Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “reweighting”

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.

9 records
All sourcesNASA NTRSDOE OSTI
NASA NTRS1993

Non-oscillatory and non-diffusive solution of convection problems by the iteratively reweighted least-squares finite element method

A comparative description is presented for the least-squares FEM (LSFEM) for 2D steady-state pure convection problems. In addition to exhibiting better control of the streamline derivative than the streamline upwinding Petrov-Galerkin method, numerical convergence rates are obtained which show the LSFEM to be virtually optimal. The LSFEM is used as a framework for an iteratively reweighted LSFEM yielding nonoscillatory and nondiffusive solutions for problems with contact discontinuities; this method is shown to convect contact discontinuities without error when using triangular and bilinear elements.

Jiang, Bo-Nan↗
NASA NTRS1991

The L sub 1 finite element method for pure convection problems

The least squares (L sub 2) finite element method is introduced for 2-D steady state pure convection problems with smooth solutions. It is proven that the L sub 2 method has the same stability estimate as the original equation, i.e., the L sub 2 method has better control of the streamline derivative. Numerical convergence rates are given to show that the L sub 2 method is almost optimal. This L sub 2 method was then used as a framework to develop an iteratively reweighted L sub 2 finite element method to obtain a least absolute residual (L sub 1) solution for problems with discontinuous solutions. This L sub 1 finite element method produces a nonoscillatory, nondiffusive and highly accurate numerical solution that has a sharp discontinuity in one element on both coarse and fine meshes. A robust reweighting strategy was also devised to obtain the L sub 1 solution in a few iterations. A number of examples solved by using triangle and bilinear elements are presented.

Jiang, Bo-Nan↗
NASA NTRS1991

The L sub 1 finite element method for pure convection problems

The least squares (L sub 2) finite element method is introduced for 2-D steady state pure convection problems with smooth solutions. It is proven that the L sub 2 method has the same stability estimate as the original equation, i.e., the L sub 2 method has better control of the streamline derivative. Numerical convergence rates are given to show that the L sub 2 method is almost optimal. This L sub 2 method was then used as a framework to develop an iteratively reweighted L sub 2 finite element method to obtain a least absolute residual (L sub 1) solution for problems with discontinuous solutions. This L sub 1 finite element method produces a nonoscillatory, nondiffusive and highly accurate numerical solution that has a sharp discontinuity in one element on both coarse and fine meshes. A robust reweighting strategy was also devised to obtain the L sub 1 solution in a few iterations. A number of examples solved by using triangle and bilinear elements are presented.

Jiang, Bo-Nan↗
NASA NTRS2002

Sensorimotor integration in human postural control

It is generally accepted that human bipedal upright stance is achieved by feedback mechanisms that generate an appropriate corrective torque based on body-sway motion detected primarily by visual, vestibular, and proprioceptive sensory systems. Because orientation information from the various senses is not always available (eyes closed) or accurate (compliant support surface), the postural control system must somehow adjust to maintain stance in a wide variety of environmental conditions. This is the sensorimotor integration problem that we investigated by evoking anterior-posterior (AP) body sway using pseudorandom rotation of the visual surround and/or support surface (amplitudes 0.5-8 degrees ) in both normal subjects and subjects with severe bilateral vestibular loss (VL). AP rotation of body center-of-mass (COM) was measured in response to six conditions offering different combinations of available sensory information. Stimulus-response data were analyzed using spectral analysis to compute transfer functions and coherence functions over a frequency range from 0.017 to 2.23 Hz. Stimulus-response data were quite linear for any given condition and amplitude. However, overall behavior in normal subjects was nonlinear because gain decreased and phase functions sometimes changed with increasing stimulus amplitude. "Sensory channel reweighting" could account for this nonlinear behavior with subjects showing increasing reliance on vestibular cues as stimulus amplitudes increased. VL subjects could not perform this reweighting, and their stimulus-response behavior remained quite linear. Transfer function curve fits based on a simple feedback control model provided estimates of postural stiffness, damping, and feedback time delay. There were only small changes in these parameters with increasing visual stimulus amplitude. However, stiffness increased as much as 60% with increasing support surface amplitude. To maintain postural stability and avoid resonant behavior, an increase in stiffness should be accompanied by a corresponding increase in damping. Increased damping was achieved primarily by decreasing the apparent time delay of feedback control rather than by changing the damping coefficient (i.e., corrective torque related to body-sway velocity). In normal subjects, stiffness and damping were highly correlated with body mass and moment of inertia, with stiffness always about 1/3 larger than necessary to resist the destabilizing torque due to gravity. The stiffness parameter in some VL subjects was larger compared with normal subjects, suggesting that they may use increased stiffness to help compensate for their loss. Overall results show that the simple act of standing quietly depends on a remarkably complex sensorimotor control system.

NASA Discipline Neuroscience↗
NASA NTRS1992

Multivariable frequency domain identification via 2-norm minimization

The author develops a computational approach to multivariable frequency domain identification, based on 2-norm minimization. In particular, a Gauss-Newton (GN) iteration is developed to minimize the 2-norm of the error between frequency domain data and a matrix fraction transfer function estimate. To improve the global performance of the optimization algorithm, the GN iteration is initialized using the solution to a particular sequentially reweighted least squares problem, denoted as the SK iteration. The least squares problems which arise from both the SK and GN iterations are shown to involve sparse matrices with identical block structure. A sparse matrix QR factorization method is developed to exploit the special block structure, and to efficiently compute the least squares solution. A numerical example involving the identification of a multiple-input multiple-output (MIMO) plant having 286 unknown parameters is given to illustrate the effectiveness of the algorithm.

Bayard, David S.↗
NASA NTRS1999

Yaw and pitch visual-vestibular interaction in weightlessness

Both yaw and pitch visual-vestibular interactions at two separate frequencies of chair rotation (0.2 and 0.8 Hz) in combination with a single velocity of optokinetic stimulus (36 degrees/s) were used to investigate the effects of sustained weightlessness on neural strategies adopted by astronaut subjects to cope with the stimulus rearrangement of spaceflight. Pitch and yaw oscillation in darkness at 0.2 and 0.8 Hz without optokinetic stimulation, and constant velocity linear optokinetic stimulation at 18, 36, and 54 degrees/s presented relative to the head with the subject stationary, were used as controls for the visual-vestibular interactions. The results following 8 days of space flight showed no significant changes in: (1) either the horizontal and vertical vestibulo-ocular reflex (VOR) gain, phase, or bias; (2) the yaw visual-vestibular response (VVR); or (3) the horizontal or vertical optokinetic (OKN) slow phase velocity (SPV). However, significant changes were observed: (1) when during pitch VVR at 0.2 Hz late inflight, the contribution of the optokinetic input to the combined oculomotor response was smaller than during the stationary OKN SPV measurements, followed by an increased contribution during the immediate postflight testing; and (2) when during pitch VVR at 0.8 Hz, the component of the combined oculomotor response due to the underlying vertical VOR was more efficiently suppressed early inflight and less suppressed immediately postflight compared with preflight observations. The larger OKN response during pitch VVR at 0.2 Hz and the better suppression of VOR during pitch VVR at 0.8 Hz postflight are presumably due to the increased role of vision early inflight and immediately after spaceflight, as previously observed in various studies. These results suggest that the subjects adopted a neural strategy to structure their spatial orientation in weightlessness by reweighting visual, otolith, and perhaps tactile/somatic signals.

NASA Center JSC↗
NASA NTRS2011

Feature Acquisition with Imbalanced Training Data

This work considers cost-sensitive feature acquisition that attempts to classify a candidate datapoint from incomplete information. In this task, an agent acquires features of the datapoint using one or more costly diagnostic tests, and eventually ascribes a classification label. A cost function describes both the penalties for feature acquisition, as well as misclassification errors. A common solution is a Cost Sensitive Decision Tree (CSDT), a branching sequence of tests with features acquired at interior decision points and class assignment at the leaves. CSDT's can incorporate a wide range of diagnostic tests and can reflect arbitrary cost structures. They are particularly useful for online applications due to their low computational overhead. In this innovation, CSDT's are applied to cost-sensitive feature acquisition where the goal is to recognize very rare or unique phenomena in real time. Example applications from this domain include four areas. In stream processing, one seeks unique events in a real time data stream that is too large to store. In fault protection, a system must adapt quickly to react to anticipated errors by triggering repair activities or follow- up diagnostics. With real-time sensor networks, one seeks to classify unique, new events as they occur. With observational sciences, a new generation of instrumentation seeks unique events through online analysis of large observational datasets. This work presents a solution based on transfer learning principles that permits principled CSDT learning while exploiting any prior knowledge of the designer to correct both between-class and withinclass imbalance. Training examples are adaptively reweighted based on a decomposition of the data attributes. The result is a new, nonparametric representation that matches the anticipated attribute distribution for the target events.

Thompson, David R.↗
NASA NTRS2018

Global High Resolution Crustal Magnetic Field at the Surface of the Moon from Low-Altitude Lunar Prospector Magnetic Gradient Data

We derived new vector gradients based models of crustal magnetic field at the lunar surface with data from the Lunar Prospector (LP) satellite using two model parameterization approaches: a global set of 35820 1° spaced (~30 km) equal area monopoles at 20 km below the surface (O’Brien and Parker, 1994; Olsen et al., 2017) and combined results of subsets of 100000 0.66° spaced monopoles at the same depth. We use the scheme of iteratively reweighted least-squares inversion to compute the initial model. Then the amplitudes of these monopoles are determined by minimizing the misfit to the components together with the global average of |Br| at the ellipsoid surface (i.e. applying a L1 model regularization of Br). In previous approaches using vector fields for modeling, we found that external field contamination leads to spurious anomalies in the downward continued field models even with stringent data selection criteria and ad-hoc noise removal techniques (e.g., satellite’s position in the Moon’s wake w.r.t. the solar wind and in the Earth’s magnetotail, internal/external dipoles fields removal, low-order polynomial removal, joint equivalent source cross-validation technique and visually removing remaining anomalous segments). On the other hand, with the use of gradients-only data (along-track first finite differences), we were able to completely bypass the ad-hoc techniques. Similar processing of Kaguya magnetic data, which have only higher altitude coverage over the most of the Moon except in the region of the South Pole–Aitken (SPA) basin, completely misses some of the anomalies seen in the Lunar Prospector data. The combined Lunar Prospector and Kaguya gradient-based models also severely degrade the derivation of the anomaly fields in many regions. Euler analysis of isolated anomaly features from the Reiner Gamma swirl suggests top depths of about 0.3 to 1.5 km and center depths of 10-14 km; in the region between Stein and Vallier craters north of the SPA basin our analysis suggests top depths of around 1.5 km and center depths of 13-15 km. With the spectral depth determination techniques, the SPA basin region yields depths to the base of magnetization ranging between 15 and 40 km. Three-dimensional modeling and the bulk magnetization determinations of the sources constrained by the Euler and spectral methods is underway.

Gradient↗