SEARCH · Engineering Papers

Results for “Separable neural networks”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 73 records · Page 4

All sources NASA NTRS DOE OSTI

DOE OSTI2025

Emulation of the calculations of final r -process abundance patterns with a neural network

This work explores the construction of a fast emulator for the calculation of the final pattern of nucleosynthesis in the rapid neutron capture process (the r-process). An emulator is built using a feed-forward artificial neural network (ANN). We train the ANN with nuclear data and relative abundance patterns. We take as input the β-decay half-lives and the one-neutron separation energy of the nuclei in the rare-earth region. The output is the final isotopic abundance pattern. In this work, we focus on the nuclear data and abundance patterns in the rare-earth region to reduce the dimension of the input and output space. We show that the ANN can capture the effect of the changes in the nuclear physics inputs on the final r-process abundance pattern in the adopted astrophysical conditions. We employ the deep ensemble method to quantify the prediction uncertainty of the neural network emulator. The emulator achieves a speed-up by a factor of about 20 000 in obtaining a final abundance pattern in the rare-earth region. The emulator may be utilized in statistical analyses such as uncertainty quantification, inverse problems, and sensitivity analysis.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

DOE OSTI2026

Application of machine learning techniques for fast MeV x-ray spectra unfolding from filter stack spectrometer data

Recovery of MeV x-ray spectra from detector signals is difficult because the response matrix inversion is ill-conditioned and current methods are too slow for high-repetition-rate experiments. In this work, we make use of neural networks to unfold MeV x-ray spectra from measurements obtained with a filter stack spectrometer at rates of near 40 Hz. The neural network was trained on synthetic data and tested on both synthetic and experimental data, the latter obtained in two separate experiments performed at the Omega EP laser facility. We show here that this unfolding method has good performance on synthetic data and that it is a promising option for experimental data of up to 40 MeV. The accuracy on experimental data is verified by using a simple forward model to compare against measured values.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DOE OSTI2024

Distributed-Memory Sparse Deep Neural Network Inference Using Global Arrays

Partitioned Global Address Space (PGAS) models exhibit tremendous promise in developing efficient and productive distributed-memory parallel applications. They have been used extensively in scientific computations due to conveniently offering a ``shared-memory''-like model and convenient interfaces that separate communication with synchronization. Traditionally, PGAS communication models have been applied to dense/contiguously distributed data, but most modern applications depict varied levels of sparsity. Existing PGAS models require certain adaptations to support distributed sparse computations, since associated computations often require matrix arithmetic, in addition to data movement. The Global Arrays toolkit from Pacific Northwest National Laboratory (PNNL) is one of the earliest PGAS models to combine one-sided data communication and distributed matrix operations and is still used in the popular NWChem quantum chemistry suite. Recently, we have expanded the Global Arrays toolkit to support common sparse operations, like sparse matrix-dense matrix multiplies (SpMM), sparse matrix-sparse matrix multiplication (SpGEMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM). As it turns out, these operations are the bedrock of sparse Deep Learning (DL); sparse deep neural networks and Graph Neural Networks (GNNs) have gained increasing attention recently in achieving speedups on training and inference with reduced memory footprints. Unlike scientific applications in High Performance Computing (HPC), modern (distributed-memory capable) DL toolkits often rely on non-standardized and closed-source vendor software optimizations, creating challenges in software-hardware co-design at scale. Our goal is to support a variety of distributed-memory sparse matrix operations and helper functions in the newly created Sparse Global Arrays (SGA), such that it is possible to build portable and productive Machine Learning scenarios for algorithm/software and hardware codesign purposes. Contemporary data-parallel schemes for training/inference are undergoing a major overhaul since model replication limits scalability and causes resource inefficiencies. As such, we have adopted tensor parallelism in decomposing the model and inputs, to mitigate memory issues. Current implementation is built on top of MPI and uses CPUs to maximize the portability across the platforms.

Distributed computing, machine learning↗

NASA NTRS1991

Target Detection Using Fractal Geometry

The concepts and theory of fractal geometry were applied to the problem of segmenting a 256 x 256 pixel image so that manmade objects could be extracted from natural backgrounds. The two most important measurements necessary to extract these manmade objects were fractal dimension and lacunarity. Provision was made to pass the manmade portion to a lookup table for subsequent identification. A computer program was written to construct cloud backgrounds of fractal dimensions which were allowed to vary between 2.2 and 2.8. Images of three model space targets were combined with these backgrounds to provide a data set for testing the validity of the approach. Once the data set was constructed, computer programs were written to extract estimates of the fractal dimension and lacunarity on 4 x 4 pixel subsets of the image. It was shown that for clouds of fractal dimension 2.7 or less, appropriate thresholding on fractal dimension and lacunarity yielded a 64 x 64 edge-detected image with all or most of the cloud background removed. These images were enhanced by an erosion and dilation to provide the final image passed to the lookup table. While the ultimate goal was to pass the final image to a neural network for identification, this work shows the applicability of fractal geometry to the problems of image segmentation, edge detection and separating a target of interest from a natural background.

Fuller, J. Joseph↗

NASA NTRS2022

Artificial Neural Network Modeling for Airline Disruption Management

Since the 1970s, most airlines have incorporated computerized support for managing disruptions during flight schedule execution. However, existing platforms for airline disruption management (ADM) employ monolithic system design methods that rely on the creation of specific rules and requirements through explicit optimization routines, before a system that meets the specifications is designed. Thus, current platforms for ADM are unable to readily accommodate additional system complexities resulting from the introduction of new capabilities, such as the introduction of unmanned aerial systems (UAS), operations and infrastructure, to the system. To this end, we use historical data on airline scheduling and operations recovery to develop a system of artificial neural networks (ANNs), which describe a predictive transfer function model (PTFM) for promptly estimating the recovery impact of disruption resolutions at separate phases of flight schedule execution during ADM. Furthermore, we provide a modular approach for assessing and executing the PTFM by employing a parallel ensemble method to develop generative routines that amalgamate the system of ANNs. Our modular approach ensures that current industry standards for tardiness in flight schedule execution during ADM are satisfied, while accurately estimating appropriate time-based performance metrics for the separate phases of flight schedule execution.

Kolawole Ogunsina↗

NASA NTRS2002

Fatigue Crack Measurement in Composite Materials by Ultrasonic Methods

The nondestructive detection of intra-ply microcracking in unlined pressure vessels fabricated from composite materials is critical to ensuring mission success. Microcracking in composite structures due to combined fatigue and cryogenic thermal loading can be very troublesome to detect in-service and when it begins to link through the thickness can cause leakage and failure of the structure. These leaks may lead to loss of pressure/propellant, increased risk of explosion and possible cryo-pumping. The work presented herein develops a method and an instrument to locate and measure intraply fatigue cracking through the thickness of laminated composite material by means of correlation with ultrasonic resonance. Resonant ultrasound spectroscopy provides measurements which are, sensitive to both the microscopic and macroscopic properties of an object. Elastic moduli, acoustic attenuation, and geometry can all be probed. The approach is based on the premise of half-wavelength resonance. The method injects a broadband ultrasonic wave into the test structure using a swept frequency technique. This method provides dramatically increased energy input into the test article, as compared to conventional spike pulsed ultrasonics. This relative energy increase improves the ability to measure finer details in the materials character, such as micro-cracking and porosity. As the micro-crack density increases, more interactions occur with the higher frequency (small wavelength) components of the signal train causing the spectrum to shift toward lower frequencies. Preliminary experiments have verified a measurable effect on the resonance spectrum of the ultrasonic data to detect microcracking. Methods involving self organizing neural networks and other clustering algorithms show that the resonance ultrasound signatures from composites vary with the degree of microcracking and can be separated and identified.

Walker, James L.↗

NASA NTRS2000

Neural Network Based Representation of UH-60A Pilot and Hub Accelerations

Neural network relationships between the full-scale, experimental hub accelerations and the corresponding pilot floor vertical vibration are studied. The present physics-based, quantitative effort represents an initial systematic study on the UH-60A Black Hawk hub accelerations. The NASA/Army UH-60A Airloads Program flight test database was used. A 'maneuver-effect-factor (MEF)', derived using the roll-angle and the pitch-rate, was used. Three neural network based representation-cases were considered. The pilot floor vertical vibration was considered in the first case and the hub accelerations were separately considered in the second case. The third case considered both the hub accelerations and the pilot floor vertical vibration. Neither the advance ratio nor the gross weight alone could be used to predict the pilot floor vertical vibration. However, the advance ratio and the gross weight together could be used to predict the pilot floor vertical vibration over the entire flight envelope. The hub accelerations data were modeled and found to be of very acceptable quality. The hub accelerations alone could not be used to predict the pilot floor vertical vibration. Thus, the hub accelerations alone do not drive the pilot floor vertical vibration. However, the hub accelerations, along with either the advance ratio or the gross weight or both, could be used to satisfactorily predict the pilot floor vertical vibration. The hub accelerations are clearly a factor in determining the pilot floor vertical vibration.

Kottapalli, Sesi↗

NASA NTRS2002

Prediction of Aerodynamic Coefficient using Genetic Algorithm Optimized Neural Network for Sparse Data

Wind tunnels use scale models to characterize aerodynamic coefficients, Wind tunnel testing can be slow and costly due to high personnel overhead and intensive power utilization. Although manual curve fitting can be done, it is highly efficient to use a neural network to define the complex relationship between variables. Numerical simulation of complex vehicles on the wide range of conditions required for flight simulation requires static and dynamic data. Static data at low Mach numbers and angles of attack may be obtained with simpler Euler codes. Static data of stalled vehicles where zones of flow separation are usually present at higher angles of attack require Navier-Stokes simulations which are costly due to the large processing time required to attain convergence. Preliminary dynamic data may be obtained with simpler methods based on correlations and vortex methods; however, accurate prediction of the dynamic coefficients requires complex and costly numerical simulations. A reliable and fast method of predicting complex aerodynamic coefficients for flight simulation I'S presented using a neural network. The training data for the neural network are derived from numerical simulations and wind-tunnel experiments. The aerodynamic coefficients are modeled as functions of the flow characteristics and the control surfaces of the vehicle. The basic coefficients of lift, drag and pitching moment are expressed as functions of angles of attack and Mach number. The modeled and training aerodynamic coefficients show good agreement. This method shows excellent potential for rapid development of aerodynamic models for flight simulation. Genetic Algorithms (GA) are used to optimize a previously built Artificial Neural Network (ANN) that reliably predicts aerodynamic coefficients. Results indicate that the GA provided an efficient method of optimizing the ANN model to predict aerodynamic coefficients. The reliability of the ANN using the GA includes prediction of aerodynamic coefficients to an accuracy of 110% . In our problem, we would like to get an optimized neural network architecture and minimum data set. This has been accomplished within 500 training cycles of a neural network. After removing training pairs (outliers), the GA has produced much better results. The neural network constructed is a feed forward neural network with a back propagation learning mechanism. The main goal has been to free the network design process from constraints of human biases, and to discover better forms of neural network architectures. The automation of the network architecture search by genetic algorithms seems to have been the best way to achieve this goal.

Rajkumar, T.↗

DOE OSTI2024

Object Detection and Recognition with PointPillars in LiDAR Point Clouds – Comparisions

In the field of autonomous systems, neural networks have been leveraged for object detection and recognition in 2-dimensional images captured by cameras. Other types of sensors are available for sensing surroundings, including LiDAR sensors, and corresponding networks have been developed to perform detection and recognition in the point clouds generated by these sensors. The approaches are similar, both perform convolutions, but have distinct characteristics and challenges. In designing and configuring autonomous systems, a variety of LiDAR sensors are available, along with configurable deep neural networks to leverage their data. This work presents a review of the PointPillars network, an evolution of the seminal PointNet, comparing accuracy and training time relative to different LiDAR sensors, network and training parameters, CPU and GPU hardware, and the criticality of the use of reflective intensity as a feature. The value of using reflectivity as a predictive feature is explored and quantified to determine if it makes a significant difference in accuracy of the PointPillars network. Two separate LiDAR sensors are utilized, a 16-plane and a 32-plane, and corresponding accuracies and training times with the PointPillars network are evaluated.

LiDAR, machine learning, neural network, object re↗

NASA NTRS1991

Simplified Learning Scheme For Analog Neural Network

Synaptic connections adjusted one at a time in small increments. Simplified gradient-descent learning scheme for electronic neural-network processor less efficient than better-known back-propagation scheme, but offers two advantages: easily implemented in circuitry because data-access circuitry separated from learning circuitry; and independence of data-access circuitry makes possible to implement feedforward as well as feedback networks, including those of multiple-attractor type. Important in such applications as recognition of patterns.

Eberhardt, Silvio P.↗

NASA NTRS1996

CI Controls for Energy and Environment

Computational intelligence (CI) is a rapidly evolving field that utilizes life imitating metaphors for guiding model building including, but not limited to neural networks, fuzzy logic, genetic algorithms, artificial life, and hybrid CI paradigms. Although the boundaries between artificial intelligence (AI) and CI are not distinct, their research communities are separate and distinct. CI researchers tend to focus on processing numerical data from sensors, while the AI community generally relies on symbolic computing to capture human knowledge. In both areas, there is a great deal of interest and activity in hybrid systems that can offset the limitations of individual methods, extend their capabilities, and create new capabilities. Examples of the benefits that can accrue from hybrid systems are contained.

Biondo, Samuel J.↗

NASA NTRS2002

Determination of Storm Flashing/Non-Flashing Condition From Convective and Environmental Observations

A simple and fundamental problem in cloud electrification is whether or not a cloud can be determined to be producing lightning or not producing lightning, based solely on knowledge of its microphysical (and perhaps environmental) state. A merged database of TRMM radar, microwave and lightning observations and NCEP reanalysis environmental parameters is used to answer this question, for the tropics. The formal skill of traditional, univariate rule-based approaches (e.g., 35 dBZ occurrence at 6 km altitude) is quantified (via the probability of detection (POD), false alarm rate (FAR) and critical skill index (CSI)). Under indiscriminate application to the tropics, peak rule-based CSI for categorization of flashing storms is approximately 50%, with peak POD approximately 67% and minimum FAR approximately 33%, with peak CSI found for radar reflectivity-based parameters at 7-7.5 km altitude (near -15C). Separation of land and ocean domains yields approximately 5-10% gains in CSI over land. Conventional multivariate categorization techniques (discriminant analysis) are then applied, and less conventional (neural network) categorization techniques are also discussed.

Boccippio, Dennis J.↗

NASA NTRSDate not supplied

Artificial Neural Network (ANN) Surface Longwave and Shortwave Fluxes Trained on CERES Observations

The Clouds and Earth’s Radiant Energy System (CERES) project provides satellite-based observations of the radiative fluxes and clouds systems. CERES climate quality data products typically take several months of calibration and validation before release to the public. The Fast Longwave and Shortwave Radiative Flux (FLASHFlux) data product was developed to provide key data for the applied sciences and educational users within a week of observation. FLASHFlux achieves this by using simplified calibration, an operational meteorological product from Global Modeling and Assimilation Office (GMAO), and its own surface parameterizations model. The CERES FLASHFlux provides two data products: 1) an hourly Level 2 Single Scanner Footprint (SSF) data separately for Terra and NOAA-20 observations, and 2) a daily Level 3 Time Interpolated and Spatially Averaged (TISA) 1o x 1o gridded data that combines Terra and NOAA-20 observations. Currently, FLASHFlux uses the Langley Parameterized Shortwave Algorithm (LPSA) and Langley Parameterized Longwave Algorithm (LPLA) to derive its surface fluxes (Kratz et al., 2010; Gupta et al, 2001). A new Machine Learning (ML) based approach using Artificial Neural Networks to derive Surface Longwave (LW) & Shortwave (SW) fluxes based on training data from the CERES Clouds Radiative Swath (CRS) product is being investigated to replace LPSA and LPLA in the SSF surface flux products. One of the biggest hurdles in training ML model is model fitting. To overcome the problem of overfitting we use feature engineering that helps in finding the important feature and remove features that are irrelevant to the model. In our training we employed the Leave-One-Feature-Out Importance (LOFO) to evaluate the significance of each feature in our training. We intercompare ANN fluxes against surface fluxes produced from the Fu-Liou model in CRS and the LPSA/LPLA in FLASHFlux SSF. Furthermore, we validated ANN derived fluxes to the Baseline Surface Radiation Network (BSRN).

P C Sawaengphokhai↗

NASA NTRS1990

Memory switches based on metal oxide thin films

MnO.sub.2-x thin films (12) exhibit irreversible memory switching (28) with an OFF/ON resistance ratio of at least about 10.sup.3 and the tailorability of ON state (20) resistance. Such films are potentially extremely useful as a connection element in a variety of microelectronic circuits and arrays (24). Such films provide a pre-tailored, finite, non-volatile resistive element at a desired place in an electric circuit, which can be electrically turned OFF (22) or disconnected as desired, by application of an electrical pulse. Microswitch structures (10) constitute the thin film element, contacted by a pair of separate electrodes (16a, 16b) and have a finite, pre-selected ON resistance which is ideally suited, for example, as a programmable binary synaptic connection for electronic implementation of neural network architectures. The MnO.sub.2-x microswitch is non-volatile, patternable, insensitive to ultraviolet light, and adherent to a variety of insulating substrates (14), such as glass and silicon dioxide-coated silicon substrates.

Ramesham, Rajeshuni↗

NASA NTRS1990

Adaptive pattern recognition by mini-max neural networks as a part of an intelligent processor

In this decade and progressing into 21st Century, NASA will have missions including Space Station and the Earth related Planet Sciences. To support these missions, a high degree of sophistication in machine automation and an increasing amount of data processing throughput rate are necessary. Meeting these challenges requires intelligent machines, designed to support the necessary automations in a remote space and hazardous environment. There are two approaches to designing these intelligent machines. One of these is the knowledge-based expert system approach, namely AI. The other is a non-rule approach based on parallel and distributed computing for adaptive fault-tolerances, namely Neural or Natural Intelligence (NI). The union of AI and NI is the solution to the problem stated above. The NI segment of this unit extracts features automatically by applying Cauchy simulated annealing to a mini-max cost energy function. The feature discovered by NI can then be passed to the AI system for future processing, and vice versa. This passing increases reliability, for AI can follow the NI formulated algorithm exactly, and can provide the context knowledge base as the constraints of neurocomputing. The mini-max cost function that solves the unknown feature can furthermore give us a top-down architectural design of neural networks by means of Taylor series expansion of the cost function. A typical mini-max cost function consists of the sample variance of each class in the numerator, and separation of the center of each class in the denominator. Thus, when the total cost energy is minimized, the conflicting goals of intraclass clustering and interclass segregation are achieved simultaneously.

Szu, Harold H.↗

NASA NTRS2001

Implementation of a General Real-Time Visual Anomaly Detection System Via Soft Computing

The intelligent visual system detects anomalies or defects in real time under normal lighting operating conditions. The application is basically a learning machine that integrates fuzzy logic (FL), artificial neural network (ANN), and generic algorithm (GA) schemes to process the image, run the learning process, and finally detect the anomalies or defects. The system acquires the image, performs segmentation to separate the object being tested from the background, preprocesses the image using fuzzy reasoning, performs the final segmentation using fuzzy reasoning techniques to retrieve regions with potential anomalies or defects, and finally retrieves them using a learning model built via ANN and GA techniques. FL provides a powerful framework for knowledge representation and overcomes uncertainty and vagueness typically found in image analysis. ANN provides learning capabilities, and GA leads to robust learning results. An application prototype currently runs on a regular PC under Windows NT, and preliminary work has been performed to build an embedded version with multiple image processors. The application prototype is being tested at the Kennedy Space Center (KSC), Florida, to visually detect anomalies along slide basket cables utilized by the astronauts to evacuate the NASA Shuttle launch pad in an emergency. The potential applications of this anomaly detection system in an open environment are quite wide. Another current, potentially viable application at NASA is in detecting anomalies of the NASA Space Shuttle Orbiter's radiator panels.

Dominguez, Jesus A.↗

DOE OSTI2026

Advancing Artificial Intelligence with Liquid Argon Neutrino Experiments (Technical Report)

The grant allowed two main contributions: 1) The development of a first successful demonstration of the employment of Optimal Transport in liquid argon time projection chamber neutrino detectors. Optimal Transport, used in other contexts and specifically with LHC calorimetric data, was adapted to address a key particle identification challenge in LArTPCs: the separation of pi0 backgrounds from single-electrons produced in charged-current electron neutrino interactions. The work, leveraging ML methods such as k-nearest-neighbor (kNN) and support-vector-machine (SVM), showed an increase in background rejection of a factor of two or more. Work is now ongoing to incorporate this development in physics analyses for LArTPC experiments and more broadly expand the use of OT in LArTPC detectors including DUNE. This work was done in collaboration with the phenomenology group led by Nathaniel Craig at UCSB. 2) The deployment of NuGraph2, a graph neural network developed for LArTPC reconstruction, in the MicroBooNE experiment. NuGraph2 uses novel graph-neural-network methods on the rather simple LArTPC inputs of reconstructed hits, greatly simplifying the workflow compared to the use of waveform or signal-deconvolved wire ROIs. The network performed particle classification and was shown to address many challenging problems in LArTPC imaging including track-shower separation and the identification of protons and charged pions from primary muons. Our group collaborated with Giuseppe Cerati (FNAL scientist) who is one of the core developers of NuGraph2 to integrate this tool in MicroBooNE’s analysis framework. This consisted in tow key contributions: a) Studying performance on real data, which came with several months of iterations because the MC-trained version of the network was found to show significant bias that our group investigated and addressed. b) Integrating the output hit labeling of NuGraph2 into the existing particle tracking and shower reconstruction code. As a result of this work led by our team NuGraph2 is now enabling a suite of new analyses which benefit from enhanced capabilities and thus broader physics reach. The grant supported primarily the salary of UCSB graduate student Chuyue “Michaelia” Fang as well as partial summer salary support for PI Caratelli. Some funds were used for travel by Michaelia to ML related schools and conferences.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

NASA NTRS2002

Method and system for training dynamic nonlinear adaptive filters which have embedded memory

Described herein is a method and system for training nonlinear adaptive filters (or neural networks) which have embedded memory. Such memory can arise in a multi-layer finite impulse response (FIR) architecture, or an infinite impulse response (IIR) architecture. We focus on filter architectures with separate linear dynamic components and static nonlinear components. Such filters can be structured so as to restrict their degrees of computational freedom based on a priori knowledge about the dynamic operation to be emulated. The method is detailed for an FIR architecture which consists of linear FIR filters together with nonlinear generalized single layer subnets. For the IIR case, we extend the methodology to a general nonlinear architecture which uses feedback. For these dynamic architectures, we describe how one can apply optimization techniques which make updates closer to the Newton direction than those of a steepest descent method, such as backpropagation. We detail a novel adaptive modified Gauss-Newton optimization technique, which uses an adaptive learning rate to determine both the magnitude and direction of update steps. For a wide range of adaptive filtering applications, the new training algorithm converges faster and to a smaller value of cost than both steepest-descent methods such as backpropagation-through-time, and standard quasi-Newton methods. We apply the algorithm to modeling the inverse of a nonlinear dynamic tracking system 5, as well as a nonlinear amplifier 6.

Rabinowitz, Matthew↗