Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Dynamic equivalent”

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 793 records · Page 44
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Numerical and experimental investigation of the flame kernel growth in a methane/air mixture near the lean flammability limit

Lean combustion has the potential to improve the thermal efficiency of spark-ignition engines, but it faces the significant challenge of increased cycle-to-cycle variation due to low mixture reactivity and unstable flame dynamics. Computational fluid dynamics (CFD) employing predictive models can guide engine design and optimize operating strategies for lean combustion. However, ignition and combustion models have rarely been validated at fuel-lean conditions, and a fundamental understanding of the early flame kernel growth process is also lacking for a successful sub-model development. Here, the present study develops a numerical simulation framework used to investigate early flame kernel growth in methane/air mixtures. A nanosecond-pulsed discharge (NPD) approach is employed to effectively decouple the flame kernel growth from the electrical discharge due to their difference in timescales, and equivalence ratios near the experimentally measured lean flammability limit (LFL) are selected to focus on challenging mixture conditions. Three numerical investigations, such as the choice of turbulence modeling, grid size, and grid control strategies, are examined to match both LFL and flame kernel structure measured from experiments. It is demonstrated that a quasi-direct numerical simulation (QDNS) with a fixed grid embedding of 10 μm can predict the LFL as φ CFD =0.61 and match the displacement speed of the kernel’s boundary marked in schlieren images. To predict the LFL and flame kernel shape, a fine grid (Δ≤12.5 μm) is needed to capture the consumption of formaldehyde (CH 2 O) in kernel’s reaction branches attached to the anode, and adaptive mesh refinement is replaced with the fixed embedding due to loss of simulation accuracy. Also, it is found that a large-eddy simulation (LES) using the Dynamic Structure model is not suitable for the NPD-induced flame kernel simulation because artificial sub-grid turbulent kinetic energy induced by shock dynamics alters the flow velocity calculation, resulting in divergence of LES from QDNS. Lastly, the simulation well matches the experimental data for the flame kernel evolution in three mixture conditions (φ = 0.7, 0.61, 0.55), showing toroidal flame kernel expansion and flame kernel growth/extinction.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2021

In-pixel automatic threshold calibration for the CMS Endcap Timing Layer readout chip

We present the implementation and verification of an in-pixel automatic threshold calibration circuit for the CMS Endcap Timing Layer (ETL) in the High-Luminosity LHC upgrade. The discriminator threshold of the ETL readout chip (ETROC) needs to be calibrated regularly to mitigate the circuit baseline change. Traditional methods need a lot of communication through a slow control system hence are time-consuming. This paper describes an in-pixel automatic scheme with improvements in operating time and usability. In this scheme, a sample-accumulation circuit is used to measure the average discriminator output. A binary successive approximation and linear combination scan are applied to find the equivalent baseline. The actual calibration procedure has been first implemented in FPGA firmware and tested with the ETROC front-end prototype chip (ETROC0). The calibration circuit has been implemented with Triple Modular Redundancy (TMR) and verified with Single Event Effects (SEEs) simulation. A complete calibration process lasts 35 ms with a 40 MHz clock. In the worst case, the dynamic and static power consumption are estimated to be 300 uW and 10.4 uW, respectively. The circuit design, implemented in a 65 CMOS technology, will be integrated into ETROC2, the next iteration of the ETROC with a 16 x16 pixel matrix.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2021

The Trinity High Explosive Implosion System: The Foundation for Precision Explosive Applications

This article is set during the 1944 and 1945 final push to complete Project Y—the Manhattan Project at Los Alamos—and focuses primarily on overcoming the challenge of creating and demonstrating a successful convergent explosive implosion to turn a subcritical quantity of plutonium into a critical mass. The critical mass would then efficiently yield kilotons of trinitrotoluene (TNT)-equivalent energy in about a microsecond, demonstrating the implosion atomic bomb concept. This work culminated in the Trinity atomic test near Alamogordo on July 16, 1945. This implosion effect demarcated the approach to explosive science and technology the Laboratory has followed ever since, including development of high-explosive synthesis and formulation, small and large test and diagnostic facilities, shock dynamics theory, high-explosive system design engineering, and three-dimensional implosion modeling and simulation using some of the fastest computers in the world. This work also ushered in a period of broader application of precision high explosives in conventional munitions, demolition, mining and oil exploration, and space travel.

45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEF↗
NASA NTRS1990

State-space design of low-order stabilizers

An algorithm is presented for stabilizing a linear multivariable system with a controller of fixed dynamic order. This is an output feedback stabilization problem. The algorithm attempts to solve this by a sequence of approximate pole assignment problems. The approximation is driven by the optimization of an objective function consisting of a weighted sum of the condition number of the closed-loop eigenvectors and the norm of the difference between the computed and actual equivalent state-feedback controls.

Keel, L. H.↗
NASA NTRS1991

A study of hydrogen diffusion flames using PDF turbulence model

The application of probability density function (pdf) turbulence models is addressed. For the purpose of accurate prediction of turbulent combustion, an algorithm that combines a conventional computational fluid dynamic (CFD) flow solver with the Monte Carlo simulation of the pdf evolution equation was developed. The algorithm was validated using experimental data for a heated turbulent plane jet. The study of H2-F2 diffusion flames was carried out using this algorithm. Numerical results compared favorably with experimental data. The computations show that the flame center shifts as the equivalence ratio changes, and that for the same equivalence ratio, similarity solutions for flames exist.

Hsu, Andrew T.↗
DOE OSTI2023

Effects of unburnt reaction progress on stretch flame dynamics under elevated temperatures

Flames in practical combustors and engines are inevitably subject to the effects of upstream chemical reaction progress and stretch, induced by elevated thermodynamic conditions and flow non-uniformities, respectively. Recent shock tube experiments and simulation studies on flame propagation have shown that flame speed under engine-relevant conditions can be enhanced with non-negligible upstream chemical reaction progress, especially when low-temperature heat release is involved in the unburnt mixture. On the other hand, depending on the mixture equivalence ratio and diluents, nonequidiffusion (including the non-unity Lewis number effect and the preferential diffusion effect) can couple with the flame stretch to fundamentally affect the flame propagation, which can manifest as either facilitation or suppression. Depending on the transport property of a reacting mixture, there hence can be either inhibition or promotion from unburnt reaction progress and stretch effects on flame propagation. In the current work, through one-dimensional numerical simulations of transient planar and spherical flames of n-heptane/air under elevated thermodynamic conditions, the combined effects of upstream chemical reaction progress and stretch on flame propagation are investigated. Results show that for both lean and rich n-heptane/air mixtures, flame speed can be substantially promoted with reaction progress, while the rich mixtures can exhibit opposite stretch dependence after first-stage ignition. Different definitions of effective Lewis number are adopted to explain the change in Markstein length for spherical flame in lean and rich mixtures reformed by the low temperature chemistry. Further, this work fills an important gap in laminar premixed flame research relevant to practical combustion systems and can provide useful insight into local turbulent flame behaviors and phi-sensitivity in engine combustion.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2024

Singular Perturbation-Based Large-Signal Order Reduction of Microgrids for Stability and Accuracy Synthesis With Control

The increasing penetration of distributed energy resources (DERs) highlights the growing importance of microgrids (MGs) in enhancing power system reliability. Employing electromagnetic transient (EMT) analysis in MGs becomes crucial for controlling the rapid transients. However, this requires an accurate but high-order model of power electronics and their underlying control loops, complexifying the stability analysis from the viewpoint of a higher control level. To overcome these challenges, this paper proposes a large-signal order reduction (LSOR) method for MGs with considerations of external control inputs and the detailed dynamics of underlying control levels based on singular perturbation theory (SPT). Specially, we innovatively proposed and strictly proved a general stability and accuracy assessment theorem that allows us to analyze the dynamic stability of a full-order nonlinear system by only leveraging our derived reduced-order model (ROM) and boundary layer model (BLM). Furthermore, this theorem furnishes a set of conditions that determine the accuracy of the developed ROM. Lastly, by embedding such a theorem into the SPT, we propose a novel LSOR approach with guaranteed accuracy and stability analysis equivalence. Case studies are conducted on MG systems to show the effectiveness of the proposed approach.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2020

Numerical Investigation of a Central Fuel Property Hypothesis Under Boosted Spark-Ignition Conditions

In the present work, a central fuel property hypothesis (CFPH), which states that fuel properties are sufficient to provide an indication of a fuel’s performance irrespective of its chemical composition, was numerically investigated. In particular, the objective of the study was to determine whether Research Octane Number (RON) and Motor Octane Number (MON), as fuel properties, are sufficient to describe a fuel’s knock-limited performance under boosted spark-ignition (SI) conditions within the framework of CFPH. To this end, four TPRF-bioblendstock surrogates having different compositions but matched RON (=98) and MON (=90), were first generated using a non-linear regression model based on artificial neural network (ANN). Additionally, three unconventional bioblendstocks were included in the analysis: di-isobutylene (DIB), isobutanol, and Anisole. Skeletal reaction mechanisms were generated for the TPRF-DIB, TPRF-isobutanol, and TPRF-anisole blends from a detailed kinetic mechanism. Thereafter, numerical simulations were performed for the fuel surrogates using the skeletal mechanisms and a virtual cooperative fuel research (CFR) engine model, under a representative boosted operating condition. In the computational fluid dynamics (CFD) model, the G-equation approach was employed to track the turbulent flame front and the well-stirred reactor model combined with the multi-zone binning strategy was used to capture auto-ignition in the end-gas. In addition, laminar flame speed (LFS) was tabulated for each blend as a function of pressure, temperature, and equivalence ratio a priori, and the lookup tables were used to prescribe laminar flame speed as an input to the G-equation model. Parametric spark timing sweeps were performed for each fuel blend to determine the corresponding knock-limited spark advance (KLSA) and 50% burn point (CA50) at the respective KLSA timing. It was observed that despite same RON, MON, and engine operating conditions, the TPRF-anisole blend exhibited markedly different knock-limited performance from the other three blends. This deviation from the octane index (OI) expectation was shown to be caused by differences in laminar flame speed. However, it was found that relatively large fuel-specific differences in LFS (>20%) would have to be present to cause any appreciable deviation from the OI framework. Otherwise, RON and MON would still be robust enough to predict a fuel’s knock-limited performance.

42 ENGINEERING↗
DOE OSTI2022

OC6 Phase Ia: CFD Simulations of the Free-Decay Motion of the DeepCwind Semisubmersible

Currently, the design of floating offshore wind systems is primarily based on mid-fidelity models with empirical drag forces. The tuning of the model coefficients requires data from either experiments or high-fidelity simulations. As part of the OC6 (Offshore Code Comparison Collaboration, Continued, with Correlation, and unCertainty (OC6) is a project under the International Energy Agency Wind Task 30 framework) project, the present investigation explores the latter option. A verification and validation study of computational fluid dynamics (CFD) models of the DeepCwind semisubmersible undergoing free-decay motion is performed. Several institutions provided CFD results for validation against the OC6 experimental campaign. The objective is to evaluate whether the CFD setups of the participants can provide valid estimates of the hydrodynamic damping coefficients needed by mid-fidelity models. The linear and quadratic damping coefficients and the equivalent damping ratio are chosen as metrics for validation. Large numerical uncertainties are estimated for the linear and quadratic damping coefficients; however, the equivalent damping ratios are more consistently predicted with lower uncertainty. Some difference is observed between the experimental and CFD surge-decay motion, which is caused by mechanical damping not considered in the simulations that likely originated from the mooring setup, including a Coulomb-friction-type force. Overall, the simulations and the experiment show reasonable agreement, thus demonstrating the feasibility of using CFD simulations to tune mid-fidelity models.

17 WIND ENERGY↗
NASA NTRS1988

Slewing and vibration control of the SCOLE

A discussion of Slewing and Vibration Control makes the following conclusions: (1) A 2-stage approach is feasible and promising for rapid slewing and precision pointing of SCOLE; (2) Not all bang-bang type of time-minimized slew maneuvers will excite large structural vibrations in SCOLE; and (3) Modal dashpots can be a concentrated high-power vibration control, as well as the usual diffuse (broadband, low-power (low-authority) control. The following recommendations are made: (1) Limit the magnitude of applied forces on reflector to either the 25-lb limit of vernier thrusters on the real Space Shuttle or the 150-lb level equivalent to the cold-gas jets of laboratory SCOLE; (2) to complete stage 2, add an integrated design of LQF/LTR (Linear-Quadratic-Gaussian/Loop-Transfer Recovery) and Modal Dashpots; and, (3) Validate the 2-stage approach using the SCOLE laboratory facility with a comprehensive sequence of integrated designs and experiments coupling nonlinear rigid-body motions with flexible-body dynamics.

Lin, Jiguan Gene↗
NASA NTRS1993

Damage detection using experimentally measured mass and stiffness matrices

A method is presented for locating physical damage or change in a structure using experimentally measured mass and stiffness matrices. The approach uses a recently developed algorithm for transforming a state-space realization into a second order structural model with physical displacements as the generalized coordinates. This is accomplished by first rotating a state-space model of the identified structural dynamics into modal coordinates and approximating the mass normalized modal vectors for the output measurement set. Next, the physical mass, damping and stiffness matrices are synthesized directly from the measured modal parameters. This yields experimental mass and stiffness matrices for the structure without the use of a finite element model or a numerical search. The computed mass and stiffness are asymptotically equivalent to a static condensation of the global physical coordinate model. Techniques for solving the inverse connectivity problem are then developed whereby it is possible to assess the stiffness in a region of the structure bounded by several sensors. Applications to both simulated data and experimental data are used to discuss the effectiveness of the approach.

Peterson, L. D.↗
NASA NTRS2011

Gravito-Inertial Force Resolution in Perception of Synchronized Tilt and Translation

Natural movements in the sagittal plane involve pitch tilt relative to gravity combined with translation motion. The Gravito-Inertial Force (GIF) resolution hypothesis states that the resultant force on the body is perceptually resolved into tilt and translation consistently with the laws of physics. The purpose of this study was to test this hypothesis for human perception during combined tilt and translation motion. EXPERIMENTAL METHODS: Twelve subjects provided verbal reports during 0.3 Hz motion in the dark with 4 types of tilt and/or translation motion: 1) pitch tilt about an interaural axis at +/-10deg or +/-20deg, 2) fore-aft translation with acceleration equivalent to +/-10deg or +/-20deg, 3) combined "in phase" tilt and translation motion resulting in acceleration equivalent to +/-20deg, and 4) "out of phase" tilt and translation motion that maintained the resultant gravito-inertial force aligned with the longitudinal body axis. The amplitude of perceived pitch tilt and translation at the head were obtained during separate trials. MODELING METHODS: Three-dimensional mathematical modeling was performed to test the GIF-resolution hypothesis using a dynamical model. The model encoded GIF-resolution using the standard vector equation, and used an internal model of motion parameters, including gravity. Differential equations conveyed time-varying predictions. The six motion profiles were tested, resulting in predicted perceived amplitude of tilt and translation for each. RESULTS: The modeling results exhibited the same pattern as the experimental results. Most importantly, both modeling and experimental results showed greater perceived tilt during the "in phase" profile than the "out of phase" profile, and greater perceived tilt during combined "in phase" motion than during pure tilt of the same amplitude. However, the model did not predict as much perceived translation as reported by subjects during pure tilt. CONCLUSION: Human perception is consistent with the GIF-resolution hypothesis even when the gravito-inertial force vector remains aligned with the body during periodic motion. Perception is also consistent with GIF-resolution in the opposite condition, when the gravito-inertial force vector angle is enhanced by synchronized tilt and translation.

Wood, Scott J.↗
DOE OSTI2026

Experimental and Computational Study of Weld Drop-Through Hydraulic Effects in Multi-Layer Concentric Pipes Servicing Cryogenic Liquid Hydrogen

The Cryogenic Moderator System (CMS) at the Oak Ridge National Laboratory (ORNL)-Spallation Neutron Source (SNS) supplies liquid hydrogen, through vacuum insulated transfer lines to the moderators. The welding process used to make the transfer lines produces weld drop-through, which can constrict the flow of hydrogen and/or cause thermal bridges between adjacent layers. To ensure proper operation of the CMS it is therefore necessary to conduct a flow test to verify the pressure drop caused by the welds. Measurements in as-built piping sections intended to provide liquid hydrogen service to the SNS-CMS system were performed to test the effect of weld penetrations into the fluid flow area focusing on pressure drop effects. The reference test sections were designed using prototypical piping dimensions, materials and expected penetration welds. A numerical study using Computational Fluid Dynamics (CFD) tools in the transfer lines provided a conversion factor from the pressure drop measured in the lines with water to the expected pressure drop using liquid cryogenic hydrogen. The results show evidence of a maximum pressure drop modification in water of 78 KPa in the tested parts, which scales to liquid hydrogen as PHydrogen = 5.2 KPa. The maximum measured equivalent blockage area percentage was found in the as-built TDM transfer line spool 2 with a value of 31%. The measured values obtained in the as-built transfer lines provided accurate values to estimate the expected total pressure drop in the CMS system providing operational limits for the current recirculatory capacity. The effect of the weld drop-through in hydrogen evacuation capacity scenarios was better evaluated as a result of this study.

Dominguez-Ontiveros, Elvis [ORNL] (ORCID:000000018↗
DOE OSTI2021

Columnar Grain-driven Plasticity and Cracking in Nanotwinned FCC Metals

The mechanisms of strengthening and plasticity in columnar-grained metals with preferentially oriented nano-sized twins have been examined traditionally by considering dislocation processes, but rarely from the perspective of grain boundary (GB) deformation. Here, the effects of GB strain accommodation on plastic deformation in four different columnar-grained nanocrystalline nanotwinned (nt) face-centered-cubic metals (Cu, Ag, Al, and Ni) were studied by large-scale molecular dynamics simulations. It is observed that in tensile deformation parallel to coherent twin boundaries (CTBs), the dislocation mechanisms in each metal are identical and associated with GB emissions of jog and threading dislocations at small and large CTB spacings, respectively. However, CTB strengthening effects are increasingly more pronounced in columnar-grained nt metals as their shear modulus increases, which is rationalized by the dependence of GB stress concentrations on twin size, metal type and strain rate. Also, while flow stresses in nt-Cu, nt-Ag, and nt-Al metals increase linearly with decreasing CTB spacing, a maximum strength limit is reached in nt-Ni below a critical CTB spacing of 6 nm. The strength limit in nt-Ni results from columnar GB cracking induced by prominent GB sliding. For columnar-grained microstructures, GB sliding is equivalent in nt-Ag and nt-Al and slightly lower in nt-Cu but markedly higher in nt-Ni. These findings underscore the importance of new GB deformation mechanisms on plasticity and fracture in columnar-grained nt metals and enrich our understanding of CTB strengthening in fcc metals synthesized in the literature.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Morphotype-resolved characterization of microalgal communities in a nutrient recovery process with ARTiMiS flow imaging microscopy

Microalgae-driven nutrient recovery represents a promising technology for phosphorus removal from wastewater while simultaneously generating biomass that can be valorized to offset treatment costs. As full-scale processes come online, system parameters including biomass composition must be carefully monitored to optimize performance and prevent culture crashes. In this study, flow imaging microscopy (FIM) was leveraged to characterize microalgal community composition in near real-time at a full-scale municipal wastewater treatment plant (WWTP) in Wisconsin, USA, and population and morphotype dynamics were examined to identify relationships between water chemistry, biomass composition, and system performance. Two FIM technologies, FlowCam and ARTiMiS, were evaluated as monitoring tools. ARTiMiS provided a more accurate estimate of total system biomass, and estimates derived from particle area as a proxy for biovolume yielded better approximations than particle counts. Deep learning classification models trained on annotated image libraries demonstrated equivalent performance between FlowCam and ARTiMiS, and convolutional neural network (CNN) classifiers proved significantly more accurate when compared to feature table-based dense neural network (DNN) models. Across a two-year study period, Scenedesmus spp. appeared most important for phosphorus removal, and were negatively impacted by elevated temperatures and increase in nitrite/nitrate concentrations. Chlorella and Monoraphidium also played an important role in phosphorus removal. For both Scenedesmus and Chlorella, smaller morphological types were more often associated with better system performance, whereas larger morphotypes likely associated with stress response(s) correlated with poor phosphorus recovery rates. Furthermore, these results demonstrate the potential of FIM as a critical technology for high-resolution characterization of industrial microalgal processes.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2021

Retreat and Regrowth of the Greenland Ice Sheet During the Last Interglacial as Simulated by the CESM2‐CISM2 Coupled Climate–Ice Sheet Model

Abstract During the Last Interglacial, approximately 129 to 116 ka (thousand years ago), the Arctic summer climate was warmer than the present, and the Greenland Ice Sheet retreated to a smaller extent than its current state. Previous model‐derived and geological reconstruction estimates of the sea‐level contribution of the Greenland Ice Sheet during the Last Interglacial vary widely. Here, we conduct a transient climate simulation from 127 to 119 ka using the Community Earth System Model (CESM2), which includes a dynamic ice sheet component (the Community Ice Sheet Model, CISM2) that is interactively coupled to the atmosphere, land, ocean, and sea ice components. Vegetation distribution is updated every 500 years based on biomes simulated using a monthly climatology to force the BIOME4 equilibrium vegetation model. Results show a substantial retreat of the Greenland Ice Sheet, reaching a minimum extent at 121.9 ka, equivalent to a 3.0 m rise in sea level relative to the present day, followed by gradual regrowth. In contrast, a companion simulation employing static vegetation based on pre‐industrial conditions shows a much smaller ice‐sheet retreat, highlighting the importance of the changes in high‐latitude vegetation distribution for amplifying the ice‐sheet response.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Quantum spin ice response to a magnetic field in the dipole-octupole pyrochlore Ce 2 Zr 2 O 7

The pyrochlore magnet Ce 2 Zr 2 O 7 has attracted much attention as a quantum spin ice candidate whose novelty derives in part from the dipolar-octupolar nature of the Ce 3+ pseudospin-1/2 degrees of freedom it possesses. We report heat capacity measurements on single crystal samples of Ce 2 Zr 2 O 7 down to T~0.1K in a magnetic field along the [$1,\bar{1},0$] direction. These measurements show that the broad hump in the zero-field heat capacity moves higher in temperature with increasing field strength and is split into two separate humps by the [$1,\bar{1},0$] magnetic field at ~2T. These separate features are due to the decomposition of the pyrochlore lattice into effectively decoupled chains for fields in this direction: One set of chains (α chains) is polarized by the field while the other (β chains) remains free. This situation is similar to that observed in the classical spin ices Ho 2 Ti 2 O 7 and Dy 2 Ti 2 O 7 , but with the twist that here the strong transverse exchange interactions produce substantial quantum effects. Our theoretical modeling suggests that the β chains are close to a critical state, with nearly-gapless excitations. We also report elastic and inelastic neutron scattering measurements on single crystal Ce 2 Zr 2 O 7 in [$1,\bar{1},0$] and [0,0,1] magnetic fields at temperatures down to T=0.03K. The elastic scattering behaves consistently with the formation of independent chains for a [$1,\bar{1},0$] field, while the [0,0,1] field produces a single field-induced elastic magnetic Bragg peak at (0,2,0) and equivalent wavevectors, indicating a polarized spin ice state for fields above ~3T. For both [$1,\bar{1},0$] and [0,0,1] magnetic fields, our inelastic neutron scattering results show an approximately dispersionless continuum of scattering that increases in both energy and intensity with increasing field strength. By modeling the complete set of experimental data using numerical linked cluster and semiclassical molecular dynamics calculations, we demonstrate the dominantly multipolar nature of the exchange interactions in Ce 2 Zr 2 O 7 and the smallness of the parameter θ, which controls the mixing between dipolar and octupolar degrees of freedom. Finally, these results support previous estimates of the microscopic exchange parameters and place strong constraints on the theoretical description of this prominent spin ice candidate.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2022

Nonlocal Kernel Network (NKN): a Stable and Resolution-Independent Deep Neural Network.

Neural operators have recently become popular tools for designing solution maps between function spaces in the form of neural networks. Differently from classical scientific machine learning approaches that learn parameters of a known partial differential equation (PDE) for a single instance of the input parameters at a fixed resolution, neural operators approximate the solution map of a family of PDEs [6, 7]. Despite their success, the uses of neural operators are so far restricted to relatively shallow neural networks and confined to learning hidden governing laws. In this work, we propose a novel nonlocal neural operator, which we refer to as nonlocal kernel network (NKN), that is resolution independent, characterized by deep neural networks, and capable of handling a variety of tasks such as learning governing equations and classifying images. Our NKN stems from the interpretation of the neural network as a discrete nonlocal diffusion reaction equation that, in the limit of infinite layers, is equivalent to a parabolic nonlocal equation, whose stability is analyzed via nonlocal vector calculus. The resemblance with integral forms of neural operators allows NKNs to capture long-range dependencies in the feature space, while the continuous treatment of node-to-node interactions makes NKNs resolution independent. The resemblance with neural ODEs, reinterpreted in a nonlocal sense, and the stable network dynamics between layers allow for generalization of NKN’s optimal parameters from shallow to deep networks. This fact enables the use of shallow-to-deep initialization techniques [8]. Our tests show that NKNs outperform baseline methods in both learning governing equations and image classification tasks and generalize well to different resolutions and depths.

97 MATHEMATICS AND COMPUTING↗