Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Dimension reduction”

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 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
NASA NTRS1973

Direct numerical solution of three-dimensional equations containing elliptic operators.

A direct three-dimensional elliptic solver is presented for application in a wide class of numerical methods for solving partial differential equations in physics and engineering. The derived algorithm and FORTRAN code implement Buzbee, Golub and Nielson's proposed extension of Buneman's Cyclic-Reduction Poisson solver to three dimensions. Both a 'most direct' cyclic reduction and a revised method (to eliminate roundoff error difficulties) are derived. Tests on an IBM 360/67 computer, using various optional combinations of subroutines, showed significant differences in accuracy and computing time, with the optimum subroutine combination depending on mesh size.

Martin, E. D.↗
NASA NTRS1973

Digital processing of radiographic images

Some techniques are presented and the software documentation for the digital enhancement of radiographs. Both image handling and image processing operations are considered. The image handling operations dealt with are: (1) conversion of format of data from packed to unpacked and vice versa; (2) automatic extraction of image data arrays; (3) transposition and 90 deg rotations of large data arrays; (4) translation of data arrays for registration; and (5) reduction of the dimensions of data arrays by integral factors. Both the frequency and the spatial domain approaches are presented for the design and implementation of the image processing operation. It is shown that spatial domain recursive implementation of filters is much faster than nonrecursive implementations using fast fourier transforms (FFT) for the cases of interest in this work. The recursive implementation of a class of matched filters for enhancing image signal to noise ratio is described. Test patterns are used to illustrate the filtering operations. The application of the techniques to radiographic images of metallic structures is demonstrated through several examples.

Bond, A. D.↗
NASA NTRS1981

Radiatively coupled thermionic power system concept

It is shown that thermionic converters at moderate emitter temperatures of about 1600 K can be designed for a radiatively coupled 100 kWe device. A nuclear reactor is a primary heat source, with heat pipes extracting heat from the reactor and distributing it over a large surface opposite an array of thermionic energy converters. The radiative heat transfer across the vacuum gap heats up the thermionic emitters, and excess heat from the converters is radiated from the collector electrodes to the vacuum of space; the heat transfer is controlled by the energy density to be transferred, and by the temperature differential between the heat source and the heat receiver. It is concluded that this system achieves isolation of power converter modules from the heat source, elimination of additional radiators, and reduction of converter dimensions.

Shimada, K.↗
NASA NTRS1981

Grid evolution in time asymptotic problems

The selection of the proper coordinate system in solving any fluid flow or heat transfer problem is a very important consideration. A new technique of moving mesh points in physical space is introduced so as to reduce the error in a computed asymptotic solution relative to that obtained using a fixed mesh. The technique has been used to solve the simple viscous Burgers' equation in one and two dimensions. Substantial error reductions are demonstrated. The treatment of boundary points and the effect of using different error criteria in generating grids are discussed.

Rai, M. M.↗
NASA NTRS1985

An atlas of objectively analyzed atmospheric cross sections, 1973-1980

Atmospheric variability over time scales greater than one month is conceptually simplified and readily recognized from vertical cross-sections of zonal-monthly mean data. The reduction to two dimensions, latitude and height, explicitly eliminates all zonal waves but implicity retains their effects on the thermal-pressure fields and the dynamically related zonal wind fields. This atlas contains 96 examples, spanning all latitudes in both the northern and southern hemispheres and two decades in pressure, from 1000 to 10 mb. Four analyses, representing each month from January 1973 through December 1980, depicts the potential virtual temperature, the observed zonal wind velocity, the virtual temperature and the geostrophic zonal wind velocity. Each variable is contoured at a close interval to facilitate visual estimates of stability and vorticity via their gradients. The analyses are generated and contoured by objective computer methods from just one data source: in situ measurements from the conventional rawin-radiosonde system. Although the analyses are independently made at constant pressure levels (the mandatory levels) the cross-sections are drawn with geopotential height as the ordinate. With this ordinate one can observe the seasonal expansion and contraction of the earth's atmosphere, especially that of the polar stratosphere. Also, the quasi-biannual cycle can be identified and studied directly from successive cross-sections.

Goodman, J.↗
NASA NTRS1989

Flexible robot control: Modeling and experiments

Described here is a model and its use in experimental studies of flexible manipulators. The analytical model uses the equivalent of Rayleigh's method to approximate the displaced shape of a flexible link as the static elastic displacement which would occur under end rotations as applied at the joints. The generalized coordinates are thereby expressly compatible with joint motions and rotations in serial link manipulators, because the amplitude variables are simply the end rotations between the flexible link and the chord connecting the end points. The equations for the system dynamics are quite simple and can readily be formulated for the multi-link, three-dimensional case. When the flexible links possess mass and (polar moment of) inertia which are small compared to the concentrated mass and inertia at the joints, the analytical model is exact and displays the additional advantage of reduction in system dimension for the governing equations. Four series of pilot tests have been completed. Studies on a planar single-link system were conducted at Carnegie-Mellon University, and tests conducted at Toshiba Corporation on a planar two-link system were then incorporated into the study. A single link system under three-dimensional motion, displaying biaxial flexure, was then tested at Carnegie-Mellon.

Oppenheim, Irving J.↗
NASA NTRS1990

Extended inflation from higher dimensional theories

The possibility is considered that higher dimensional theories may, upon reduction to four dimensions, allow extended inflation to occur. Two separate models are analayzed. One is a very simple toy model consisting of higher dimensional gravity coupled to a scalar field whose potential allows for a first-order phase transition. The other is a more sophisticated model incorporating the effects of non-trivial field configurations (monopole, Casimir, and fermion bilinear condensate effects) that yield a non-trivial potential for the radius of the internal space. It was found that extended inflation does not occur in these models. It was also found that the bubble nucleation rate in these theories is time dependent unlike the case in the original version of extended inflation.

Holman, Richard↗
NASA NTRS1993

Modal ring method for the scattering of sound

The modal element method for acoustic scattering can be simplified when the scattering body is rigid. In this simplified method, called the modal ring method, the scattering body is represented by a ring of triangular finite elements forming the outer surface. The acoustic pressure is calculated at the element nodes. The pressure in the infinite computational region surrounding the body is represented analytically by an eigenfunction expansion. The two solution forms are coupled by the continuity of pressure and velocity on the body surface. The modal ring method effectively reduces the two-dimensional scattering problem to a one-dimensional problem capable of handling very high frequency scattering. In contrast to the boundary element method or the method of moments, which perform a similar reduction in problem dimension, the model line method has the added advantage of having a highly banded solution matrix requiring considerably less computer storage. The method shows excellent agreement with analytic results for scattering from rigid circular cylinders over a wide frequency range (1 is equal to or less than ka is less than or equal to 100) in the near and far fields.

Baumeister, Kenneth J.↗
NASA NTRS2011

Carbon Nanofiber-Based, High-Frequency, High-Q, Miniaturized Mechanical Resonators

High Q resonators are a critical component of stable, low-noise communication systems, radar, and precise timing applications such as atomic clocks. In electronic resonators based on Si integrated circuits, resistive losses increase as a result of the continued reduction in device dimensions, which decreases their Q values. On the other hand, due to the mechanical construct of bulk acoustic wave (BAW) and surface acoustic wave (SAW) resonators, such loss mechanisms are absent, enabling higher Q-values for both BAW and SAW resonators compared to their electronic counterparts. The other advantages of mechanical resonators are their inherently higher radiation tolerance, a factor that makes them attractive for NASA s extreme environment planetary missions, for example to the Jovian environments where the radiation doses are at hostile levels. Despite these advantages, both BAW and SAW resonators suffer from low resonant frequencies and they are also physically large, which precludes their integration into miniaturized electronic systems. Because there is a need to move the resonant frequency of oscillators to the order of gigahertz, new technologies and materials are being investigated that will make performance at those frequencies attainable. By moving to nanoscale structures, in this case vertically oriented, cantilevered carbon nanotubes (CNTs), that have larger aspect ratios (length/thickness) and extremely high elastic moduli, it is possible to overcome the two disadvantages of both bulk acoustic wave (BAW) and surface acoustic wave (SAW) resonators. Nano-electro-mechanical systems (NEMS) that utilize high aspect ratio nanomaterials exhibiting high elastic moduli (e.g., carbon-based nanomaterials) benefit from high Qs, operate at high frequency, and have small force constants that translate to high responsivity that results in improved sensitivity, lower power consumption, and im - proved tunablity. NEMS resonators have recently been demonstrated using topdown, lithographically fabricated ap - proaches to form cantilever or bridgetype structures. Top-down approaches, however, rely on complicated and expensive e-beam lithography, and often require a release mechanism. Reso - nance effects in structures synthesized using bottom-up approaches have also recently been reported based on carbon nanotubes, but such approaches have relied on a planar two-dimensional (2D) geometry. In this innovation, vertically aligned tubes synthesized using a bottom- up approach have been considered, where the vertical orientation of the tubes has the potential to increase integration density even further. The simulation of a vertically oriented, cantilevered carbon nanotube was performed using COMSOL Multi - physics, a finite element simulation package. All simulations were performed in a 2D geometry that provided consistent results and minimized computational complexity. The simulations assumed a vertically oriented, cantilevered nanotube of uniform density (1.5 g/cu cm). An elastic modulus was assumed to be 600 GPa, relative permittivity of the nanotube was assumed to be 5.0, and Poisson s ratio was assumed to be 0.2. It should be noted that the relative permittivity and Poisson s ratio for the nanotubes of interest are not known accurately. However, as in previous simulations, the relative permittivity and Poisson s ratios were treated as weak variables in the simulation, and no significant changes were recognized when these variables were varied.

Kaul, Anupama B.↗
NASA NTRS2020

Amino Acid Encoding for Deep Learning Applications

Background: The number of applications of deep learning algorithms in bioinformatics is increasing as they usually achieve superior performance over classical approaches, especially, when bigger training datasets are available. In deep learning applications, discrete data, e.g. words or n-grams in language, or amino acids or nucleotides in bioinformatics, are generally represented as a continuous vector through an embedding matrix. Recently, learning this embedding matrix directly from the data as part of the continuous iteration of the model to optimize the target prediction – a process called ‘end-to-end learning’ – has led to state-of-the-art results in many fields. Although usage of embeddings is well described in the bioinformatics literature, the potential of end-to-end learning for single amino acids, as compared to more classical manually-curated encoding strategies, has not been systematically addressed. To this end, we compared classical encoding matrices, namely one-hot, VHSE8 and BLOSUM62, to end-to-end learning of amino acid embeddings for two different prediction tasks using three widely used architectures, namely recurrent neural networks (RNN), convolutional neural networks (CNN), and the hybrid CNN-RNN. Results: By using different deep learning architectures, we show that end-to-end learning is on par with classical encodings for embeddings of the same dimension even when limited training data is available, and might allow for a reduction in the embedding dimension without performance loss, which is critical when deploying the models to devices with limited computational capacities. We found that the embedding dimension is a major factor in controlling the model performance. Surprisingly, we observed that deep learning models are capable of learning from random vectors of appropriate dimension. Conclusion: Our study shows that end-to-end learning is a flexible and powerful method for amino acid encoding. Further, due to the flexibility of deep learning systems, amino acid encoding schemes should be benchmarked against random vectors of the same dimension to disentangle the information content provided by the encoding scheme from the distinguishability effect provided by the scheme.

Deep-learning↗
NASA NTRS2020

Amino Acid Encoding for Deep Learning Applications

Background: The number of applications of deep learning algorithms in bioinformatics is increasing as they usually achieve superior performance over classical approaches, especially, when bigger training datasets are available. In deep learning applications, discrete data, e.g. words or n-grams in language, or amino acids or nucleotides in bioinformatics, are generally represented as a continuous vector through an embedding matrix. Recently, learning this embedding matrix directly from the data as part of the continuous iteration of the model to optimize the target prediction – a process called ‘end-to-end learning’ – has led to state-ofthe-art results in many fields. Although usage of embeddings is well described in the bioinformatics literature, the potential of end-to-end learning for single amino acids, as compared to more classical manually-curated encoding strategies, has not been systematically addressed. To this end, we compared classical encoding matrices, namely one-hot, VHSE8 and BLOSUM62, to end-to-end learning of amino acid embeddings for two different prediction tasks using three widely used architectures, namely recurrent neural networks (RNN), convolutional neural networks (CNN), and the hybrid CNN-RNN. Results: By using different deep learning architectures, we show that end-to-end learning is on par with classical encodings for embeddings of the same dimension even when limited training data is available, and might allow for a reduction in the embedding dimension without performance loss, which is critical when deploying the models to devices with limited computational capacities. We found that the embedding dimension is a major factor in controlling the model performance. Surprisingly, we observed that deep learning models are capable of learning from random vectors of appropriate dimension. Conclusion: Our study shows that end-to-end learning is a flexible and powerful method for amino acid encoding. Further, due to the flexibility of deep learning systems, amino acid encoding schemes should be benchmarked against random vectors of the same dimension to disentangle the information content provided by the encoding scheme from the distinguishability effect provided by the scheme.

Hesham ElAbd↗
NASA NTRS1999

Reduction of Large Dynamical Systems by Minimization of Evolution Rate

Reduction of a large system of equations to a lower-dimensional system of similar dynamics is investigated. For dynamical systems with disparate timescales, a criterion for determining redundant dimensions and a general reduction method based on the minimization of evolution rate are proposed.

Girimaji, Sharath S.↗
NASA NTRS1992

A comparison between backscattering models for rough surfaces

The ease of applicability of three scattering models is examined. This is accomplished by considering the time taken to numerically evaluate these models and by comparing their predictions as a function of surface roughness, frequency, incident angle and polarization with the moment-method solution in two dimensions. The complexity of the analytic models in three dimensions and their analytic reduction to high- and low-frequency regions are also compared. The selected models are an integral equation model (IEM), a full wave model (FWM), and the phase perturbation model (PPM). It is noted that in three dimensions, the full-wave model requires an evaluation of a tenfold integral and the phase perturbation model requires a fourfold and a twofold integral, whereas the integral equation model is an algebraic equation in like polarization under single scattering conditions. It is found that both the IEM and PPM agree with the moment-method solution from low to high frequencies numerically.

Chen, K. S.↗
NASA NTRS1988

On the collapse and violent relaxation of protoglobular clusters

During the formation of stellar systems such as globular clusters, low-mass subcondensations which eventually form stars must retain a geometric size throughout the collapse process that is small compared to the characteristic distance separating them. If the local velocity dispersion of the subcondensations is small, the overall dimension of the system can decrease substantially before reaching a dynamical equilibrium state. The maximum collapse factor is deduced by examining the growth of the velocity dispersion and the spread in arrival times at the origin caused by local and global fluctuations. It is shown, analytically as well as in a series of N-body simulations, that the maximum reduction in the characteristic dimension of a system of N fragments with an initial homogeneous distribution subject to N exp 1/2 fluctuations is proportional to N exp 1/3. Direct physical collisions between low-mass subcondensations are therefore unlikely to occur in protoglobular clusters. The results are discussed in the context of fragmentation and violent relaxation.

Aarseth, S. J.↗
NASA NTRS1995

A Reduced Dimension Static, Linearized Kalman Filter and Smoother

An approximate Kalman filter and smoother, based on approximations of the state estimation error covariance matrix, is described. Approximations include a reduction of the effective state dimension, use of a static asymptotic error limit, and a time-invariant linearization of the dynamic model for error integration. The approximations lead to dramatic computational savings in applying estimation theory to large complex systems. Examples of use come from TOPEX/POSEIDON.

Kalman filter smoother modeling ocean simulation↗
NASA NTRS1986

Calibrating an altimeter - How many tide gauges is enough?

The dominant contributor to the random error of an altimetric satellite system is the long wavelength uncertainty in the orbital radius. It is shown that calibration by a comparatively modest tide gauge system can drastically reduce the overall error in global estimates of large-scale oceanic variability. The procedure used is a form of optimal estimation. Absolute (time average) altimetric calibration is much more difficult because it requires absolute calibration of the tide gauge positions (in three dimensions) but the error reduction process would be the same.

Wunsch, Carl↗