SEARCH · Engineering Papers
Results for “neural algorithms”
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.
Using the Metropolis algorithm to explore the loss surface of a recurrent neural network
In the limit of small trial moves the Metropolis Monte Carlo algorithm is equivalent to gradient descent on the energy function in the presence of Gaussian white noise. This observation was originally used to demonstrate a correspondence between Metropolis Monte Carlo moves of model molecules and overdamped Langevin dynamics, but it also applies in the context of training a neural network: making small random changes to the weights of a neural network, accepted with the Metropolis probability, with the loss function playing the role of energy, has the same effect as training by explicit gradient descent in the presence of Gaussian white noise. We explore this correspondence in the context of a simple recurrent neural network. We also explore regimes in which this correspondence breaks down, where the gradient of the loss function becomes very large or small. In these regimes the Metropolis algorithm can still effect training, and so can be used as a probe of the loss function of a neural network in regimes in which gradient descent struggles. We also show that training can be accelerated by making purposely-designed Monte Carlo trial moves of neural-network weights.
Predictability of Seawater DMS During the North Atlantic Aerosol and Marine Ecosystem Study (NAAMES)
This work presents an overview of a unique set of surface ocean dimethylsulfide (DMS) measurements from four shipboard field campaigns conducted during the North Atlantic Aerosol and Marine Ecosystem Study (NAAMES) project. Variations in surface seawater DMS are discussed in relation to biological and physical observations. Results are considered at a range of timescales (seasons to days) and spatial scales (regional to sub-mesoscale). Elevated DMS concentrations are generally associated with greater biological productivity, although chlorophyll a (Chl) only explains a small fraction of the DMS variability (15%). Physical factors that determine the location of oceanic temperature fronts and depth of vertical mixing have an important influence on seawater DMS concentrations during all seasons. The interplay of biomass and physics influences DMS concentrations at regional/seasonal scales and at smaller spatial and shorter temporal scales. Seawater DMS measurements are compared with the global seawater DMS climatology and predictions made using a recently published algorithm and by a neural network model. The climatology is successful at capturing the seasonal progression in average seawater DMS, but does not reproduce the shorter spatial/temporal scale variability. The input terms common to the algorithm and neural network approaches are biological (Chl) and physical (mixed layer depth, photosynthetically active radiation, seawater temperature). Both models predict the seasonal North Atlantic average seawater DMS trends better than the climatology. However, DMS concentrations tend to be under-predicted and the episodic occurrence of higher DMS concentrations is poorly predicted. The choice of climatological seawater DMS product makes a substantial impact on the estimated DMS flux into the North Atlantic atmosphere. These results suggest that additional input terms are needed to improve the predictive capability of current state-of-the-art approaches to estimating seawater DMS.
Design of thrust vectoring exhaust nozzles for real-time applications using neural networks
Thrust vectoring continues to be an important issue in military aircraft system designs. A recently developed concept of vectoring aircraft thrust makes use of flexible exhaust nozzles. Subtle modifications in the nozzle wall contours produce a non-uniform flow field containing a complex pattern of shock and expansion waves. The end result, due to the asymmetric velocity and pressure distributions, is vectored thrust. Specification of the nozzle contours required for a desired thrust vector angle (an inverse design problem) has been achieved with genetic algorithms. This approach is computationally intensive and prevents the nozzles from being designed in real-time, which is necessary for an operational aircraft system. An investigation was conducted into using genetic algorithms to train a neural network in an attempt to obtain, in real-time, two-dimensional nozzle contours. Results show that genetic algorithm trained neural networks provide a viable, real-time alternative for designing thrust vectoring nozzles contours. Thrust vector angles up to 20 deg were obtained within an average error of 0.0914 deg. The error surfaces encountered were highly degenerate and thus the robustness of genetic algorithms was well suited for minimizing global errors.
Accelerating network layouts using graph neural networks
Graph layout algorithms used in network visualization represent the first and the most widely used tool to unveil the inner structure and the behavior of complex networks. Current network visualization software relies on the force-directed layout (FDL) algorithm, whose high computational complexity makes the visualization of large real networks computationally prohibitive and traps large graphs into high energy configurations, resulting in hard-to-interpret “hairball” layouts. Here we use Graph Neural Networks (GNN) to accelerate FDL, showing that deep learning can address both limitations of FDL: it offers a 10 to 100 fold improvement in speed while also yielding layouts which are more informative. We analytically derive the speedup offered by GNN, relating it to the number of outliers in the eigenspectrum of the adjacency matrix, predicting that GNNs are particularly effective for networks with communities and local regularities. Finally, we use GNN to generate a three-dimensional layout of the Internet, and introduce additional measures to assess the layout quality and its interpretability, exploring the algorithm’s ability to separate communities and the link-length distribution. The novel use of deep neural networks can help accelerate other network-based optimization problems as well, with applications from reaction-diffusion systems to epidemics.
Structured Neural Network Modeling for Developing Digital Twins Models of Hydropower Generation Units
Dynamic modeling is a key part in the development of digital twin (DT) for dynamic systems. This is true for hydropower systems, where whole system modeling including penstock, turbine and generators, etc is important in realizing actuate modeling for the real systems. On the other hand, in response to the large variations of the power demand due to increased penetration of renewables such as wind and solar, hydropower systems are now required to operate in a large power generation range. This situation triggers the nonlinear characteristics of the generation unit with respect to its models. As such, it is imperative to use data driven modeling such as neural networks to learn the nonlinear dynamics of the hydropower generation unit. To achieve this objective, this study constructs a modeling and learning algorithm integrated with multiple structured neural network models for the modeling of turbine shaft speed, penstock pressure, and generator power output based on the generator power control setpoint, field current, and field voltage. In addition, the study uses the hydropower data from Tacoma Public Utilities to train and validate the proposed neural network algorithm. The results have shown that this structured neural network modeling approach can learn the system dynamics effectively by using the real-time data collected from the hydropower system with the desired modeling results.
Approximating Nash Equilibrium in Day-ahead Electricity Market Bidding with Multi-agent Deep Reinforcement Learning
In this paper, a day-ahead electricity market bidding problem with multiple strategic generation company (GEN-CO) bidders is studied. The problem is formulated as a Markov game model, where GENCO bidders interact with each other todevelop their optimal day-ahead bidding strategies. Considering unobservable information in the problem, a model-free and data-driven approach, known as multi-agent deep deterministic policy gradient (MADDPG), is applied for approximating the Nash equilibrium (NE) in the above Markov game. The MADDPG algorithm has the advantage of generalization due to the automatic feature extraction ability of the deep neural networks. The algorithm is tested on an IEEE 30-bus system with three competitive GENCO bidders in both an uncongested caseand a congested case. Comparisons with a truthful bidding strategy and state-of-the-art deep reinforcement learning methods including deep Q network and deep deterministic policy gradient (DDPG) demonstrate that the applied MADDPG algorithm can find a superior bidding strategy for all the market participants with increased profit gains. In addition, the comparison with a conventional model-based method shows that the MADDPG algorithm has higher computational efficiency, which is feasible for real-world applications.
SDR Input Power Estimation Algorithms
The General Dynamics (GD) S-Band software defined radio (SDR) in the Space Communications and Navigation (SCAN) Testbed on the International Space Station (ISS) provides experimenters an opportunity to develop and demonstrate experimental waveforms in space. The SDR has an analog and a digital automatic gain control (AGC) and the response of the AGCs to changes in SDR input power and temperature was characterized prior to the launch and installation of the SCAN Testbed on the ISS. The AGCs were used to estimate the SDR input power and SNR of the received signal and the characterization results showed a nonlinear response to SDR input power and temperature. In order to estimate the SDR input from the AGCs, three algorithms were developed and implemented on the ground software of the SCAN Testbed. The algorithms include a linear straight line estimator, which used the digital AGC and the temperature to estimate the SDR input power over a narrower section of the SDR input power range. There is a linear adaptive filter algorithm that uses both AGCs and the temperature to estimate the SDR input power over a wide input power range. Finally, an algorithm that uses neural networks was designed to estimate the input power over a wide range. This paper describes the algorithms in detail and their associated performance in estimating the SDR input power.
Mapping Stochastic Devices to Probabilistic Algorithms
Probabilistic and Bayesian neural networks have long been proposed as a method to incorporate uncertainty about the world (both in training data and operation) into artificial intelligence applications. One approach to making a neural network probabilistic is to leverage a Monte Carlo sampling approach that samples a trained network while incorporating noise. Such sampling approaches for neural networks have not been extensively studied due to the prohibitive requirement of many computationally expensive samples. While the development of future microelectronics platforms that make this sampling more efficient is an attractive option, it has not been immediately clear how to sample a neural network and what the quality of random number generation should be. This research aimed to start addressing these two fundamental questions by examining basic “off the shelf” neural networks can be sampled through a few different mechanisms (including synapse “dropout” and neuron “dropout”) and examine how these sampling approaches can be evaluated both in terms of evaluating algorithm effectiveness and the required quality of random numbers.
Stiffness and Fatigue Life Estimator for Polymer Composite Laminates Using Machine Learning
Machine learning (ML) models are increasingly being used in many engineering fields due to the advancements in ML algorithms and availability of high-speed computing power. One of the most popular ML class of models is artificial neural networks (ANN). ML is increasingly being used in the design and analysis of composite materials and structures, specifically in the constitutive modeling of composite materials with the focus on greatly accelerating multiscale analyses of composite materials and structures through development of surrogate models. Towards that end, Python-based neural nets have been developed to predict initial stiffness and fatigue life of an eight-ply symmetric polymer matrix composite laminate. Two types of neural networks, a Multilayer Perceptron (MLP) and a Recurrent Neural Network (RNN), have been established. Results show that both neural net type algorithms can provide an excellent estimate of initial laminate stiffness as well as fatigue life of eight-ply symmetric polymer matrix composite laminates (PMCs). RNNs are better able to capture the shape of the fatigue curve of a laminate. The resulting tool and GUI can be very useful for system level studies to obtain an estimate of desired properties and life of PMC composite laminates. Further, the associated surrogate models can also be used in composite multiscale analyses to replace the actual physics-based calculations at lower scales and thereby significantly increase the computational efficiency of such analyses and thus make micromechanics-based multiscale analyses a viable industrial tool for large scale structural problems.
Stiffness and Fatigue Life Estimator for Polymer Composite Laminates Using Machine Learning
Machine learning (ML) models are increasingly being used in many engineering fields due to the advancements in ML algorithms and availability of high-speed computing power. One of the most popular ML class of models is artificial neural networks (ANN). ML is increasingly being used in the design and analysis of composite materials and structures, specifically in the constitutive modeling of composite materials with the focus on greatly accelerating multiscale analyses of composite materials and structures through development of surrogate models. Towards that end, Python-based neural nets have been developed to predict initial stiffness and fatigue life of an eight-ply symmetric polymer matrix composite laminate. Two types of neural networks, a Multilayer Perceptron (MLP) and a Recurrent Neural Network (RNN), have been established. Results show that both neural net type algorithms can provide an excellent estimate of initial laminate stiffness as well as fatigue life of eight-ply symmetric polymer matrix composite laminates (PMCs). RNNs are better able to capture the shape of the fatigue curve of a laminate. The resulting tool and GUI can be very useful for system level studies to obtain an estimate of desired properties and life of PMC composite laminates. Further, the associated surrogate models can also be used in composite multiscale analyses to replace the actual physics-based calculations at lower scales and thereby significantly increase the computational efficiency of such analyses and thus make micromechanics-based multiscale analyses a viable industrial tool for large scale structural problems.
Neural network based adaptive control of nonlinear plants using random search optimization algorithms
This paper presents a method for utilizing artificial neural networks for direct adaptive control of dynamic systems with poorly known dynamics. The neural network weights (controller gains) are adapted in real time using state measurements and a random search optimization algorithm. The results are demonstrated via simulation using two highly nonlinear systems.
Millimeter-Wave Radar Field Measurements and Inversion of Cloud Parameters for the 1999 Mt. Washington Icing Sensors Project
The Mount Washington Icing Sensors Project (MWISP) was a multi-investigator experiment with participants from Quadrant Engineering, NOAA Environmental Technology Laboratory (NOAA/ETL), the Microwave Remote Sensing Laboratory (MIRSL) of the University of Massachusetts (UMass), and others. Radar systems from UMass and NOAA/ETL were used to measure X-, Ka-, and W-band backscatter data from the base of Mt. Washington, while simultaneous in-situ particle measurements were made from aircraft and from the observatory at the summit. This report presents range and time profiles of liquid water content and particle size parameters derived from range profiles of radar reflectivity as measured at X-, Ka-, and W-band (9.3, 33.1, and 94.9 GHz) using an artificial neural network inversion algorithm. In this report, we provide a brief description of the experiment configuration, radar systems, and a review of the artificial neural network used to extract cloud parameters from the radar data. Time histories of liquid water content (LWC), mean volume diameter (MVD) and mean Z diameter (MZD) are plotted at 300 m range intervals for slant ranges between 1.1 and 4 km. Appendix A provides details on the extraction of radar reflectivity from measured radar power, and Appendix B provides summary logs of the weather conditions for each day in which we processed data.
A Comparison between Invariant and Equivariant Classical and Quantum Graph Neural Networks
Machine learning algorithms are heavily relied on to understand the vast amounts of data from high-energy particle collisions at the CERN Large Hadron Collider (LHC). The data from such collision events can naturally be represented with graph structures. Therefore, deep geometric methods, such as graph neural networks (GNNs), have been leveraged for various data analysis tasks in high-energy physics. One typical task is jet tagging, where jets are viewed as point clouds with distinct features and edge connections between their constituent particles. The increasing size and complexity of the LHC particle datasets, as well as the computational models used for their analysis, have greatly motivated the development of alternative fast and efficient computational paradigms such as quantum computation. In addition, to enhance the validity and robustness of deep networks, we can leverage the fundamental symmetries present in the data through the use of invariant inputs and equivariant layers. In this paper, we provide a fair and comprehensive comparison of classical graph neural networks (GNNs) and equivariant graph neural networks (EGNNs) and their quantum counterparts: quantum graph neural networks (QGNNs) and equivariant quantum graph neural networks (EQGNN). The four architectures were benchmarked on a binary classification task to classify the parton-level particle initiating the jet. Based on their area under the curve (AUC) scores, the quantum networks were found to outperform the classical networks. However, seeing the computational advantage of quantum networks in practice may have to wait for the further development of quantum technology and its associated application programming interfaces (APIs).
Linear and nonlinear trending and prediction for AVHRR time series data
The variability of AVHRR calibration coefficient in time was analyzed using algorithms of linear and non-linear time series analysis. Specifically we have used the spline trend modeling, autoregressive process analysis, incremental neural network learning algorithm and redundancy functional testing. The analysis performed on available AVHRR data sets revealed that (1) the calibration data have nonlinear dependencies, (2) the calibration data depend strongly on the target temperature, (3) both calibration coefficients and the temperature time series can be modeled, in the first approximation, as autonomous dynamical systems, (4) the high frequency residuals of the analyzed data sets can be best modeled as an autoregressive process of the 10th degree. We have dealt with a nonlinear identification problem and the problem of noise filtering (data smoothing). The system identification and filtering are significant problems for AVHRR data sets. The algorithms outlined in this study can be used for the future EOS missions. Prediction and smoothing algorithms for time series of calibration data provide a functional characterization of the data. Those algorithms can be particularly useful when calibration data are incomplete or sparse.
A neural network-based estimator for the mixture ratio of the Space Shuttle Main Engine
In order to properly utilize the available fuel and oxidizer of a liquid propellant rocket engine, the mixture ratio is closed loop controlled during main stage (65 percent - 109 percent power) operation. However, because of the lack of flight-capable instrumentation for measuring mixture ratio, the value of mixture ratio in the control loop is estimated using available sensor measurements such as the combustion chamber pressure and the volumetric flow, and the temperature and pressure at the exit duct on the low pressure fuel pump. This estimation scheme has two limitations. First, the estimation formula is based on an empirical curve fitting which is accurate only within a narrow operating range. Second, the mixture ratio estimate relies on a few sensor measurements and loss of any of these measurements will make the estimate invalid. In this paper, we propose a neural network-based estimator for the mixture ratio of the Space Shuttle Main Engine. The estimator is an extension of a previously developed neural network based sensor failure detection and recovery algorithm (sensor validation). This neural network uses an auto associative structure which utilizes the redundant information of dissimilar sensors to detect inconsistent measurements. Two approaches have been identified for synthesizing mixture ratio from measurement data using a neural network. The first approach uses an auto associative neural network for sensor validation which is modified to include the mixture ratio as an additional output. The second uses a new network for the mixture ratio estimation in addition to the sensor validation network. Although mixture ratio is not directly measured in flight, it is generally available in simulation and in test bed firing data from facility measurements of fuel and oxidizer volumetric flows. The pros and cons of these two approaches will be discussed in terms of robustness to sensor failures and accuracy of the estimate during typical transients using simulation data.
Computation and parallel implementation for early vision
The problem of early vision is to transform one or more retinal illuminance images-pixel arrays-to image representations built out of such primitive visual features such as edges, regions, disparities, and clusters. These transformed representations form the input to later vision stages that perform higher level vision tasks including matching and recognition. Researchers developed algorithms for: (1) edge finding in the scale space formulation; (2) correlation methods for computing matches between pairs of images; and (3) clustering of data by neural networks. These algorithms are formulated for parallel implementation of SIMD machines, such as the Massively Parallel Processor, a 128 x 128 array processor with 1024 bits of local memory per processor. For some cases, researchers can show speedups of three orders of magnitude over serial implementations.
Phase Space Reconstruction from Accelerator Beam Measurements Using Neural Networks and Differentiable Simulations
Characterizing the phase space distribution of particle beams in accelerators is a central part of accelerator understanding and performance optimization. However, conventional reconstruction-based techniques either use simplifying assumptions or require specialized diagnostics to infer high-dimensional (> $2D$) beam properties. In this Letter, we introduce a general-purpose algorithm that combines neural networks with differentiable particle tracking to efficiently reconstruct high-dimensional phase space distributions without using specialized beam diagnostics or beam manipulations. Furthermore, we demonstrate that our algorithm accurately reconstructs detailed 4D phase space distributions with corresponding confidence intervals in both simulation and experiment using a single focusing quadrupole and diagnostic screen. This technique allows for the measurement of multiple correlated phase spaces simultaneously, which will enable simplified 6D phase space distribution reconstructions in the future.