing Layer Laden with Evaporating Drops: A Posteriori Modelling - Part 2

Large-eddy simulation (LES) is conducted of a three-dimensional temporal mixing layer whose lower stream is initially laden with liquid drops which may evaporate during the simulation. The gas-phase equations are written in an Eulerian frame for two perfect gas species (carrier gas and vapour emanating from the drops), while the liquid-phase equations are written in a Lagrangian frame. The effect of drop evaporation on the gas phase is considered through mass, species, momentum and energy source terms. The drop evolution is modelled using physical drops, or using computational drops to represent the physical drops. Simulations are performed using various LES models previously assessed on a database obtained from direct numerical simulations (DNS). These LES models are for: (i) the subgrid-scale (SGS) fluxes and (ii) the filtered source terms (FSTs) based on computational drops. The LES, which are compared to filtered-and-coarsened (FC) DNS results at the coarser LES grid, are conducted with 64 times fewer grid points than the DNS, and up to 64 times fewer computational than physical drops. It is found that both constant-coefficient and dynamic Smagorinsky SGS-flux models, though numerically stable, are overly dissipative and damp generated small-resolved-scale (SRS) turbulent structures. Although the global growth and mixing predictions of LES using Smagorinsky models are in good agreement with the FC-DNS, the spatial distributions of the drops differ significantly. In contrast, the constant-coefficient scale-similarity model and the dynamic gradient model perform well in predicting most flow features, with the latter model having the advantage of not requiring a priori calibration of the model coefficient. The ability of the dynamic models to determine the model coefficient during LES is found to be essential since the constant-coefficient gradient model, although more accurate than the Smagorinsky model, is not consistently numerically stable despite using DNS-calibrated coefficients. With accurate SGS-flux models, namely scale-similarity and dynamic gradient, the FST model allows up to a 32-fold reduction in computational drops compared to the number of physical drops, without degradation of accuracy; a 64-fold reduction leads to a slight decrease in accuracy.

two-phase flow↗
DOE OSTI2023

Tailoring material microstructure and property in wire-laser directed energy deposition through a wiggle deposition strategy

Developing effective strategies to directly control material microstructure, property, and anisotropy is an active research area in metal additive manufacturing. This work develops a wiggle deposition pattern for wire-laser directed energy deposition (DED) of 316L stainless steel (SS) to modify the solidification texture, particularly in the building direction, in as-deposited samples. Through multi-physics simulation, operando near-infrared imaging, and synchrotron x-ray characterization, it is found that the wiggle deposition strategy induces highly dynamic melt flow and oscillating thermal gradient in the melt pool, which is responsible for the variation of preferable grain growth direction and crystallographic texture in the sample. The specific texture reduces the anisotropy in the tensile strength of as-printed 316L SS samples cut along different directions. Also, it largely increases the ductility along the build direction. Crystal plasticity simulation is performed to correlate the sample texture with mechanical property. In conclusion, this work offers a unique approach for tailoring local properties through the control of melt pool instability by applying different tool paths.

36 MATERIALS SCIENCE↗
NASA NTRS1985

Estimation of particle size based on LDV measurements in a de-accelerating flow field

The accuracy of velocity measurements made with a laser velocimeter is strongly dependent upon the response of the seeding particles to the dynamics of the flow field. The smaller the particle the better the response to flow fluctuations and gradients and therefore the more accurate velocity measurement. In direct conflict is the requirement of light scattering efficiency to obtain signals with the laser velocimeter which, in general, is better as the particle size is increased. In low speed flow fields these two requirements on particle size overlap and accurate measurements may be obtained. However in high speed flows, where the velocity gradients may be severe, very small particles are required to maintain sufficient dynamic response characteristics to follow the flow. Therefore if velocity measurements are to be made in these flows, the laser velocimeter must be designed with sufficient sensitivity to obtain signals from these small particles. An insitu determination of the size distribution of kaolin particles (Al2O3, .2 + or - SiO2 . 2H2O) in the 16-foot Transonic Tunnel and the sensitivity characteristics of the laser velocimeter system is described.

Meyers, J. F.↗
NASA NTRS2002

Temporal gradients in shear stimulate osteoblastic proliferation via ERK1/2 and retinoblastoma protein

Bone cells are subject to interstitial fluid flow (IFF) driven by venous pressure and mechanical loading. Rapid dynamic changes in mechanical loading cause transient gradients in IFF. The effects of pulsatile flow (temporal gradients in fluid shear) on rat UMR106 cells and rat primary osteoblastic cells were studied. Pulsatile flow induced a 95% increase in S-phase UMR106 cells compared with static controls. In contrast, ramped steady flow stimulated only a 3% increase. Similar patterns of S-phase induction were also observed in rat primary osteoblastic cells. Pulsatile flow significantly increased relative UMR106 cell number by 37 and 62% at 1.5 and 24 h, respectively. Pulsatile flow also significantly increased extracellular signal-regulated kinase (ERK1/2) phosphorylation by 418%, whereas ramped steady flow reduced ERK1/2 activation to 17% of control. Correspondingly, retinoblastoma protein was significantly phosphorylated by pulsatile fluid flow. Inhibition of mitogen-activated protein (MAP)/ERK kinase (MEK)1/2 by U0126 (a specific MEK1/2 inhibitor) reduced shear-induced ERK1/2 phosphorylation and cell proliferation. These findings suggest that temporal gradients in fluid shear stress are potent stimuli of bone cell proliferation.

NASA Discipline Cell Biology↗
NASA NTRS1988

Possible explanations of north-south plasma flow in the outer heliosphere and meridional transport of magnetic flux

Observed meridional plasma flow and its connection with other plasma parameters in the outer heliosphere are discussed. The dynamics of the flow are examined locally and compared with observed plasma parameters and a global flow model which predicts such flows in a steady solar wind. The observational evidence supports stream dynamics and associated pressure gradients as responsible for driving the flow. Such a meridional flow may result in a net transport of magnetic flux from regions near the heliographic equator. The amplitude of the observed meridional component of solar wind flow is consistent with observed magnetic flux deficits in the outer heliosphere. The limited coverage of heliographic latitude by Voyager 2 precludes a direct measurement of the full flow pattern; however, the magnitude of reported magnetic flux deficits and the unambiguous, regular variations in the meridional flow suggest that the stream interactions do produce a net movement of magnetic flux away from the heliographic equator.

Mcnutt, R. L., Jr.↗
DOE OSTI2025

Impact of Different Thermal Gradients on the Dynamics of Cylindrical Lithium-ion Cells Subject to Accelerated Aging and on Module Performance

This study investigates the impacts of applying different thermal gradient patterns to cylindrical lithium-ion cells in a module on cell dynamics (temperatures, current flows, state of charge), module performance (evolution of resistance, capacity, and energy versus cycle number), and module lifetime. The thermal gradients were generated using cooling plates (CPs) with three different flow-field designs, namely, straight, perpendicular, and U-turn. The study uses computational fluid dynamics (CFD), the pseudo-two-dimensional (P2D) battery model, capacity loss and increased impedance due to the growth of a solid-electrolyte-interphase, and the electric current distribution from module terminals to cells that depends on the series-parallel electrical connections among the cells. The impact of the thermal gradient (resulting from the CP designs) on the variability in resistance, current, state of charge, and voltage among the cells was analyzed and linked to differences in the module's performance. Applying a thermal gradient to parallel-connected strings of series-connected cells led to variation in the current through each parallel string and an imbalance in the voltage of series-connected cells. Module performance is poorer when the thermal gradient causes a voltage imbalance than when it causes a current imbalance. Module performance becomes the worst when both current variation and voltage imbalance happen together. For instance, the module's lifetime (estimated as reaching 80% of its initial capacity) varied by 5% to 17.5%, depending on the magnitude and pattern of the imposed thermal gradient. As the relative orientation between thermal gradients and cells' electrical connectivity influences the module's performance, appropriate consideration should be given to the choice of the CP, especially if large thermal gradients are allowed.

Battery thermal management↗
NASA NTRS1991

Modeling turbulent boundary layers in adverse pressure gradients

Many of the turbulent layers encountered in practical flows develop in adverse pressure gradients; hence, the dynamics of the thickening and possible separation of the boundary layer has important implications for design practices. What are the key physical processes that govern how a turbulent boundary layer responds to an adverse pressure gradient, and how should these processes be modeled? Despite the ubiquity of such flows in engineering and nature, these equations remain largely unanswered. The turbulence closure models presently used to describe these flows commonly use 'wall functions' that have ad hoc corrections for the effects of pressure gradients. There is, therefore, a practical and theoretical need to examine the effects of adverse pressure gradients on wall bounded turbulent flows in order to develop models based on sound physical principle. The evolution of a turbulent boundary layer on a flat wall with an externally imposed pressure gradient is studied.

Belcher, Stephen E.↗