Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Neural Network”

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 865 records · Page 48

Resilience and Robustness of Spiking Neural Networks for Neuromorphic Systems

Though robustness and resilience are commonly quoted as features of neuromorphic computing systems, the expected performance of neuromorphic systems in the face of hardware failures is not clear. In this work, we study the effect of failures on the performance of four different training algo-rithms for spiking neural networks on neuromorphic systems: two back-propagation-based training approaches (Whetstone and SLAYER), a liquid state machine or reservoir computing approach, and an evolutionary optimization-based approach (EONS). We show that these four different approaches have very different resilience characteristics with respect to simulated hardware failures. We then analyze an approach for training more resilient spiking neural networks using the evolutionary optimization approach. We show how this approach produces more resilient networks and discuss how it can be extended to other spiking neural network training approaches as well.

Schuman, Catherine↗

Permutation invariant polynomial neural network approach to fitting potential energy surfaces. II. Four-atom systems

A rigorous, general, and simple method to fit global and permutation invariant potential energy surfaces (PESs) using neural networks (NNs) is discussed. This so-called permutation invariant polynomial neural network (PIP-NN) method imposes permutation symmetry by using in its input a set of symmetry functions based on PIPs. For systems with more than three atoms, it is shown that the number of symmetry functions in the input vector needs to be larger than the number of internal coordinates in order to include both the primary and secondary invariant polynomials. This PIP-NN method is successfully demonstrated in three atom-triatomic reactive systems, resulting in full-dimensional global PESs with average errors on the order of meV. These PESs are used in full-dimensional quantum dynamical calculations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Aerodynamic Design Using Neural Networks

The design of aerodynamic components of aircraft, such as wings or engines, involves a process of obtaining the most optimal component shape that can deliver the desired level of component performance, subject to various constraints, e.g., total weight or cost, that the component must satisfy. Aerodynamic design can thus be formulated as an optimization problem that involves the minimization of an objective function subject to constraints. A new aerodynamic design optimization procedure based on neural networks and response surface methodology (RSM) incorporates the advantages of both traditional RSM and neural networks. The procedure uses a strategy, denoted parameter-based partitioning of the design space, to construct a sequence of response surfaces based on both neural networks and polynomial fits to traverse the design space in search of the optimal solution. Some desirable characteristics of the new design optimization procedure include the ability to handle a variety of design objectives, easily impose constraints, and incorporate design guidelines and rules of thumb. It provides an infrastructure for variable fidelity analysis and reduces the cost of computation by using less-expensive, lower fidelity simulations in the early stages of the design evolution. The initial or starting design can be far from optimal. The procedure is easy and economical to use in large-dimensional design space and can be used to perform design tradeoff studies rapidly. Designs involving multiple disciplines can also be optimized. Some practical applications of the design procedure that have demonstrated some of its capabilities include the inverse design of an optimal turbine airfoil starting from a generic shape and the redesign of transonic turbines to improve their unsteady aerodynamic characteristics.

Rai, Man Mohan↗

Determination of Infinite Dilution Activity Coefficients of Molecular Solutes in Ionic Liquids and Deep Eutectic Solvents by Factorization-Machine-Based Neural Networks

Widely known as “green solvents,” ionic liquids (ILs) and deep eutectic solvents (DESs) have been used as substitutes for traditional organic solvents in separation science. To achieve better separations using ILs and DESs, this work aimed to predict the infinite dilution activity coefficients (IDACs) of molecular solutes in these solvents with the state-of-the-art factorization-machine-based neural network (DeepFM). DeepFM combines the benefits of factorization machines and deep neural networks to learn both low-order and high-order interactions among features. The IDAC prediction model was established with 52,372 experimental IDAC datapoints including 260 solvents (252 ILs and 8 DESs) and 112 molecular solutes collected at various temperatures from 288.15 to 428.15 K. Chemical information describing the ILs, DESs, and molecular solutes was included in the IDAC prediction model, including chemical functional groups, molecular weights, and Abraham solvation parameters. The IDAC prediction model showed an improved accuracy compared with alternative models; additionally, DESs were included in the IDAC prediction model for the first time. Here, the model will reduce the energy and resources needed to optimize the selection of ILs and DESs for specific separations, which will promote the development of these green solvents for sustainable chemical processes.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Enhanced physics-constrained deep neural networks for modeling vanadium redox flow battery

Numerical simulation has become indispensable in advancing cost-effective process optimization and control of flow batteries. We propose an enhanced version of the physics-constrained deep neural network (PCDNN) approach to provide high-accuracy voltage predictions in the vanadium redox flow batteries (VRFBs). The purpose of the PCDNN approach is to enforce the physics-based zero-dimensional (0D) VRFB model in a neural network to assure model generalization for various battery operation conditions. However, limited by the simplifications of the 0D model, the PCDNN cannot capture sharp voltage changes in the extreme SOC regions. To improve the accuracy of voltage prediction at extreme ranges, we introduce a second (enhanced) DNN to mitigate the prediction errors carried from the 0D model itself and call the resulting approach enhanced PCDNN (ePCDNN). By comparing with experimental data, we demonstrate that the ePCDNN approach can accurately capture the voltage response throughout the charge–discharge cycle, including the tail region of the voltage discharge curve. The loss function for training the ePCDNN is designed to be flexible by adjusting the weights of the physics-constrained DNN and the enhanced DNN. In conclusion, this allows the ePCDNN framework to be transferable to battery systems with variable physical model fidelity.

25 ENERGY STORAGE↗

Improving Precipitation Estimation Using Convolutional Neural Network

Precipitation process is generally considered to be poorly represented in numerical weather/climate models. Statistical downscaling (SD) methods, which relate precipitation with model resolved dynamics, often provide more accurate precipitation estimates compared to model's raw precipitation products. Herein we introduce the convolutional neural network model to foster this aspect of SD for daily precipitation prediction. Specifically, we restrict the predictors to the variables that are directly resolved by discretizing the atmospheric dynamics equations. In this sense, our model works as an alternative to the existing precipitation-related parameterization schemes for numerical precipitation estimation. We train the model to learn precipitation-related dynamical features from the surrounding dynamical fields by optimizing a hierarchical set of spatial convolution kernels. We test the model at 14 geogrid points across the contiguous United States. Results show that provided with enough data, precipitation estimates from the convolutional neural network model outperform the reanalysis precipitation products, as well as SD products using linear regression, nearest neighbor, random forest, or fully connected deep neural network. Evaluation for the test set suggests that the improvements can be seamlessly transferred to numerical weather modeling for improving precipitation prediction. Based on the default network, we examine the impact of the network architectures on model performance. Also, we offer simple visualization and analyzing approaches to interpret the models and their results. Our study contributes to the following two aspects: First, we offer a novel approach to enhance numerical precipitation estimation; second, the proposed model provides important implications for improving precipitation-related parameterization schemes using a data-driven approach.

54 ENVIRONMENTAL SCIENCES↗

All Optical Neural Networks for Low Power Edge Computing

We developed a simplistic physics-based model of an all-optical neural network that mimics the encoder part of an autoencoder neural network for image compression. Our approach relies on the generation of a MATLAB-based model for both data compression and decompression and utilizes MATLAB's built-in autoencoder networks in combination with simple propagation of optical fields between layers constituting phase elements via Fourier transform. We optimize the phase elements using the particle swarm optimization technique and using our model, we demonstrate a compression ratio of 25% for 2828-pixel input images containing numeric digits from 0 to 9.

97 MATHEMATICS AND COMPUTING↗

Grad–Shafranov equilibria via data-free physics informed neural networks

A large number of magnetohydrodynamic (MHD) equilibrium calculations are often required for uncertainty quantification, optimization, and real-time diagnostic information, making MHD equilibrium codes vital to the field of plasma physics. In this paper, we explore a method for solving the Grad–Shafranov equation by using physics-informed neural networks (PINNs). For PINNs, we optimize neural networks by directly minimizing the residual of the partial differential equation as a loss function. We show that PINNs can accurately and effectively solve the Grad–Shafranov equation with several different boundary conditions, making it more flexible than traditional solvers. This method is flexible as it does not require any mesh and basis choice, thereby streamlining the computational process. We also explore the parameter space by varying the size of the model, the learning rate, and boundary conditions to map various tradeoffs such as between reconstruction error and computational speed. Additionally, we introduce a parameterized PINN framework, expanding the input space to include variables such as pressure, aspect ratio, elongation, and triangularity in order to handle a broader range of plasma scenarios within a single network. Parameterized PINNs could be used in future work to solve inverse problems such as shape optimization.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations

In the last 50 years there has been a tremendous progress in solving numerically the Navier-Stokes equations using finite differences, finite elements, spectral, and even meshless methods. Yet, in many real cases, we still cannot incorporate seamlessly (multi-fidelity) data into existing algorithms, and for industrial-complexity applications the mesh generation is time consuming and still an art. Moreover, solving ill-posed problems (e.g., lacking boundary conditions) or inverse problems is often prohibitively expensive and requires different formulations and new computer codes. Here, we employ physics-informed neural networks (PINNs), encoding the governing equations directly into the deep neural network via automatic differentiation, to overcome some of the aforementioned limitations for simulating incompressible laminar and turbulent flows. We develop the Navier-Stokes flow nets (NSFnets) by considering two different mathematical formulations of the Navier-Stokes equations: the velocity-pressure (VP) formulation and the vorticity-velocity (VV) formulation. Since this is a new approach, we first select some standard benchmark problems to assess the accuracy, convergence rate, computational cost and flexibility of NSFnets; analytical solutions and direct numerical simulation (DNS) databases provide proper initial and boundary conditions for the NSFnet simulations. The spatial and temporal coordinates are the inputs of the NSFnets, while the instantaneous velocity and pressure fields are the outputs for the VP-NSFnet, and the instantaneous velocity and vorticity fields are the outputs for the VV-NSFnet. This is unsupervised learning and, hence, no labeled data are required beyond boundary and initial conditions and the fluid properties. The residuals of the VP or VV governing equations, together with the initial and boundary conditions, are embedded into the loss function of the NSFnets. No data is provided for the pressure to the VP-NSFnet, which is a hidden state and is obtained via the incompressibility constraint without extra computational cost. Unlike the traditional numerical methods, NSFnets inherit the properties of neural networks (NNs), hence the total error is composed of the approximation, the optimization, and the generalization errors. Here, we empirically attempt to quantify these errors by varying the sampling (“residual”) points, the iterative solvers, and the size of the NN architecture. For the laminar flow solutions, we show that both the VP and the VV formulations are comparable in accuracy but their best performance corresponds to different NN architectures. The initial convergence rate is fast but the error eventually saturates to a plateau due to the dominance of the optimization error. For the turbulent channel flow, we show that NSFnets can sustain turbulence at , but due to expensive training we only consider part of the channel domain and enforce velocity boundary conditions on the subdomain boundaries provided by the DNS data base. We also perform a systematic study on the weights used in the loss function for balancing the data and physics components, and investigate a new way of computing the weights dynamically to accelerate training and enhance accuracy. In the last part, we demonstrate how NSFnets should be used in practice, namely for ill-posed problems with incomplete or noisy boundary conditions as well as for inverse problems. We obtain reasonably accurate solutions for such cases as well without the need to change the NSFnets and at the same computational cost as in the forward well-posed problems. As a result, we also present a simple example of transfer learning that will aid in accelerating the training of NSFnets for different parameter settings.

97 MATHEMATICS AND COMPUTING↗

Training the Recurrent neural network by the Fuzzy Min-Max algorithm for fault prediction

In this paper, we present a training technique of a Recurrent Radial Basis Function neural network for fault prediction. We use the Fuzzy Min-Max technique to initialize the k-center of the RRBF neural network. The k-means algorithm is then applied to calculate the centers that minimize the mean square error of the prediction task. The performances of the k-means algorithm are then boosted by the Fuzzy Min-Max technique.

97 MATHEMATICS AND COMPUTING↗

Using a Neural Network to Determine the Hatch Status of the AERI at the ARM North Slope of Alaska Site

The fore-optics of the Atmospheric Emitted Radiance Interferometer (AERI) are protected by an automated hatch to prevent precipitation from fouling the instrument's scene mirror (Knuteson et al. 2004). Limit switches connected with the hatch controller provide a signal of the hatch state: open, closed, undetermined (typically associated with the hatch being between fully open or fully closed during the instrument's sky view period), or an error condition. The instrument then records the state of the hatch with the radiance data so that samples taken when the hatch is not open can be removed from any subsequent analysis. However, the hatch controller suffered a multi-year failure for the AERI located at the ARM North Slope of Alaska (NSA) Central Facility in Barrow, Alaska, from July 2006-February 2008. The failure resulted in misreporting the state of the hatch in the 'hatchOpen' field within the AERI data files. With this error there is no simple solution to translate what was reported back to the correct hatch status, thereby making it difficult for an analysis to determine when the AERI was actually viewing the sky. As only the data collected when the hatch is fully open are scientifically useful, an algorithm was developed to determine whether the hatch was open or closed based on spectral radiance data from the AERI. Determining if the hatch is open or closed in a scene with low clouds is non-trivial, as low opaque clouds may look very similar spectrally as the closed hatch. This algorithm used a backpropagation neural network; these types of neural networks have been used with increasing frequency in atmospheric science applications.

47 OTHER INSTRUMENTATION↗

Neural networks for data compression and invariant image recognition

An approach to invariant image recognition (I2R), based upon a model of biological vision in the mammalian visual system (MVS), is described. The complete I2R model incorporates several biologically inspired features: exponential mapping of retinal images, Gabor spatial filtering, and a neural network associative memory. In the I2R model, exponentially mapped retinal images are filtered by a hierarchical set of Gabor spatial filters (GSF) which provide compression of the information contained within a pixel-based image. A neural network associative memory (AM) is used to process the GSF coded images. We describe a 1-D shape function method for coding of scale and rotationally invariant shape information. This method reduces image shape information to a periodic waveform suitable for coding as an input vector to a neural network AM. The shape function method is suitable for near term applications on conventional computing architectures equipped with VLSI FFT chips to provide a rapid image search capability.

Gardner, Sheldon↗

A design philosophy for multi-layer neural networks with applications to robot control

A system is proposed which receives input information from many sensors that may have diverse scaling, dimension, and data representations. The proposed system tolerates sensory information with faults. The proposed self-adaptive processing technique has great promise in integrating the techniques of artificial intelligence and neural networks in an attempt to build a more intelligent computing environment. The proposed architecture can provide a detailed decision tree based on the input information, information stored in a long-term memory, and the adapted rule-based knowledge. A mathematical model for analysis will be obtained to validate the cited hypotheses. An extensive software program will be developed to simulate a typical example of pattern recognition problem. It is shown that the proposed model displays attention, expectation, spatio-temporal, and predictory behavior which are specific to the human brain. The anticipated results of this research project are: (1) creation of a new dynamic neural network structure, and (2) applications to and comparison with conventional multi-layer neural network structures. The anticipated benefits from this research are vast. The model can be used in a neuro-computer architecture as a building block which can perform complicated, nonlinear, time-varying mapping from a multitude of input excitory classes to an output or decision environment. It can be used for coordinating different sensory inputs and past experience of a dynamic system and actuating signals. The commercial applications of this project can be the creation of a special-purpose neuro-computer hardware which can be used in spatio-temporal pattern recognitions in such areas as air defense systems, e.g., target tracking, and recognition. Potential robotics-related applications are trajectory planning, inverse dynamics computations, hierarchical control, task-oriented control, and collision avoidance.

Vadiee, Nader↗

Symplectic neural network and its application to charged particle dynamics in electromagnetic fields

Recently, machine learning models have shown many successes in various applications in science and technology. In this work, we focus on the charged particle dynamics, with the development of a class of symplectic neural networks, including a linear version, SympMat, and a nonlinear version, HénonNet. Both are designed to preserve the structure of Hamiltonian systems. We show that they can be used to model relevant Hamiltonian systems of interest in plasma physics and astrophysics, for linear and nonlinear charged particle dynamics, with the potential to bridge multi-scale simulations. These symplectic neural networks are adapted to the applications in plasma simulations and particle-wave interaction with parametric dependence and periodicity, where we have investigated their performance and accuracy. In particular, SympMat is shown to outperform the traditional Boris particle pusher down to the sub-gyroperiod scale in the case of charged particles in uniform magnetic fields. HénonNet successfully predicts the hot electron distribution, which is validated against theoretical results. These results highlight the potential of symplectic neural networks as a trajectory integrator for particle-in-cell simulations or a fast surrogate to replace conventional numerical schemes.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Real-time capable modeling of neutral beam injection on NSTX-U using neural networks

A new model of heating, current drive, torque and other effects of neutral beam injection on NSTX-U that uses neural networks has been developed. The model has been trained and tested on the results of the Monte Carlo code NUBEAM for the database of experimental discharges taken during the first operational campaign of NSTX-U. By projecting flux surface quantities onto empirically derived basis functions, the model is able to efficiently and accurately reproduce the behavior of both scalars, like the total neutron rate and shine through, and profiles, like beam current drive and heating. The model has been tested on the NSTX-U real-time computer, demonstrating a rapid execution time orders of magnitude faster than the Monte Carlo code that is well suited for the iterative calculations needed to interpret experimental results, optimization during scenario development activities, and real-time plasma control applications. Simulation results of a proposed design for a nonlinear observer that embeds the neural network calculations to estimate the poloidal flux profile evolution, as well as effective charge and fast ion diffusivity, are presented.

Neural networks↗

Multi-Objective Hyperparameter Optimization for Spiking Neural Network Neuroevolution

Neuroevolution has had significant success over recent years, but there has been relatively little work applying neuroevolution approaches to spiking neural networks (SNNs). SNNs are a type of neural network that includes temporal processing component, are not easily trained using other methods, and can be deployed into energy-efficient neuromorphic hardware. In this work, we investigate two evolutionary approaches for training SNNs. We explore the impact of the hyperparameters of the evolutionary approaches, including tournament size, population size, and representation type, on the performance of the algorithms. We present a multi-objective Bayesian-based hyperparameter optimization approach to tune the hyperparameters to produce the most accurate and smallest SNNs. We show that the hyperparameters can significantly affect the performance of these algorithms. We also perform sensitivity analysis and demonstrate that every hyperparameter value has the potential to perform well, assuming other hyperparameter values are set correctly.

Parsa, Maryam↗

Conjugate-Gradient Neural Networks in Classification of Multisource and Very-High-Dimensional Remote Sensing Data

Application of neural networks to classification of remote sensing data is discussed. Conventional two-layer backpropagation is found to give good results in classification of remote sensing data but is not efficient in training. A more efficient variant, based on conjugate-gradient optimization, is used for classification of multisource remote sensing and geographic data and very-high-dimensional data. The conjugate-gradient neural networks give excellent performance in classification of multisource data, but do not compare as well with statistical methods in classification of very-high-dimentional data.

Benediktsson, J. A.↗

Predicting carcinogenicity of diverse chemicals using probabilistic neural network modeling approaches

Robust global models capable of discriminating positive and non-positive carcinogens; and predicting carcinogenic potency of chemicals in rodents were developed. The dataset of 834 structurally diverse chemicals extracted from Carcinogenic Potency Database (CPDB) was used which contained 466 positive and 368 non-positive carcinogens. Twelve non-quantum mechanical molecular descriptors were derived. Structural diversity of the chemicals and nonlinearity in the data were evaluated using Tanimoto similarity index and Brock–Dechert–Scheinkman statistics. Probabilistic neural network (PNN) and generalized regression neural network (GRNN) models were constructed for classification and function optimization problems using the carcinogenicity end point in rat. Validation of the models was performed using the internal and external procedures employing a wide series of statistical checks. PNN constructed using five descriptors rendered classification accuracy of 92.09% in complete rat data. The PNN model rendered classification accuracies of 91.77%, 80.70% and 92.08% in mouse, hamster and pesticide data, respectively. The GRNN constructed with nine descriptors yielded correlation coefficient of 0.896 between the measured and predicted carcinogenic potency with mean squared error (MSE) of 0.44 in complete rat data. The rat carcinogenicity model (GRNN) applied to the mouse and hamster data yielded correlation coefficient and MSE of 0.758, 0.71 and 0.760, 0.46, respectively. The results suggest for wide applicability of the inter-species models in predicting carcinogenic potency of chemicals. Both the PNN and GRNN (inter-species) models constructed here can be useful tools in predicting the carcinogenicity of new chemicals for regulatory purposes. - Graphical abstract: Figure (a) shows classification accuracies (positive and non-positive carcinogens) in rat, mouse, hamster, and pesticide data yielded by optimal PNN model. Figure (b) shows generalization and predictive abilities of the interspecies GRNN model to predict the carcinogenic potency of diverse chemicals. - Highlights: • Global robust models constructed for carcinogenicity prediction of diverse chemicals. • Tanimoto/BDS test revealed structural diversity of chemicals and nonlinearity in data. • PNN/GRNN successfully predicted carcinogenicity/carcinogenic potency of chemicals. • Developed interspecies PNN/GRNN models for carcinogenicity prediction. • Proposed models can be used as tool to predict carcinogenicity of new chemicals.

60 APPLIED LIFE SCIENCES↗