Search NASA⌕ Search

SEARCH · Search NASA

Results for “filter 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 1,639 records · Page 91

Recent activities within the Aeroservoelasticity Branch at the NASA Langley Research Center

The objective of research in aeroservoelasticity at the NASA Langley Research Center is to enhance the modeling, analysis, and multidisciplinary design methodologies for obtaining multifunction digital control systems for application to flexible flight vehicles. Recent accomplishments are discussed, and a status report on current activities within the Aeroservoelasticity Branch is presented. In the area of modeling, improvements to the Minimum-State Method of approximating unsteady aerodynamics are shown to provide precise, low-order aeroservoelastic models for design and simulation activities. Analytical methods based on Matched Filter Theory and Random Process Theory to provide efficient and direct predictions of the critical gust profile and the time-correlated gust loads for linear structural design considerations are also discussed. Two research projects leading towards improved design methodology are summarized. The first program is developing an integrated structure/control design capability based on hierarchical problem decomposition, multilevel optimization and analytical sensitivities. The second program provides procedures for obtaining low-order, robust digital control laws for aeroelastic applications. In terms of methodology validation and application the current activities associated with the Active Flexible Wing project are reviewed.

Noll, Thomas E.↗

Recent activities within the aeroservoelasticity branch at the NASA Langley Research Center

The objective of research in aeroservoelasticity at the NASA Langley Research Center is to enhance the modeling, analysis, and multidisciplinary design methodologies for obtaining multifunction digital control systems for application to flexible flight vehicles. Recent accomplishments are discussed, and a status report on current activities within the Aeroservoelasticity Branch is presented. In the area of modeling, improvements to the Minimum-State Method of approximating unsteady aerodynamics are shown to provide precise, low-order aeroservoelastic models for design and simulation activities. Analytical methods based on Matched Filter Theory and Random Process Theory to provide efficient and direct predictions of the critical gust profile and the time-correlated gust loads for linear structural design considerations are also discussed. Two research projects leading towards improved design methodology are summarized. The first program is developing an integrated structure/control design capability based on hierarchical problem decomposition, multilevel optimization and analytical sensitivities. The second program provides procedures for obtaining low-order, robust digital control laws for aeroelastic applications. In terms of methodology validation and application the current activities associated with the Active Flexible Wing project are reviewed.

Noll, Thomas↗

A Predictive Attitude Determination Algorithm

In this paper, a new and efficient algorithm is developed for attitude determination from vector observations. The new algorithm, called the Predictive Attitude Determination (PAD) algorithm, is derived from a general nonlinear predictive filter approach. Traditional deterministic algorithms are shown to be suboptimal for anisotropic measurement errors. The major advantage of the PAD algorithm is that it can be easily applied to the case where anisotropic measurement errors exist. Also, an analytical expression is derived for the steady-state attitude error covariance, which is shown to be equivalent to the optimal covariance derived from maximum likelihood techniques. Simulation studies indicate that the new algorithm is able to accurately determine the attitude of a spacecraft, even for radically anisotropic measurement errors.

Markley, F. Landis↗

Airborne and Ground-Based Measurements Using a High-Performance Raman Lidar

A high-performance Raman lidar operating in the UV portion of the spectrum has been used to acquire, for the first time using a single lidar, simultaneous airborne profiles of the water vapor mixing ratio, aerosol backscatter, aerosol extinction, aerosol depolarization and research mode measurements of cloud liquid water, cloud droplet radius, and number density. The Raman Airborne Spectroscopic Lidar (RASL) system was installed in a Beechcraft King Air B200 aircraft and was flown over the mid-Atlantic United States during July August 2007 at altitudes ranging between 5 and 8 km. During these flights, despite suboptimal laser performance and subaperture use of the telescope, all RASL measurement expectations were met, except that of aerosol extinction. Following the Water Vapor Validation Experiment Satellite/Sondes (WAVES_2007) field campaign in the summer of 2007, RASL was installed in a mobile trailer for groundbased use during the Measurements of Humidity and Validation Experiment (MOHAVE-II) field campaign held during October 2007 at the Jet Propulsion Laboratory s Table Mountain Facility in southern California. This ground-based configuration of the lidar hardware is called Atmospheric Lidar for Validation, Interagency Collaboration and Education (ALVICE). During theMOHAVE-II field campaign, during which only nighttime measurements were made, ALVICE demonstrated significant sensitivity to lower-stratospheric water vapor. Numerical simulation and comparisons with a cryogenic frost-point hygrometer are used to demonstrate that a system with the performance characteristics of RASL ALVICE should indeed be able to quantify water vapor well into the lower stratosphere with extended averaging from an elevated location like Table Mountain. The same design considerations that optimize Raman lidar for airborne use on a small research aircraft are, therefore, shown to yield significant dividends in the quantification of lower-stratospheric water vapor. The MOHAVE-II measurements, along with numerical simulation, were used to determine that the likely reason for the suboptimal airborne aerosol extinction performance during theWAVES_2007 campaign was amisaligned interference filter. With full laser power and a properly tuned interference filter,RASL is shown to be capable ofmeasuring themain water vapor and aerosol parameters with temporal resolutions of between 2 and 45 s and spatial resolutions ranging from 30 to 330 m from a flight altitude of 8 km with precision of generally less than 10%, providing performance that is competitive with some airborne Differential Absorption Lidar (DIAL) water vapor and High Spectral Resolution Lidar (HSRL) aerosol instruments. The use of diode-pumped laser technology would improve the performance of an airborne Raman lidar and permit additional instrumentation to be carried on board a small research aircraft. The combined airborne and ground-based measurements presented here demonstrate a level of versatility in Raman lidar that may be impossible to duplicate with any other single lidar technique.

Whiteman, David N.↗

Modeling of RMS Current in CSI Filter Capacitor and Minimum Conduction Loss Operation of CSI-Fed PMSM Drives for Traction Applications

This paper presents a new minimum conduction loss (MCL) torque control algorithm for current source inverter- (CSI-) fed permanent-magnet synchronous machine drives that improves the overall machine drive efficiency by minimizing the combined conduction losses in the inverter and machine at each operating point. First, analytical models of conduction losses in CSI-fed motor drives are presented, and a closed-form expression for optimal d-axis stator current to achieve MCL operation is derived. An expression for the rms current in both wye- and delta-connected CSI output capacitors is derived. A detailed simulation model that emulates the operation of a 100 kW SiC CSI-fed integrated machine drive (IMD) has been developed based on experimental results and finite element analysis. This simulation model is used to evaluate the proposed MCL control algorithm applied to a CSI-fed IMD system for a battery-electric vehicle traction drive, and the predicted total drive system loss with MCL control is compared to predicted losses with maximum-torque-per-ampere (MTPA) and minimum dc-link current (MDCC) control. Results show that MCL control can achieve loss reductions compared to the other two control algorithms over a wide range of operating conditions, with significant loss reductions >20% in the medium-speed regime.

42 ENGINEERING↗

SU-G-TeP2-11: Initial Evaluation of a Novel Split-Filter Dual-Energy CT for Use in Radiation Oncology

Purpose: To perform an initial evaluation of a novel split-filter dual-energy CT (DECT) system with the goal of understanding the clinical utility and limitations of the system for radiation therapy. Methods: Several phantoms were imaged using the split-filter DECT technique on the Siemens Edge CT scanner using a range of clinically-relevant doses. The optimum-contrast reconstruction, the mixed reconstruction, and the monoenergetic reconstructions (ranging from 40 keV to 190 keV) were evaluated. Each image was analyzed for CT number accuracy, uniformity, noise, low-contrast visibility (LCV), spatial resolution and geometric distortion. For comparison purposes, all parameters were evaluated on 120 kVp single-energy CT (SECT) scans used for treatment planning, as well as, a sequential-scan DECT technique for corresponding doses. Results: For all DECT reconstructions no observable geometric distortion was found. Both the optimal-contrast and mixed images demonstrated slight improvements in LCV and noise when compared to the SECT, and slight reductions in CT number accuracy and spatial resolution. The CT numbers trended as expected for the monoenergetic reconstructions, with CT number accuracy within 50 HU for materials of density <2 g/cm3. Spatial resolution increased with energy, and for monoenergetic reconstructions >70 keV the spatial resolution exceeded that of the SECT. The noise in the monoenergetic reconstructions increased with decreasing energy. Thus, the image uniformity, signal-to-noise ratio and LCV were diminished at lower energies (70 keV). Applying iterative reconstruction techniques to the low-energy images reduced noise and improved LCV. The signal-to-noise ratio was stable for energies >100 keV. Conclusion: The initial commissioning of the novel split-filter DECT technology demonstrated favorable results for clinical implementation. The mixed reconstruction showed potential as a replacement for the treatment planning SECT. The image parameters for the monoenergetic reconstructions varied appropriately with energy. This work provides an initial understanding of the limitations and potential applications for monoenergetic imaging.

60 APPLIED LIFE SCIENCES↗

High-Temperature Superconducting/Ferroelectric Structures for Tunable Microwave Components

At the NASA Lewis Research Center, ferroelectric films, such as SrTiO3 and Ba(x)Sr(1-x)TiO3, are being used in conjunction with YBa2Cu3O(7-delta) high-temperature superconducting (HTS) thin films to fabricate tunable microwave components, such as filters, varactors, and local oscillators. These structures capitalize on the variation of the dielectric constant of the ferroelectric film upon the application of a dc electric field as well as on the low microwave losses exhibited by the high-temperature superconducting films relative to their conventional conductor counterparts. (For example, the surface resistance for a YBa2Cu3O(7-delta) thin film at 10 GHz and 77 K is more than two orders of magnitude lower than that of copper at the same frequency and temperature.) SrTiO3 and Ba(x)Sr(1-x)TiO3 films are used because their crystal structure and lattice parameters are similar to those of YBa2Cu3O(7-delta), thus enabling the growth of highly textured YBa2Cu3O(7-delta) films with high critical current densities on the underlying ferroelectric film, or alternatively, of highly textured ferroelectric film on the underlying YBa2Cu3O(7-delta) film. Our efforts have been concentrated so far in determining the deposition parameters required for optimal ferroelectric thin-film growth (i.e., maximum tunability and lowest loss) and in investigating different varactor configurations to determine which geometry is the most advantageous in terms of tunability, losses, and required bias for a given communication application. For example, we have observed that for optimized SrTiO3 films in a parallel plate capacitor, tunabilities of up to 47 percent and dissipation losses (tan d) of 0.05 are attainable at 1 MHz , 80 K, and within the 0- to 5-V bias range. In contrast, for an interdigital configuration, tunabilities of up to 70 percent and tan d ranging from 0.015 to 0.001 (depending on bias) have been observed at 1 MHz and 77 K within the 0- to 100-V bias range. Efforts are underway to use these results in developing tunable receiver front-end preselect filters as well as in low-phase noise, tunable local oscillators for K-band applications. These components represent a hitherto unavailable technology to meet the stringent performance requirements of foreseeable satellite and wireless communication systems (e.g., bandwidth, in-band insertion losses, out-of-band rejection, and noise, amongst others) in a more advantageous fashion than with currently available technology (e.g., dielectric-filled cavity and waveguide filters, and dielectric resonator oscillators). Prototypes of high-temperature superconducting/ferroelectric tunable components such as a low-phase noise K-band local oscillator, a preselect C-band filter, and a low-loss K-band phase shifter are under development at NASA Lewis.

Miranda, Felix A.↗

Optimal space communications techniques

Digital multiplication of two waveforms using delta modulation (DM) is discussed. It is shown that while conventional multiplication of two N bit words requires N2 complexity, multiplication using DM requires complexity which increases linearly with N. Bounds on the signal-to-quantization noise ratio (SNR) resulting from this multiplication are determined and compared with the SNR obtained using standard multiplication techniques. The phase locked loop (PLL) system, consisting of a phase detector, voltage controlled oscillator, and a linear loop filter, is discussed in terms of its design and system advantages. Areas requiring further research are identified.

Schilling, D. L.↗

Co-optimized machine-learned manifold models for large eddy simulation of turbulent combustion

Many modeling approaches in large eddy simulation (LES) of turbulent combustion employ a projection of the thermochemical state onto a low-dimensional manifold within state space to reduce the number of transported variables and hence computational cost. Flamelet-generated manifolds (FGM) is an example of a well-established, physics-based approach, but increasingly, principal component analysis (PCA) is being used as a data-driven method for generating manifold models. For both approaches, the nonlinear relationship between the location on the predefined manifold and the outputs of interest, such as reaction rates, can be tabulated or encoded in a neural network. This work proposes a new approach for manifold modeling that extends these existing approaches. A modified neural network structure simultaneously encodes the definition of the manifold variables, the nonlinear mapping, and the subfilter closure for LES. This allows all three of these aspects of the model to be co-optimized, generating a model from any source of combustion thermochemical state data. The manifold parameterizing variables are constrained to be linear combinations of species, as in FGM and PCA-based models, to aid in interpretability and implementation. For LES, subfilter variances of the manifold variables are also included as inputs. Two types of a priori analysis are performed to evaluate the new approach. In the first, the model is trained on data from one-dimensional premixed flames. In this case, the approach recovers the behavior of flamelet-based manifold approaches, and in fact slightly improves performance by identifying an optimized progress variable. The approach is also applied to data from direct numerical simulations of spherical ignition kernels in isotropic turbulence. For any specified manifold dimensionality, the new approach provides substantially lower prediction errors than a PCA-based model developed from the same data set. Additionally, the LES formulation of the new approach can provide accurate predictions for filtered reaction rates across a variety of filter widths.

97 MATHEMATICS AND COMPUTING↗

Morphology of electrospun nanofibres of polyhydroxybutyrate

This research work focuses on process characteristics of polymer solutions, such as viscosity and electrical conductivity, as well as the parameters of electrospinning using poly-3-hydroxybutyrate modified by titanium dioxide nanoparticles, which have been optimized. Both physical-mechanical characteristics and photooxidation stability of materials have been improved. The structure of materials has been examined by means of X-ray diffraction, differential scanning calorimetry (DSC), IR-spectroscopy, and physical-mechanical testing. The fibrous materials obtained can find a wide application in medicine and filtration techniques as scaffolds for cell growth, filters for body fluids and gas-air media, and sorbents.

36 MATERIALS SCIENCE↗

On-board data processing for the IRAS telescope system

Onboard data processing has been developed for the Infrared Astronomical Satellite (IRAS) telescope system to effectively utilize the RF bandwidth and minimize the ground-data processing costs. Use of filtering techniques, data deglitching circuits, and data encoding algorithms are combined with a microprocessor-controlled telescope system to achieve a system dynamic range of 100,000 with a noise equivalent flux density of no less than 10 to the minus 19th W/sq cm. Extraneous data, e.g., noise due to radiation hits, will be removed prior to transmission, thereby optimizing ground-data processing for the highest sensitivity and reliability. Simulation results are presented which demonstrate the advantages gained through the use of onboard data processing and illustrate the possible merging of these techniques with an array for the future evolution of 'smart' sensors.

Lum, H.↗

The Biolink Implantable Telemetry System

Most biotelemetry applications deal with the moderated data rates of biological signals. Few people have studied the problem of transcutaneous data transmission at the rates required by NASA's Life Sciences-Advanced BioTelemetry System (LS-ABTS). Implanted telemetry eliminate the problems associated with wire breaking the skin, and permits experiments with awake and unrestrained subjects. Our goal is to build a low-power 174-216MHz Radio Frequency (RF) transmitter suitable for short range biosensor and implantable use. The BioLink Implantable Telemetry System (BITS) is composed of three major units: an Analog Data Module (ADM), a Telemetry Transmitter Module (TTM), and a Command Receiver Module (CRM). BioLink incorporates novel low-power techniques to implement a monolithic digital RF transmitter operating at 100kbps, using quadrature phase shift keying (QPSK) modulation in the 174-216MHz ISM band. As the ADM will be specific for each application, we focused on solving the problems associated with a monolithic implementation of the TTM and CRM, and this is the emphasis of this report. A system architecture based on a Frequency-Locked Loop (FLL) Frequency Synthesizer is presented, and a novel differential frequency that eliminates the need for a frequency divider is also shown. A self sizing phase modulation scheme suitable for low power implementation was also developed. A full system-level simulation of the FLL was performed and loop filter parameters were determined. The implantable antenna has been designed, simulated and constructed. An implant package compatible with the ABTS requirements is also being proposed. Extensive work performed at 200MHz in 0.5um complementary metal oxide semiconductors (CMOS) showed the feasibility of integrating the RF transmitter circuits in a single chip. The Hajimiri phase noise model was used to optimize the Voltage Controlled Oscillator (VCO) for minimum power consumption. Two test chips were fabricated in a 0.5pm, 3V CMOS process. Measured phase noise for a 1.5mW, 200MHz ring oscillator VCO is -80dBc/Hz at 100KHZ offset, showing good agreement with the theory. We also propose a novel superregenerative receiver architecture for implementing the command receiver. The superregenerative receiver's simplicity, low cost, and low power consumption has made it the receiver of choice for short-distance data communications, remote control and home automation. We present the design of a superregenerative AM receiver implemented in a 0.5um CMOS technology that operates at 433.92MHz and dissipates only 300uW. Further work entails detailed transistor-level design of the FLL and superregenerative receiver and a monolithic implementation of an implantable transceiver in 0.5um CMOS technology.

Betancourt-Zamora, Rafael J.↗

Single Pixel Characterization of X-Ray TES Microcalorimeter Under AC Bias at MHz Frequencies

In this paper we present the progress made at SRON in the read-out of GSFC x-ray transition-edge sensor (TES) micro-calorimeters in the frequency domain. The experiments reported so far, whose aim was to demonstrate an energy resolution of 2eV at 6 keV with a TES acting as a modulator, were carried out at frequencies below 700 kHz using a standard flux locked loop (FLL) SQUID read-out scheme. The TES read-out suffered from the use of sub-optimal circuit components, large parasitic inductances, low quality factor resonators and poor magnetic field shielding. We have developed a novel experimental set-up, which allows us to test several read-out schemes in a single cryogenic run. In this set-up, the TES pixels are coupled via superconducting transformers to 18 high-Q lithographic LC filters with resonant frequencies ranging between 2 and 5 MHz. The signal is amplified by a two-stage SQUID current sensor and baseband feedback is used to overcome the limited SQUID dynamic range. We study the single pixel performance as a function of TES bias frequency, voltage and perpendicular magnetic field.

Gottardi, L.↗

CFD Based Design of a Filming Injector for N+3 Combustors

An effort was undertaken to perform CFD analysis of fluid flow in Lean-Direct Injection (LDI) combustors with axial swirl-venturi elements for next-generation LDI-3 combustor design. The National Combustion Code (NCC) was used to perform non-reacting and two-phase reacting flow computations for a newly-designed pre-filming type fuel injector LDI-3 injector, in a single-injector and a five-injector array configuration. All computations were performed with a consistent approach of mesh-optimization, spray-modeling, ignition and kinetics-modeling. Computational predictions of the aerodynamics of the single-injector were used to arrive at an optimized main-injector design that meets effective area and fuel-air mixing criteria. Emissions (EINOx) characteristics were predicted for a medium-power engine cycle condition, and will be compared with data when it is made available from experimental measurements. The use of a PDF-like turbulence-chemistry interaction model with NCC's Time-Filtered Navier-Stokes (TFNS) solver is shown to produce a significant impact on the CFD results, when compared with a laminar-chemistry TFNS approach for the five-injector computations.

combustion chambers↗

A PCR Based Microbial Monitoring Alternative Method of Detection and Identification of Microbes Aboard ISS

Previous research has shown that microorganisms and potential human pathogens have been detected on the International Space Station (ISS) with additional introduction of new microflora occurring with every exchange of crew or addition of equipment and supplies. These microbes are readily transferred between crew and subsystems (i.e. ECLSS, environmental control and life support systems). As this can be detrimental to astronaut health and optimal performance of ISS systems, monitoring of systems such as ECLSS to include identification of microbial contaminants could prevent adverse effects on human health and life support systems. Current monitoring on ISS is laborious and utilizes culture based methods followed by sample return to Earth for complete analysis. Future, long-distance spaceflight missions will require real-time monitoring capabilities that enable efficient and rapid assessments of the microbial environment allowing for expedited decisions and more targeted response to cope with anomalies. Polymerase chain reaction (PCR), a molecular microbial monitoring method was chosen and numerous PCR instruments investigated for their potential to perform in microgravity conditions. Using ISS as a test bed for PCR verification in microgravity will enable NASA to assess whether molecular based microbiological sensors may be components of reliable, closed-loop life support and habitation systems in spacecraft, enhancing infrastructure capabilities through increased efficiency, reliability, and time savings by enabling sample analysis on orbit. NASA selected the Water Monitoring Suite as one of the rapid spaceflight hardware demonstration activities utilizing a streamlined process to minimize the time required to fly experimental flight hardware. The RAZOR EX (BioFire Defense, Salt Lake City, UT) system was part of the water monitoring suite and is a commercial off-the-shelf (COTS) real-time PCR instrument designed for field work. The RAZOR EX was originally designed for Department of Defense (DoD) under a small business innovative research (SBIR) grant and is ruggedized, compact and provides a rapid, sample to answer in less than an hour. PCR assays using a fluorescent probe were optimized and spiked with known concentrations of DNA (Pseudomonas aeruginosa) ranging from 0.002 to 20 ng. PCR reagents were lyophilized and configured in customized pouches and tested for flight readiness. Three types of water were used to rehydrate the reagents and demonstrate the fidelity of the PCR reaction in microgravity. Molecular grade deionized water served as a control while filtered and unfiltered ISS potable water served to test for chemical or biological inhibitors. All three types were compared to parallel ground test results. Nine tests were run on ISS (3 of each water type) and the critical threshold cycle (Ct) was compared to parallel ground tests completed at Kennedy Space Center, FL and Johnson Space Center, TX. All concentrations of Pseudomonas aeruginosa DNA were detected. A comparison of the Ct produced in real time PCR indicated similarity between flight and ground samples. There appeared to be no significant difference between flight or ground PCR reactions or between any of the three water types. This testing demonstrated the ability to perform molecular testing during spaceflight operations with similar sensitivity. It will allow for future ground development of molecular protocols and minimize the need for spaceflight testing. Future testing will include development of additional targets including environmental and health related organisms.

Christina Khodadad↗

Development of the GeoNEX Level 2G Products: Exploiting the Diurnal Variability of TOA Reflectance in Atmospheric Correction

This study develops a new atmospheric correction algorithm to generate the Level 2G products, in particular the gap-filled Surface Reflectance at 10-minute time steps, for the Geostationary-NASA Earth Exchange (GeoNEX) project. The algorithm is based on the MODIS MAIAC (Multi-Angle Implementation of Atmospheric Correction) framework but with significant modifications to exploit angular/temporal information from the diurnal variability of the GeoNEX L1G TOA (Top-of-Atmosphere) reflectance. The algorithm starts by evaluating the roughness/smoothness of the diurnal time series of the TOA reflectance. Because rapid changes in TOA reflectance are generally caused by passing clouds or shadows, rough segments of the time series are automatically filtered out while the smooth segments are further tested for brightness and temperature to identify clear-sky and snow-free observations. Next the algorithm runs the MAIAC RTM (Radiative Transfer Model) to retrieve the Ross-Thick-Li-Sparse (RTLS) BRDF model parameters and the daily-mean atmospheric optical depth (AOD) that allow the RTM to optimally simulate the observed diurnal variability of clear-sky TOA reflectance. Once the initial RTLS parameters are retrieved after the algorithm’s burn-in period, they are used as the prior information to predict the AOD level for the next days, while the subsequent clear-sky observations are used to make necessary adjustments to the RTLS parameters in an continuous fashion. This “prediction-analysis” cycle is then iterated to process the full time series of the L1G data, skipping only total-cloudy days or when surface snow is detected. We tested the algorithm over a list of selected AERONET sites. The retrieved results (the daily mean AOD and the RTLS parameters) reasonably agree with the ground-based measurements. Importantly, the results indicate that the diurnal cycles of surface reflectance are continuous functions of the illumination-view geometry. Thus we can use the retrieved RTLS model to accurately fill in data gaps on partial cloudy days. Also, our algorithm is totally independent from the traditional approaches based on the use of spectral band ratios between the shortwave infrared (e.g., 2.2µm) and the visible (e.g., 0.47µm and 0.64µm) bands. Our results thus demonstrate that the high-frequent diurnal geostationary observations contain unique information that helps us improve atmospheric correction of remote sensing data.

GeoNEX↗

A Generalized Framework for Auto-tuning Stencil Computations

This work introduces a generalized framework for automatically tuning stencil computations to achieve superior performance on a broad range of multicore architectures. Stencil (nearest-neighbor) based kernels constitute the core of many important scientific applications involving block-structured grids. Auto-tuning systems search over optimization strategies to find the combination of tunable parameters that maximizes computational efficiency for a given algorithmic kernel. Although the auto-tuning strategy has been successfully applied to libraries, generalized stencil kernels are not amenable to packaging as libraries. Studied kernels in this work include both memory-bound kernels as well as a computation-bound bilateral filtering kernel. We introduce a generalized stencil auto-tuning framework that takes a straightforward Fortran expression of a stencil kernel and automatically generates tuned implementations of the kernel in C or Fortran to achieve performance portability across diverse computer architectures.

97 MATHEMATICS AND COMPUTING↗

DPM: A deep learning PDE augmentation method with application to large-eddy simulation

A framework is introduced that leverages known physics to reduce overfitting in machine learning for scientific applications. The partial differential equation (PDE) that expresses the physics is augmented with a neural network that uses available data to learn a description of the corresponding unknown or unrepresented physics. Training within this combined system corrects for missing, unknown, or erroneously represented physics, including discretization errors associated with the PDE's numerical solution. For optimization of the network within the PDE, an adjoint PDE is solved to provide high-dimensional gradients, and a stochastic adjoint method (SAM) further accelerates training. Additionally, the approach is demonstrated for large-eddy simulation (LES) of turbulence. High-fidelity direct numerical simulations (DNS) of decaying isotropic turbulence provide the training data used to learn sub-filter-scale closures for the filtered Navier–Stokes equations. Out-of-sample comparisons show that the deep learning PDE method outperforms widely-used models, even for filter sizes so large that they become qualitatively incorrect. It also significantly outperforms the same neural network when a priori trained based on simple data mismatch, not accounting for the full PDE. Measures of discretization errors, which are well-known to be consequential in LES, point to the importance of the unified training formulation's design, which without modification corrects for them. For comparable accuracy, simulation runtime is significantly reduced. A relaxation of the typical discrete enforcement of the divergence-free constraint in the solver is also successful, instead allowing the DPM to approximately enforce incompressibility physics. Since the training loss function is not restricted to correspond directly to the closure to be learned, training can incorporate diverse data, including experimental data.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗