SEARCH · Search NASA
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.
The Massively Parallel Processor (MPP) is an ideal machine for computer experiments with simulated neural nets as well as more general cellular automata. Experiments using the MPP with a formal model neural network are described. The results on problem mapping and computational efficiency apply equally well to the neural nets of Hopfield, Hinton et al., and Geman and Geman.
We use neural nets to construct nonlinear models to forecast the AL index given solar wind and interplanetary magnetic field (IMF) data. We follow two approaches: (1) the state space reconstruction approach, which is a nonlinear generalization of autoregressive-moving average models (ARMA) and (2) the nonlinear filter approach, which reduces to a moving average model (MA) in the linear limit. The database used here is that of Bargatze et al. (1985).
Networks of Boolean programmable logic modules are presented as one purely digital class of artificial neural nets. The approach contrasts with the continuous analog framework usually suggested. Programmable logic networks are capable of handling many neural-net applications. They avoid some of the limitations of threshold logic networks and present distinct opportunities. The network nodes are called dynamically programmable logic modules. They can be implemented with digitally controlled demultiplexers. Each node performs a Boolean function of its inputs which can be dynamically assigned. The overall network is therefore a combinational circuit and its outputs are Boolean global functions of the network's input variables. The approach offers definite advantages for VLSI implementation, namely, a regular architecture with limited connectivity, simplicity of the control machinery, natural modularity, and the support of a mature technology.
Document analysis is one of the main applications of machine vision today and offers great opportunities for neural net circuits. Despite more and more data processing with computers, the number of paper documents is still increasing rapidly. A fast translation of data from paper into electronic format is needed almost everywhere, and when done manually, this is a time consuming process. Markets range from small scanners for personal use to high-volume document analysis systems, such as address readers for the postal service or check processing systems for banks. A major concern with present systems is the accuracy of the automatic interpretation. Today's algorithms fail miserably when noise is present, when print quality is poor, or when the layout is complex. A common approach to circumvent these problems is to restrict the variations of the documents handled by a system. In our laboratory, we had the best luck with circuits implementing basic functions, such as convolutions, that can be used in many different algorithms. To illustrate the flexibility of this approach, three applications of the NET32K circuit are described in this short viewgraph presentation: locating address blocks, cleaning document images by removing noise, and locating areas of interest in personal checks to improve image compression. Several of the ideas realized in this circuit that were inspired by neural nets, such as analog computation with a low resolution, resulted in a chip that is well suited for real-world document analysis applications and that compares favorably with alternative, 'conventional' circuits.
The implementation of an earlier introduced neural net model for pattern classification is considered. Data flow principles are employed in the development of a machine that efficiently implements the model and can be useful for real time classification tasks. Further enhancement with optical computing structures is also considered.
This paper describes a feed-forward neural net approach for detection of abnormal system behavior based upon sensor data analyses. A new dynamical invariant representing structural parameters of the system is introduced in such a way that any structural abnormalities in the system behavior are detected from the corresponding changes to the invariant.
A calibration process is presented that uses optical measurements alone to calibrate a neural-net based NDE method. The method itself detects small changes in the vibration mode shapes of structures. The optics-only calibration process confirms previous work that the sensitivity to vibration-amplitude changes can be as small as 10 nanometers. A more practical value in an NDE service laboratory is shown to be 50 nanometers. Both model-generated and experimental calibrations are demonstrated using two implementations of the calibration technique. The implementations are based on previously published demonstrations of the NDE method and an alternative calibration procedure that depends on comparing neural-net and point sensor measurements. The optics-only calibration method, unlike the alternative method, does not require modifications of the structure being tested or the creation of calibration objects. The calibration process can be used to test improvements in the NDE process and to develop a vibration-mode-independence of damagedetection sensitivity. The calibration effort was intended to support NASA s objective to promote safety in the operations of ground test facilities or aviation safety, in general, by allowing the detection of the gradual onset of structural changes and damage.
The biological motivation for Artificial Neural Net developments is briefly discussed, and the most popular paradigm, the feedforward supervised learning net with error back propagation training algorithm, is introduced. Possible approaches for utilization in structural optimization is illustrated through simple examples. Other currently ongoing developments for application in structural mechanics are also mentioned.
The National Aeronautics and Space Administration (NASA) at the Dryden Flight Research Center (DFRC) has been conducting flight-test research using an F-15 aircraft (figure 1). This aircraft has been specially modified to interface a neural net (NN) controller as part of a single-string Airborne Research Test System (ARTS) computer with the existing quad-redundant flight control system (FCC) shown in figure 2. The NN commands are passed to FCC channels 2 and 4 and are cross channel data linked (CCDL) to the other computers as shown. Numerous types of fault-detection monitors exist in the FCC when the NN mode is engaged; these monitors would cause an automatic disengagement of the NN in the event of a triggering fault. Unfortunately, these monitors still may not prevent a possible NN hard-over command from coming through to the control laws. Therefore, an additional and unique safety monitor was designed for a single-string source that allows authority at maximum actuator rates but protects the pilot and structural loads against excessive g-limits in the case of a NN hard-over command input. This additional monitor resides in the FCCs and is executed before the control laws are computed. This presentation describes a floating limiter (FL) concept1 that was developed and successfully test-flown for this program (figure 3). The FL computes the rate of change of the NN commands that are input to the FCC from the ARTS. A window is created with upper and lower boundaries, which is constantly floating and trying to stay centered as the NN command rates are changing. The limiter works by only allowing the window to move at a much slower rate than those of the NN commands. Anywhere within the window, however, full rates are allowed. If a rate persists in one direction, it will eventually hit the boundary and be rate-limited to the floating limiter rate. When this happens, a persistent counter begins and after a limit is reached, a NN disengage command is generated. The tunable metrics for the FL are (1) window size, (2) drift rate, and (3) persistence counter. Ultimate range limits are also included in case the NN command should drift slowly to a limit value that would cause the FL to be defeated. The FL has proven to work as intended. Both high-g transients and excessive structural loads are controlled with NN hard-over commands. This presentation discusses the FL design features and presents test cases. Simulation runs are included to illustrate the dramatic improvement made to the control of NN hard-over effects. A mission control room display from a flight playback is presented to illustrate the neural net fault display representation. The FL is very adaptable to various requirements and is independent of flight condition. It should be considered as a cost-effective safety monitor to control single-string inputs in general.
The San Francisco Bay Delta is a large hydrodynamic complex that incorporates the Sacramento and San Joaquin Estuaries, the Burman Marsh, and the San Francisco Bay proper. Competition exists for the use of this extensive water system both from the fisheries industry, the agricultural industry, and from the marine and estuarine animal species within the Delta. As tidal fluctuations occur, more saline water pushes upstream allowing fish to migrate beyond the Burman Marsh for breeding and habitat occupation. However, the agriculture industry does not want extensive salinity intrusion to impact water quality for human and plant consumption. The balance is regulated by pumping stations located alone the estuaries and reservoirs whereby flushing of fresh water keeps the saline intrusion at bay. The pumping schedule is driven by data collected at various locations within the Bay Delta and by numerical models that predict the salinity intrusion as part of a larger model of the system. The Interagency Ecological Program (IEP) for the San Francisco Bay/Sacramento-San Joaquin Estuary collects, monitors, and archives the data, and the Department of Water Resources provides a numerical model simulation (DSM2) from which predictions are made that drive the pumping schedule. A problem with this procedure is that the numerical simulation takes roughly 16 hours to complete a C:~ prediction. We have created a neural net, optimized with a genetic algorithm, that takes as input the archived data from multiple stations and predicts stage, salinity, and flow at the Carquinez Straits (at the downstream end of the Burman Marsh). This model seems to be robust in its predictions and operates much faster than the current numerical DSM2 model. Because the system is strongly tidal driven, we used both Principal Component Analysis and Fast Fourier Transforms to discover dominant features within the IEP data. We then filtered out the dominant tidal forcing to discover non-primary tidal effects, and used this to enhance the neural network by mapping input-output relationships in a more efficient manner. Furthermore, the neural network implicitly incorporates both the hydrodynamic and water quality models into a single predictive system. Although our model has not yet been enhanced to demonstrate improve pumping schedules, it has the possibility to support better decision-making procedures that may then be implemented by State agencies if desired. Our intention is now to use this model in the smaller Elkhorn Slough complex near Monterey Bay where no such hydrodynamic model currently exists. At the Elkhorn Slough, we are fusing the neural net model of tidally-driven flow with in situ flow data and airborne and satellite remote sensation data. These further constrain the behavior of the model in predicting the longer-term health and future of this vital estuary.
The space vehicle guidance problem is formulated using a neural network approach, and the appropriate neural net architecture for modeling optimum guidance trajectories is investigated. In particular, an investigation is made of the incorporation of prior knowledge about the characteristics of the optimal guidance solution into the neural network architecture. The online classification performance of the developed network is demonstrated using a synthesized network trained with a database of optimum guidance trajectories. Such a neural-network-based guidance approach can readily adapt to environment uncertainties such as those encountered by an AOTV during atmospheric maneuvers.
The capacity of classical neurocomputers is limited by the number of classical degrees of freedom which is roughly proportional to the size of the computer. By Contrast, a Hypothetical quantum neurocomputer can implement an exponentially large number of the degrees of freedom within the same size. In this paper an attempt is made to reconcile linear reversible structure of quantum evolution with nonlinear irreversible dynamics for neural nets.
To facilitate and solve the prediction problem, awareness of the current health state of the system is key, since it is necessary to perform condition-based predictions. To accurately predict the future state of any system, it is required to possess knowledge of its current health state and future operational conditions. Development in data-driven algorithms in regression of complex nonlinear functions and classification tasks have generated a growing interest in artificial intelligence for industrial applications. Complex multi-physics models as well as digital twins, once purely built on physics and corresponding simplified lumped parameter iterations, can now benefit from machine learning algorithms to mitigate the lack of understanding of some complex behavior. The research work presents application of physics-informed neural nets application to a representative electric powertrain for unmanned aerial vehicles. The model is composed of physics-derived and empirical equations, integrated with connected networks that are strategically placed within the model to substitute equations that are subject to large uncertainty. Polynomial fit driven by heuristics or empirical observations can be substituted by more flexible networks that can minimize the error between model predictions and observations without being restricted to a predefined functional form. This modeling strategy allows training of networks deep inside the model and unknown parameters in a single learning stage.
Artificial neural networks have been used for a number of years to process holography-generated characteristic patterns of vibrating structures. This technology depends critically on the selection and the conditioning of the training sets. A scaling operation called folding is discussed for conditioning training sets optimally for training feed-forward neural networks to process characteristic fringe patterns. Folding allows feed-forward nets to be trained easily to detect damage-induced vibration-displacement-distribution changes as small as 10 nm. A specific application to aerospace of neural-net processing of characteristic patterns is presented to motivate the conditioning and optimization effort.
The problem of determining the composition of an unknown input mixture from its measured spectrum, given the spectra of a number of elements, is studied. The Hopfield minimization procedure was used to express the determination of the compositions as a problem suitable for solution by neural nets. A mathematical description of the problem was developed and used as a basis for a neural network solution and an optical implementation.
Neural networks have attracted much interest recently, and using parallel architectures to simulate neural networks is a natural and necessary application. The SIMD model of parallel computation is chosen, because systems of this type can be built with large numbers of processing elements. However, such systems are not naturally suited to generalized communication. A method is proposed that allows an implementation of neural network connections on massively parallel SIMD architectures. The key to this system is an algorithm permitting the formation of arbitrary connections between the neurons. A feature is the ability to add new connections quickly. It also has error recovery ability and is robust over a variety of network topologies. Simulations of the general connection system, and its implementation on the Connection Machine, indicate that the time and space requirements are proportional to the product of the average number of connections per neuron and the diameter of the interconnection network.
Neural networks have attracted much interest recently, and using parallel architectures to simulate neural networks is a natural and necessary application. The SIMD model of parallel computation is chosen, because systems of this type can be built with large numbers of processing elements. However, such systems are not naturally suited to generalized elements. A method is proposed that allows an implementation of neural network connections on massively parallel SIMD architectures. The key to this system is an algorithm permitting the formation of arbitrary connections between the neurons. A feature is the ability to add new connections quickly. It also has error recovery ability and is robust over a variety of network topologies. Simulations of the general connection system, and its implementation on the Connection Machine, indicate that the time and space requirements are proportional to the product of the average number of connections per neuron and the diameter of the interconnection network.
A system for determining the position, orientation and motion of a satellite with respect to a robotic spacecraft using video data is advanced. This system utilizes two levels of pose and motion estimation: an initial system which provides coarse estimates of pose and motion, and a second system which uses the coarse estimates and further processing to provide finer pose and motion estimates. The present paper emphasizes the initial coarse pose and motion estimation sybsystem. This subsystem utilizes novelty detection and filtering for locating novel parts and a neural net tracker to track these parts over time. Results of using this system on a sequence of images of a spin stabilized satellite are presented.