Search NASA⌕ Search

SEARCH · Search NASA

Results for “feedforward”

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 415 records · Page 23

Neutral Networks for Subpixel Classification of Multispectral Images

In this work, the implementation and use of AVHRR (Advanced Very High Resolution Radiometer) images for subpixel analysis is studied. The work consists of two parts, the first is making training data, the second is using the training data to design a neural net subpixel analyzer. Most work on subpixel analysis has been done with images with more spectral bands. AVHRR images were chosen because of their easy acquisition, and because the five spectral bands allow investigation into the development of training data. The first step in subpixel analysis is the development of training data. This consists of image to be classified, and the classification of each pixel. In order to do the classification, a high spatial resolution image is typically needed in order to manually create a classified image. It is difficult to have both the image of interest, and a high spatial resolution image of the same area taken at the same time. Thus it was studied whether a subsampled image taken from the image of interest could serve as the training data. Statistical work has been done showing the unusefulness of this approach. In the second part of the analysis, a feedforward neural net was trained and used to classify the AVHRR images. Results of these tests, comparing the neural net with typical statistical based schemes are shown.

Figueroa, Ricardo R.↗

Preliminary Work for Examining the Scalability of Reinforcement Learning

Researchers began studying automated agents that learn to perform multiple-step tasks early in the history of artificial intelligence (Samuel, 1963; Samuel, 1967; Waterman, 1970; Fikes, Hart & Nilsonn, 1972). Multiple-step tasks are tasks that can only be solved via a sequence of decisions, such as control problems, robotics problems, classic problem-solving, and game-playing. The objective of agents attempting to learn such tasks is to use the resources they have available in order to become more proficient at the tasks. In particular, each agent attempts to develop a good policy, a mapping from states to actions, that allows it to select actions that optimize a measure of its performance on the task; for example, reducing the number of steps necessary to complete the task successfully. Our study focuses on reinforcement learning, a set of learning techniques where the learner performs trial-and-error experiments in the task and adapts its policy based on the outcome of those experiments. Much of the work in reinforcement learning has focused on a particular, simple representation, where every problem state is represented explicitly in a table, and associated with each state are the actions that can be chosen in that state. A major advantage of this table lookup representation is that one can prove that certain reinforcement learning techniques will develop an optimal policy for the current task. The drawback is that the representation limits the application of reinforcement learning to multiple-step tasks with relatively small state-spaces. There has been a little theoretical work that proves that convergence to optimal solutions can be obtained when using generalization structures, but the structures are quite simple. The theory says little about complex structures, such as multi-layer, feedforward artificial neural networks (Rumelhart & McClelland, 1986), but empirical results indicate that the use of reinforcement learning with such structures is promising. These empirical results make no theoretical claims, nor compare the policies produced to optimal policies. A goal of our work is to be able to make the comparison between an optimal policy and one stored in an artificial neural network. A difficulty of performing such a study is finding a multiple-step task that is small enough that one can find an optimal policy using table lookup, yet large enough that, for practical purposes, an artificial neural network is really required. We have identified a limited form of the game OTHELLO as satisfying these requirements. The work we report here is in the very preliminary stages of research, but this paper provides background for the problem being studied and a description of our initial approach to examining the problem. In the remainder of this paper, we first describe reinforcement learning in more detail. Next, we present the game OTHELLO. Finally we argue that a restricted form of the game meets the requirements of our study, and describe our preliminary approach to finding an optimal solution to the problem.

Clouse, Jeff↗

Non-linear Post Processing Image Enhancement

A non-linear filter for image post processing based on the feedforward Neural Network topology is presented. This study was undertaken to investigate the usefulness of "smart" filters in image post processing. The filter has shown to be useful in recovering high frequencies, such as those lost during the JPEG compression-decompression process. The filtered images have a higher signal to noise ratio, and a higher perceived image quality. Simulation studies comparing the proposed filter with the optimum mean square non-linear filter, showing examples of the high frequency recovery, and the statistical properties of the filter are given,

Hunt, Shawn↗

Recent National Transonic Facility Test Process Improvements (Invited)

This paper describes the results of two recent process improvements; drag feed-forward Mach number control and simultaneous force/moment and pressure testing, at the National Transonic Facility. These improvements have reduced the duration and cost of testing. The drag feedforward Mach number control reduces the Mach number settling time by using measured model drag in the Mach number control algorithm. Simultaneous force/moment and pressure testing allows simultaneous collection of force/moment and pressure data without sacrificing data quality thereby reducing the overall testing time. Both improvements can be implemented at any wind tunnel. Additionally the NTF is working to develop and implement continuous pitch as a testing option as an additional method to reduce costs and maintain data quality.

Kilgore, W. A.↗

Gravity Sensors and the Role of 3-D Visualization and Simulation in Biomedical Research

Ross 3.0 software was developed in the Biocomputation Center for semi-automated reconstruction of objects from serial thin sections. Data are captured directly from a transmission electron microscope via a video camera to a graphics workstation where the sections are mosaicked and contours are traced, registered and displayed by semi-automated methods. For the first time, macular type II cells are described completely for their innervation patterns. The purposes are to learn more about the fundamental circuitry of the macula and to demonstrate whether the terminals are altered morphologically by space flight. Current examples, from the medial part of the macula, are from maculas collected in-flight on the Space Life Sciences-2 mission, 4.5 hrs post-flight, and from a ground control. Results show that the typical type 11 cell receives processes from up to six nearby calyces or afferents. Nearly all the processes are elongated; some have bouton-like swellings and numerous vesicles. Multiple (2 to 4) processes from a single calyx to a type II cell are common, and approx. 1/3 of the processes innervate 2 type II cells of a neighboring cluster of 3 cells. About 2% of type II cells resemble type I cells morphologically and are surrounded by demicalyces. Differences in size or shape of the terminals under flight conditions could not be determined because the sample size is still too small; but it is clear that reconstruction methods provide insights into macular circuitry not obtainable by other techniques. The results demonstrate a morphological basis for interactions between adjacent receptive fields, through feedback-feedforward connections, during preprocessing of linear acceleratory information by the maculas. While the methods are currently being tested using vestibular maculas as the model system, it is clear that the technology is applicable to any tissue that can be physically or optically sectioned. ROSS software has already been implemented for reconstructing objects from tissues studied by confocal and by transmitted light microscopy, and research into magnetic resonance imaging-computational tomography combined visualization are underway.

Ross, Muriel D.↗

Estimating Single-Trial Responses in EEG

Accurate characterization of single-trial field potential responses is critical from a number of perspectives. For example, it allows differentiation of an evoked response from ongoing EEG. We previously developed the multiple component Event Related Potential (mcERP) algorithm to improve resolution of the single-trial evoked response. The mcERP model states that multiple components, each specified by a stereotypic waveform varying in latency and amplitude from trial to trial, comprise the evoked response. Application of the mcERP algorithm to simulated data with three independent, synthetic components has shown that the model is capable of separating these components and estimating their variability. Application of the model to single trial, visual evoked potentials recorded simultaneously from all V1 laminae in an awake, fixating macaque yielded local and far-field components. Certain local components estimated by the model were distributed in both granular and supragranular laminae. This suggests a linear coupling between the responses of thalamo-recipient neuronal ensembles and subsequent responses of supragranular neuronal ensembles, as predicted by the feedforward anatomy of V1. Our results indicate that the mcERP algorithm provides a valid estimation of single-trial responses. This will enable analyses that depend on trial-to-trial variations and those that require separation of the evoked response from background EEG rhythms

Shah, A. S.↗

Single-Trial Analysis of V1 Responses Suggests Two Transmission States

Sensory processing in the visual, auditory, and somatosensory systems is often studied by recording electrical activity in response to a stimulus of interest. Typically, multiple trial responses to the stimulus are averaged to isolate the stereotypic response from noise. However, averaging ignores dynamic variability in the neuronal response, which is potentially critical to understanding stimulus-processing schemes. Thus, we developed the multiple component, Event-Related Potential (mcERP) model. This model asserts that multiple components, defined as stereotypic waveforms, comprise the stimulus-evoked response and that these components may vary in amplitude and latency from trial to trial. Application of this model to data recorded simultaneously from all six laminae of V1 in an awake, behaving monkey performing a visual discrimination yielded three components. The first component localized to granular V1, the second was located in supragranular V1, and the final component displayed a multi-laminar distribution. These modeling results, which take into account single-trial response dynamics, illustrated that the initial activation of VI occurs in the granular layer followed by activation in the supragranular layers. This finding is expected because the average response in those layers demonstrates the same progression and because anatomical evidence suggests that the feedforward input in V1 enters the granular layer and progresses to supragranular layers. In addition to these findings, the granular component of the model displayed several interesting trial-to-trial characteristics including (1) a bimodal latency distribution, (2) a latency-related variation in response amplitude, (3) a latency correlation with the supragranular component, and (4) an amplitude and latency association with the multi-laminar component. Direct analyses of the single-trial data were consistent with these model predictions. These findings suggest that V1 has at least 2 transmission states, which may be modulated by various effects such as attention, dynamics in local EEG rhythm, or variation in sensory inputs.

Shah, A. S.↗

Integration of Online Parameter Identification and Neural Network for In-Flight Adaptive Control

An indirect adaptive system has been constructed for robust control of an aircraft with uncertain aerodynamic characteristics. This system consists of a multilayer perceptron pre-trained neural network, online stability and control derivative identification, a dynamic cell structure online learning neural network, and a model following control system based on the stochastic optimal feedforward and feedback technique. The pre-trained neural network and model following control system have been flight-tested, but the online parameter identification and online learning neural network are new additions used for in-flight adaptation of the control system model. A description of the modification and integration of these two stand-alone software packages into the complete system in preparation for initial flight tests is presented. Open-loop results using both simulation and flight data, as well as closed-loop performance of the complete system in a nonlinear, six-degree-of-freedom, flight validated simulation, are analyzed. Results show that this online learning system, in contrast to the nonlearning system, has the ability to adapt to changes in aerodynamic characteristics in a real-time, closed-loop, piloted simulation, resulting in improved flying qualities.

Hageman, Jacob↗

Integration of Online Parameter Identification and Neural Network for In-Flight Adaptive Control

An indirect adaptive system has been constructed for robust control of an aircraft with uncertain aerodynamic characteristics. This system consists of a multilayer perceptron pre-trained neural network, online stability and control derivative identification, a dynamic cell structure online learning neural network, and a model following control system based on the stochastic optimal feedforward and feedback technique. The pre-trained neural network and model following control system have been flight-tested, but the online parameter identification and online learning neural network are new additions used for in-flight adaptation of the control system model. A description of the modification and integration of these two stand-alone software packages into the complete system in preparation for initial flight tests is presented. Open-loop results using both simulation and flight data, as well as closed-loop performance of the complete system in a nonlinear, six-degree-of-freedom, flight validated simulation, are analyzed. Results show that this online learning system, in contrast to the nonlearning system, has the ability to adapt to changes in aerodynamic characteristics in a real-time, closed-loop, piloted simulation, resulting in improved flying qualities.

Hageman, Jacob J.↗

Modeling beta-adrenergic control of cardiac myocyte contractility in silico

The beta-adrenergic signaling pathway regulates cardiac myocyte contractility through a combination of feedforward and feedback mechanisms. We used systems analysis to investigate how the components and topology of this signaling network permit neurohormonal control of excitation-contraction coupling in the rat ventricular myocyte. A kinetic model integrating beta-adrenergic signaling with excitation-contraction coupling was formulated, and each subsystem was validated with independent biochemical and physiological measurements. Model analysis was used to investigate quantitatively the effects of specific molecular perturbations. 3-Fold overexpression of adenylyl cyclase in the model allowed an 85% higher rate of cyclic AMP synthesis than an equivalent overexpression of beta 1-adrenergic receptor, and manipulating the affinity of Gs alpha for adenylyl cyclase was a more potent regulator of cyclic AMP production. The model predicted that less than 40% of adenylyl cyclase molecules may be stimulated under maximal receptor activation, and an experimental protocol is suggested for validating this prediction. The model also predicted that the endogenous heat-stable protein kinase inhibitor may enhance basal cyclic AMP buffering by 68% and increasing the apparent Hill coefficient of protein kinase A activation from 1.0 to 2.0. Finally, phosphorylation of the L-type calcium channel and phospholamban were found sufficient to predict the dominant changes in myocyte contractility, including a 2.6x increase in systolic calcium (inotropy) and a 28% decrease in calcium half-relaxation time (lusitropy). By performing systems analysis, the consequences of molecular perturbations in the beta-adrenergic signaling network may be understood within the context of integrative cellular physiology.

Non-NASA Center↗

Neural Networks for Rapid Design and Analysis

Artificial neural networks have been employed for rapid and efficient dynamics and control analysis of flexible systems. Specifically, feedforward neural networks are designed to approximate nonlinear dynamic components over prescribed input ranges, and are used in simulations as a means to speed up the overall time response analysis process. To capture the recursive nature of dynamic components with artificial neural networks, recurrent networks, which use state feedback with the appropriate number of time delays, as inputs to the networks, are employed. Once properly trained, neural networks can give very good approximations to nonlinear dynamic components, and by their judicious use in simulations, allow the analyst the potential to speed up the analysis process considerably. To illustrate this potential speed up, an existing simulation model of a spacecraft reaction wheel system is executed, first conventionally, and then with an artificial neural network in place.

Sparks, Dean W., Jr.↗

Generation of Higher Order Modes in a Rectangular Duct

Advanced noise control methodologies to reduce sound emission from aircraft engines take advantage of the modal structure of the noise in the duct. This noise is caused by the interaction of rotor wakes with downstream obstructions such as exit guide vanes. Mode synthesis has been accomplished in circular ducts and current active noise control work has made use of this capability to cancel fan noise. The goal of the current effort is to examine the fundamental process of higher order mode propagation through an acoustically treated, curved duct. The duct cross-section is rectangular to permit greater flexibility in representation of a range of duct curvatures. The work presented is the development of a feedforward control system to generate a user-specified modal pattern in the duct. The multiple-error, filtered-x LMS algorithm is used to determine the magnitude and phase of signal input to the loudspeakers to produce a desired modal pattern at a set of error microphones. Implementation issues, including loudspeaker placement and error microphone placement, are discussed. Preliminary results from a 9-3/8 inch by 21 inch duct, using 12 loudspeakers and 24 microphones, are presented. These results demonstrate the ability of the control system to generate a user-specified mode while suppressing undesired modes.

Gerhold, Carl H.↗

Callosally projecting neurons in the macaque monkey V1/V2 border are enriched in nonphosphorylated neurofilament protein

Previous immunohistochemical studies combined with retrograde tracing in macaque monkeys have demonstrated that corticocortical projections can be differentiated by their content of neurofilament protein. The present study analyzed the distribution of nonphosphorylated neurofilament protein in callosally projecting neurons located at the V1/V2 border. All of the retrogradely labeled neurons were located in layer III at the V1/V2 border and at an immediately adjacent zone of area V2. A quantitative analysis showed that the vast majority (almost 95%) of these interhemispheric projection neurons contain neurofilament protein immunoreactivity. This observation differs from data obtained in other sets of callosal connections, including homotypical interhemispheric projections in the prefrontal, temporal, and parietal association cortices, that were found to contain uniformly low proportions of neurofilament protein-immunoreactive neurons. Comparably, highly variable proportions of neurofilament protein-containing neurons have been reported in intrahemispheric corticocortical pathways, including feedforward and feedback visual connections. These results indicate that neurofilament protein is a prominent neurochemical feature that identifies a particular population of interhemispheric projection neurons at the V1/V2 border and suggest that this biochemical attribute may be critical for the function of this subset of callosal neurons.

Non-NASA Center↗

Morphological evidence for local microcircuits in rat vestibular maculae

Previous studies suggested that intramacular, unmyelinated segments of vestibular afferent nerve fibers and their large afferent endings (calyces) on type I hair cells branch. Many of the branches (processes) contain vesicles and are presynaptic to type II hair cells, other processes, intramacular nerve fibers, and calyces. This study used serial section transmission electron microscopy and three-dimensional reconstruction methods to document the origins and distributions of presynaptic processes of afferents in the medial part of the adult rat utricular macula. The ultrastructural research focused on presynaptic processes whose origin and termination could be observed in a single micrograph. Results showed that calyces had 1) vesiculated, spine-like processes that invaginated type I cells and 2) other, elongate processes that ended on type II cells pre- as well as postsynaptically. Intramacular, unmyelinated segments of afferent nerve fibers gave origin to branches that were presynaptic to type II cells, calyces, calyceal processes, and other nerve fibers in the macula. Synapses with type II cells occurred opposite subsynaptic cisternae (C synapses); all other synapses were asymmetric. Vesicles were pleomorphic but were differentially distributed according to process origin. Small, clear-centered vesicles, approximately 40-60 nm in diameter, predominated in processes originating from afferent nerve fibers and basal parts of calyces. Larger vesicles approximately 70-120 nm in diameter having approximately 40-80 nm electron-opaque cores were dominant in processes originating from the necks of calyces. Results are interpreted to indicate the existence of a complex system of intrinsic feedforward (postsynaptic)-feedback (presynaptic) connections in a network of direct and local microcircuits. The morphological findings support the concept that maculae dynamically preprocess linear acceleratory information before its transmission to the central nervous system.

NASA Center ARC↗

Linear and nonlinear ARMA model parameter estimation using an artificial neural network

This paper addresses parametric system identification of linear and nonlinear dynamic systems by analysis of the input and output signals. Specifically, we investigate the relationship between estimation of the system using a feedforward neural network model and estimation of the system by use of linear and nonlinear autoregressive moving-average (ARMA) models. By utilizing a neural network model incorporating a polynomial activation function, we show the equivalence of the artificial neural network to the linear and nonlinear ARMA models. We compare the parameterization of the estimated system using the neural network and ARMA approaches by utilizing data generated by means of computer simulations. Specifically, we show that the parameters of a simulated ARMA system can be obtained from the neural network analysis of the simulated data or by conventional least squares ARMA analysis. The feasibility of applying neural networks with polynomial activation functions to the analysis of experimental data is explored by application to measurements of heart rate (HR) and instantaneous lung volume (ILV) fluctuations.

NASA Discipline Regulatory Physiology↗

Intelligent, Robust Control of Deteriorated Turbofan Engines via Linear Parameter Varying Quadratic Lyapunov Function Design

A method for accommodating engine deterioration via a scheduled Linear Parameter Varying Quadratic Lyapunov Function (LPVQLF)-Based controller is presented. The LPVQLF design methodology provides a means for developing unconditionally stable, robust control of Linear Parameter Varying (LPV) systems. The controller is scheduled on the Engine Deterioration Index, a function of estimated parameters that relate to engine health, and is computed using a multilayer feedforward neural network. Acceptable thrust response and tight control of exhaust gas temperature (EGT) is accomplished by adjusting the performance weights on these parameters for different levels of engine degradation. Nonlinear simulations demonstrate that the controller achieves specified performance objectives while being robust to engine deterioration as well as engine-to-engine variations.

Turso, James A.↗

An optimal controller for an electric ventricular-assist device: theory, implementation, and testing

This paper addresses the development and testing of an optimal position feedback controller for the Penn State electric ventricular-assist device (EVAD). The control law is designed to minimize the expected value of the EVAD's power consumption for a targeted patient population. The closed-loop control law is implemented on an Intel 8096 microprocessor and in vitro test runs show that this controller improves the EVAD's efficiency by 15-21%, when compared with the performance of the currently used feedforward control scheme.

Therapy, Computer-Assisted/standards↗

Model-Trained Neural Networks and Electronic Holography Demonstrated to Detect Damage in Blades

Detect Damage in Blades Electronic holography can show damaged regions in fan blades at 30 frames/sec. The electronic holograms are transformed by finite-element-model-trained artificial neural networks to visualize the damage. The trained neural networks are linked with video and graphics to visualize the bending-induced strain distribution, which is very sensitive to damage. By contrast, it is very difficult to detect damage by viewing the raw, speckled, characteristic fringe patterns. For neural-network visualization of damage, 2 frames or 2 fields are used, rather than the 12 frames normally used to compute the displacement distribution from electronic holograms. At the NASA Lewis Research Center, finite element models are used to compute displacement and strain distributions for the vibration modes of undamaged and cracked blades. A model of electronic time-averaged holography is used to transform the displacement distributions into finite-element-resolution characteristic fringe patterns. Then, a feedforward neural network is trained with the fringe-pattern/strain-pattern pairs, and the neural network, electronic holography, and video are implemented on a workstation. Now that the neural networks have been tested successfully at 30 frames/sec on undamaged and cracked cantilevers, the electronic holography and neural-network processing are being adapted for onsite damage inspection of twisted fan blades and rotormounted blades. Our conclusion is that model-trained neural nets are effective when they are trained with good models whose application is well understood. This work supports the aeromechanical testing portion of the Advanced Subsonic Technology Project.

Decker, Arthur J.↗