Search NASA⌕ Search

SEARCH · Search NASA

Results for “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.

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

At least 1,225 records · Page 68

Inference of three-dimensional hot-spot and shell morphology in inertial confinement fusion experiments using a convolutional neural network

The performance of inertial confinement fusion (ICF) implosions is sensitive to the three-dimensional (3D) morphology of the hot-spot and shell configurations. The ability to infer shell-mass uniformity and reconstruct 3D hot spots is crucial for quantifying the degradation of ignition criteria and improving symmetry in ICF implosion experiments. In this work, we present a deep-learning convolutional neural network (CNN) for reconstructing 3D hot-spot and shell structures for ICF capsules. The 3D geometry of the hot spot is reconstructed from x-ray images measured from multiple lines of sight on OMEGA. The shell configuration is inferred indirectly through machine learning using a convolutional neural network extensively trained on a dec3d simulation database. This simulation-dependent approach yields consistent agreement between reconstructed 3D shell densities and machine-learning optimized dec3d simulation results. This work demonstrates a CNN framework that successfully reconstructs 3D capsule structures from two-dimensional images in ICF implosions.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Learning quantum computers' errors using interpretable neural networks

Learning and reducing the errors and noise in quantum computing systems is necessary for achieving quantum computation’s promise. However, rapid advances in experimental quantum computing are making this task increasingly difficult, because state-of-the-art systems now contain hundreds of qubits and many characterization techniques are hard to apply at this scale. Furthermore, complex kinds of errors in these systems, such as crosstalk and non-Markovian effects, must be understood and decreased, but these errors are challenging to study with most existing methods. In this project, we explored using neural networks for scalable characterization of complex errors in quantum computers. We proposed and demonstrated characterizing a quantum computer’s errors with neural networks that have interpretable parameters corresponding to the rates of different kinds of errors, within a sparse Lindbladian parameterization for errors. To enable scaling to many qubit systems, these networks then predict how these errors combine within quantum circuits and impact their outcomes using an efficient approximations. We demonstrated these networks ability to learn coherent crosstalk errors and context-dependent errors in a simulated 4-qubit system.

97 MATHEMATICS AND COMPUTING↗

Target Recognition Using Neural Networks for Model Deformation Measurements

Optical measurements provide a non-invasive method for measuring deformation of wind tunnel models. Model deformation systems use targets mounted or painted on the surface of the model to identify known positions, and photogrammetric methods are used to calculate 3-D positions of the targets on the model from digital 2-D images. Under ideal conditions, the reflective targets are placed against a dark background and provide high-contrast images, aiding in target recognition. However, glints of light reflecting from the model surface, or reduced contrast caused by light source or model smoothness constraints, can compromise accurate target determination using current algorithmic methods. This paper describes a technique using a neural network and image processing technologies which increases the reliability of target recognition systems. Unlike algorithmic methods, the neural network can be trained to identify the characteristic patterns that distinguish targets from other objects of similar size and appearance and can adapt to changes in lighting and environmental conditions.

Ross, Richard W.↗

Super resolution convolutional neural network for feature extraction in spectroscopic data

Two dimensional (2D) peak finding is a common practice in data analysis for physics experiments, which is typically achieved by computing the local derivatives. However, this method is inherently unstable when the local landscape is complicated or the signal-to-noise ratio of the data is low. In this work, we propose a new method in which the peak tracking task is formalized as an inverse problem, which thus can be solved with a convolutional neural network (CNN). In addition, we show that the underlying physics principle of the experiments can be used to generate the training data. By generalizing the trained neural network on real experimental data, we show that the CNN method can achieve comparable or better results than traditional derivative based methods. This approach can be further generalized in different physics experiments when the physical process is known.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Neural Network Based Optimal Control: Resilience to Missed Thrust Events for Long Duration Transfers

A growing number of spacecraft are continuing to adopt new and more efficient forms of in-space propulsion. One shared characteristic of these high-efficiency propulsion techniques is their limited thrust capabilities. This requires them to thrust continuously for long periods of time, making the spacecraft susceptible to potential missed thrust events. This work demonstrates how neural networks can autonomously correct for missed thrust events during a low-thrust long-duration transfer trajectory. The research applies and tests the developed method to au- tonomously correct a Mars return trajectory. Additionally, methods for improving the response of neural networks to missed thrust events are presented and further investigated.

Laipert, Frank E.↗

Improved particle-flow event reconstruction with scalable neural networks for current and future particle detectors

Abstract Efficient and accurate algorithms are necessary to reconstruct particles in the highly granular detectors anticipated at the High-Luminosity Large Hadron Collider and the Future Circular Collider. We study scalable machine learning models for event reconstruction in electron-positron collisions based on a full detector simulation. Particle-flow reconstruction can be formulated as a supervised learning task using tracks and calorimeter clusters. We compare a graph neural network and kernel-based transformer and demonstrate that we can avoid quadratic operations while achieving realistic reconstruction. We show that hyperparameter tuning significantly improves the performance of the models. The best graph neural network model shows improvement in the jet transverse momentum resolution by up to 50% compared to the rule-based algorithm. The resulting model is portable across Nvidia, AMD and Habana hardware. Accurate and fast machine-learning based reconstruction can significantly improve future measurements at colliders.

Physics↗

Generation of synthetic dielectric dispersion logs in organic-rich shale formations using neural-network models

Dielectric dispersion (DD) logs acquired in subsurface geologic formations generally are composed of conductivity ([Formula: see text]) and relative permittivity ([Formula: see text]) measurements at four discrete frequencies in the range of 10 MHz to 1 GHz. Acquisition of DD logs in subsurface formations is operationally challenging, and it requires a hard-to-deploy infrastructure. We developed three supervised neural-network-based predictive methods to process conventional, easy-to-acquire subsurface logs for generating the eight DD logs acquired at four frequencies. These predictive methods will improve reservoir characterization in the absence of a DD logging tool. The predictive methods are tested in three wells intersecting organic-rich shale formations of the Permian Basin and the Bakken Shale. The first method predicts the eight dispersion logs simultaneously using a single artificial neural network (ANN) model, whereas the second method simultaneously predicts the four conductivity dispersion logs using one ANN model, followed by simultaneous prediction of four permittivity dispersion logs using a second ANN model. The third method sequentially predicts the eight dispersion logs, one at a time using eight sequential ANN models, based on a predetermined ranking of the prediction accuracy for each of the eight DD logs when simultaneously generated. Considering that the conventional and DD logs are recorded more than 10,000 ft deep in the subsurface using logging tools that are run at different times in rugose boreholes for sensing the near-wellbore geologic formation, the data used in this predictive work is prone to noise and biases that tend to adversely affect the prediction performances of the proposed methods. In terms of normalized root-mean square error (Nrms error), the prediction performances of the second predictive method are 8.5% worse and 6.2% better for the conductivity and permittivity dispersion logs, respectively, as compared with those of the first predictive method. The third method has best prediction performance for permittivity dispersion logs, which is 0.089 in terms of the Nrms error.

Geochemistry & Geophysics↗

High-Performance Computing Optimization for Aladyn – Adaptive Neural Network Molecular Dynamics Mini-Application

This report provides a description and performance evaluation of the optimization techniques for high performance computing (HPC) implementation of the open source Computational Materials mini-application Aladyn (https://github.com/nasa/aladyn). Aladyn is a basic molecular dynamics code written in FORTRAN 2003, which is designed to demonstrate the use of adaptive neural networks (ANNs) in atomistic simulations. The role of ANNs is to efficiently reproduce the very complex energy landscape resulting from the atomic interactions in materials with the accuracy of the more expensive quantum mechanics-based calculations. The ANN is trained on a large set of atomic structures calculated using the density functional theory (DFT) method. While achieving orders of magnitude faster computational performance than DFT, the ANN-based approach was still very computationally demanding compared to the conventional approach of using empirically fitted energy functions. After its initial development, Aladyn was evaluated and optimized by experts at the NASA Advanced Supercomputing (NAS) division to exploit modern supercomputer architectures. The code has been optimized for execution on multicore central processing units (CPUs), including Intel® Skylake microarchitecture, and on graphic accelerators, such as Nvidia® V100 graphic processing units (GPUs), using Open Multi-Processing (OpenMP) and Open Accelerators (OpenACC) programming interfaces. The optimization achieved a speedup of 4.7 times the baseline version on CPU performance and an additional 2.4 times on CPU+GPU performance. Atomistic computer simulations are a fundamental tool in materials research to model material properties form physics-based first principles. Atomic interaction, governed by Quantum Mechanics (QM) require sophisticated and highly computationally demanding mathematical models to calculate [1]. Classical methods use approximate functional forms, empirically fitted through a set of variable parameters to emulate atomic energies as direct functions of atomic coordinates [2]. While empirical potentials are computationally much simpler, allowing simulations of large-scale systems of up to a trillion (1012) atoms [3], they are substantially less accurate compared to quantum calculations and applicable only to very specific atomic configurations or predefined crystallographic phases. A recently suggested approach is to use heuristic machine learning methods [4], such as those based on Adaptive Neural Networks (ANNs) to predict atomic energies, after being trained on a sufficiently large database of QM-calculated structures [5,6]. This approach reduces significantly the computational complexity, allowing for simulations of orders of magnitude larger systems compared to QM-based methods without compromising accuracy. Still, compared to classical methods using empirical energy functions, ANN methods remain two- to three orders of magnitude more computationally demanding. Hence, the computational cost of simulations, together with the need for extensive training of ANNs, still makes the practical implementation of ANN-based methods quite challenging. The purpose of the Aladyn mini-application software [7], available as open source at https://github.com/nasa/aladyn, is to be a testbed for exploring possible optimization strategies to develop highly scalable parallel algorithms for ANN-based atomistic simulations. Aladyn is aimed at utilizing the architecture of the high-end modern highperformance computing (HPC) hardware based on multicore central processing units (CPUs) equipped with graphic processing unit (GPU) accelerators. Specifically, the goal is to optimize the performance on a single HPC compute node, before implementing scaling to multi-node parallelization using message passing interface (MPI). At the same time, the open source code of Aladyn can serve as a training model for students and professors in academia.

Yamakov, Vesselin I.↗

Global ocean dimethyl sulfide climatology estimated from observations and an artificial neural network

Marine dimethyl sulfide (DMS) is important to climate due to the ability of DMS to alter Earth's radiation budget. Knowledge of the global-scale distribution, seasonal variability, and sea-to-air flux of DMS is needed in order to improve understanding of atmospheric sulfur, aerosol/cloud dynamics, and albedo. Here we examine the use of an artificial neural network (ANN) to extrapolate available DMS measurements to the global ocean and produce a global climatology with monthly temporal resolution. A global database of 82 996 ship-based DMS measurements in surface waters was used along with a suite of environmental parameters consisting of latitude–longitude coordinates, time of day, time of year, solar radiation, mixed layer depth, sea surface temperature, salinity, nitrate, phosphate, and silicate. Linear regressions of DMS against the environmental parameters show that on a global-scale mixed layer depth and solar radiation are the strongest predictors of DMS. These parameters capture ~9 % and ~7 % of the raw DMS data variance, respectively. Multilinear regression can capture more of the raw data variance (~39 %) but strongly underestimates DMS in high-concentration regions. In contrast, the artificial neural network captures ~66 % of the raw data variance in our database. Like prior climatologies our results show a strong seasonal cycle in surface ocean DMS with the highest concentrations and sea-to-air fluxes in the high-latitude summertime oceans. We estimate a lower global sea-to-air DMS flux (20.12±0.43 Tg S yr –1 ) than the prior estimate based on a map interpolation method when the same gas transfer velocity parameterization is used. Our sensitivity test results show that DMS concentration does not change unidirectionally with each of the environmental parameters, which emphasizes the interactions among these parameters. The ANN model suggests that the flux of DMS from the ocean to the atmosphere will increase with global warming. Given that larger DMS fluxes induce greater cloud albedo, this corresponds to a negative climate feedback.

54 ENVIRONMENTAL SCIENCES↗

CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling

Large-scale simulations with complex electron interactions remain one of the greatest challenges for atomistic modelling. Although classical force fields often fail to describe the coupling between electronic states and ionic rearrangements, the more accurate ab initio molecular dynamics suffers from computational complexity that prevents long-time and large-scale simulations, which are essential to study technologically relevant phenomena. Here we present the Crystal Hamiltonian Graph Neural Network (CHGNet), a graph neural network-based machine-learning interatomic potential (MLIP) that models the universal potential energy surface. CHGNet is pretrained on the energies, forces, stresses and magnetic moments from the Materials Project Trajectory Dataset, which consists of over 10 years of density functional theory calculations of more than 1.5 million inorganic structures. The explicit inclusion of magnetic moments enables CHGNet to learn and accurately represent the orbital occupancy of electrons, enhancing its capability to describe both atomic and electronic degrees of freedom. We demonstrate several applications of CHGNet in solid-state materials, including charge-informed molecular dynamics in Li x MnO 2 , the finite temperature phase diagram for Li x FePO 4 and Li diffusion in garnet conductors. We highlight the significance of charge information for capturing appropriate chemistry and provide insights into ionic systems with additional electronic degrees of freedom that cannot be observed by previous MLIPs.

36 MATERIALS SCIENCE↗

Structure and Dynamics of CO 2 at the Air–Water Interface from Classical and Neural Network Potentials

The accurate description of the structure and dynamics of CO 2 at the instantaneous air–water interface, along with the effects of surface fluctuations on the CO 2 -transport processes, is essential for the development of negative emission technologies aimed at minimizing climate change. In this study, we performed molecular dynamics simulations of CO 2 at the air–water interface using neural network potentials (NNPs) trained on ab initio data generated through density-functional-theory-based molecular dynamics simulations. We compared these results with classical force fields to assess their performance in modeling interfacial CO 2 behavior. Our findings revealed that the asymmetric interactions, coupled with thermal surface fluctuations at the air–water interface, significantly influence CO 2 transport into the aqueous phase. The simulations demonstrate that classical force fields underestimate both the free energy of CO 2 transport and the strength of its interactions at the interface compared with the neural network potentials. In conclusion, the free energy and the interfacial dynamics of CO 2 are primarily influenced by the distribution of water within the instantaneous interfacial water layer, responsible for creating an asymmetric intermolecular interaction environment within the interfacial region.

Ab initio molecular dynamics↗

Fabrication-conscious neural network based inverse design of single-material variable-index multilayer films

Multilayer films with continuously varying indices for each layer have attracted great deal of attention due to their superior optical, mechanical, and thermal properties. However, difficulties in fabrication have limited their application and study in scientific literature compared to multilayer films with fixed index layers. In this work we propose a neural network based inverse design technique enabled by a differentiable analytical solver for realistic design and fabrication of single material variable-index multilayer films. This approach generates multilayer films with excellent performance under ideal conditions. We furthermore address the issue of how to translate these ideal designs into practical useful devices which will naturally suffer from growth imperfections. By integrating simulated systematic and random errors just as a deposition tool would into the optimization process, we demonstrated that the same neural network that produced the ideal device can be retrained to produce designs compensating for systematic deposition errors. Furthermore, the proposed approach corrects for systematic errors even in the presence of random fabrication imperfections. The results outlined in this paper provide a practical and experimentally viable approach for the design of single material multilayer film stacks for an extremely wide variety of practical applications with high performance.

36 MATERIALS SCIENCE↗

Virtual node graph neural network for full phonon prediction

Understanding the structure-property relationship is crucial for designing materials with desired properties. The past few years have witnessed remarkable progress in machine-learning methods for this connection. However, substantial challenges remain, including the generalizability of models and prediction of properties with materials-dependent output dimensions. Here we present the virtual node graph neural network to address the challenges. By developing three virtual node approaches, we achieve Γ-phonon spectra and full phonon dispersion prediction from atomic coordinates. We show that, compared with the machine-learning interatomic potentials, our approach achieves orders-of-magnitude-higher efficiency with comparable to better accuracy. This allows us to generate databases for Γ-phonon containing over 146,000 materials and phonon band structures of zeolites. Additionally, our work provides an avenue for rapid and high-quality prediction of phonon band structures enabling materials design with desired phonon properties. The virtual node method also provides a generic method for machine-learning design with a high level of flexibility. In this study, the authors present a virtual node graph neural network to enable the prediction of material properties with variable output dimensions. This method offers fast and accurate predictions of phonon band structures in complex solids.

36 MATERIALS SCIENCE↗

Inception Based Deep Convolutional Neural Network for Remaining Useful Life Estimation of Turbofan Engines

Accurate estimation of the remaining useful life (RUL) is a key component of condition based maintenance (CBM) and prognosis and health management (PHM). Data-based models for the estimation of RUL are of particular interest because expert knowledge of systems is not always available and physical modeling is often not feasible. In this paper, a deep convolutional neural network (CNN) architecture is investigated for its ability to estimate the RUL of turbofan engines. The input to the model is a window of time series data collected from the engine under test. Inputting raw sensor data allows features to be learned instead of manually determined. To incorporate the ability to detect features of differing lengths, inception modules are used in the neural network architecture. The model is trained and tested using the new Commercial Modular Aero-Propulsion System Simulation (N-CMAPSS) data set and high prognosis accuracy was achieved. The developed model was used in the 2021 PHM Society Data Challenge and received second place, further validating its ability to accurately estimate RUL.

DeVol, Nathaniel↗

Water, Energy, and Carbon with Artificial Neural Networks (WECANN): a statistically based estimate of global surface turbulent fluxes and gross primary productivity using solar-induced fluorescence

A new global estimate of surface turbulent fluxes, latent heat flux (LE) and sensible heat flux ( H ), and gross primary production (GPP) is developed using a machine learning approach informed by novel remotely sensed solar-induced fluorescence (SIF) and other radiative and meteorological variables. This is the first study to jointly retrieve LE, H , and GPP using SIF observations. The approach uses an artificial neural network (ANN) with a target dataset generated from three independent data sources, weighted based on a triple collocation (TC) algorithm. The new retrieval, named Water, Energy, and Carbon with Artificial Neural Networks (WECANN), provides estimates of LE, H , and GPP from 2007 to 2015 at 1° × 1° spatial resolution and at monthly time resolution. The quality of ANN training is assessed using the target data, and the WECANN retrievals are evaluated using eddy covariance tower estimates from the FLUXNET network across various climates and conditions. When compared to eddy covariance estimates, WECANN typically outperforms other products, particularly for sensible and latent heat fluxes. Analyzing WECANN retrievals across three extreme drought and heat wave events demonstrates the capability of the retrievals to capture the extent of these events. Uncertainty estimates of the retrievals are analyzed and the interannual variability in average global and regional fluxes shows the impact of distinct climatic events – such as the 2015 El Niño – on surface turbulent fluxes and GPP.

54 ENVIRONMENTAL SCIENCES↗

Photon Reconstruction in the Belle II Calorimeter Using Graph Neural Networks

We present the study of a fuzzy clustering algorithm for the Belle II electromagnetic calorimeter using Graph Neural Networks. We use a realistic detector simulation including simulated beam backgrounds and focus on the reconstruction of both isolated and overlapping photons. We find significant improvements of the energy resolution compared to the currently used reconstruction algorithm for both isolated and overlapping photons of more than 30% for photons with energies E γ < 0.5 GeV and high levels of beam backgrounds. Overall, the GNN reconstruction improves the resolution and reduces the tails of the reconstructed energy distribution and therefore is a promising option for the upcoming high luminosity running of Belle II.

Calorimeter↗

Acoustic Emission Source Location in Unidirectional Carbon-Fibre-Reinforced Plastic Plates Using Virtually Trained Artificial Neural Networks

Acoustic emission source location in a unidirectional carbon-fibre-reinforced plastic plate was attempted employing Artificial Neural Network (ANN) technology. The acoustic emission events were produced by a lead break, and the response wave received by piezoelectric sensors, type VS150-M resonant at 150 kHz. The waves were detected by a Vallen AMSY4 eight-channel instrumentation. The time of arrival, determined through the conventional threshold crossing technique, was used to measure the dependence of wave velocity on fibre orientation. A simple empirical formula, relying on classical lamination and suggested by wave propagation theory, was able to accurately model the experimental trend. Based on the formula, virtual training and testing data sets were generated for the case of a plate monitored by three transducers, and adopted to select two potentially effective ANN architectures. For final validation, experimental tests were carried out, positioning the source at predetermined points evenly distributed within the plate area. A very satisfactory correlation was found between the actual source locations and the ANN predictions.

36 MATERIALS SCIENCE↗