Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “learning dynamics”

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 577 records · Page 32
All sourcesNASA NTRSDOE OSTI
NASA NTRS2002

Measurement of Meteor Impact Experiments Using Three-Component Particle Image Velocimetry

The study of hypervelocity impacts has been aggressively pursued for more than 30 years at Ames as a way to simulate meteoritic impacts. Development of experimental methods coupled with new perspectives over this time has greatly improved the understanding of the basic physics and phenomenology of the impact process. These fundamental discoveries have led to novel methods for identifying impact craters and features in craters on both Earth and other planetary bodies. Work done at the Ames Vertical Gun Range led to the description of the mechanics of the Chicxualub crater (a.k.a. K-T crater) on the Yucatan Peninsula, widely considered to be the smoking gun impact that brought an end to the dinosaur era. This is the first attempt in the world to apply three-component particle image velocimetry (3-D PIV) to measure the trajectory of the entire ejecta curtain simultaneously with the fluid structure resulting from impact dynamics. The science learned in these experiments will build understanding in the entire impact process by simultaneously measuring both ejecta and atmospheric mechanics.

Heineck, James T.↗
NASA NTRS2018

Southern California Megacity CO2, CH4, and CO Flux Estimates Using Ground- and Space-Based Remote Sensing and a Lagrangian Model

We estimate the overall CO2, CH4, and CO flux from the South Coast Air Basin using an inversion that couples Total Carbon Column Observing Network (TCCON) and Orbiting Carbon Observatory-2 (OCO-2) observations, with the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model and the Open-source Data Inventory for Anthropogenic CO2 (ODIAC). Using TCCON data we estimate the direct net CO2 flux from the So-CAB to be 104±26 Tg CO2 yr(exp -1) for the study period of July 2013–August 2016. We obtain a slightly higher estimate of 120±30 Tg CO2 yr(exp -1) using OCO-2 data. These CO2 emission estimates are on the low end of previous work. Our net CH4 (360±90 Gg CH4 y(exp -1)) flux estimate is in agreement with central values from previous top-down studies going back to 2010 (342–440 Gg CH4 yr(exp -1)). CO emissions are estimated at 487±122 Gg CO yr(exp -1), much lower than previous top-down estimates (1440 Gg CO yr(exp -1)). Given the decreasing emissions of CO, this finding is not unexpected. We perform sensitivity tests to estimate how much errors in the prior, errors in the covariance, different inversion schemes, or a coarser dynamical model influence the emission estimates. Overall, the uncertainty is estimated to be 25%, with the largest contribution from the dynamical model. Lessons learned here may help in future inversions of satellite data over urban areas.

Total Carbon Column Observing Network (TCCON)↗
NASA NTRSDate not supplied

Noble Gases in Samples Returned From Asteroids Ryugu and Bennu

Introduction: The asteroid sample return missions, JAXA’s Hayabusa2 [1] and NASA’s OSIRIS-REx [2], fromC-rich asteroids Ryugu and Bennu offer unique opportunities to sample pristine early solar system material, unaffected by terrestrial weathering and contamination, and with a known geological context. The comparison with most primi-tive members of carbonaceous –and inner solar system ordinary and enstatite– chondrite classes allows us to (i) assess the variation (or the lack of thereof) of the distribution of the most volatile elements throughout the entire early solar system, (ii) determine the effects of parent body thermal and aqueous alteration, and (iii) learn about the dynamic evolution and history of the various parent bodies, as sampled by both return missions and meteorite delivery. Here we will present noble gas data obtained from newly received particles from Ryugu and Bennu. Samples & Analysis: We received nine particles allocated by JAXA, with masses of 0.5-1.7 mg mass: four col-lected during Hayabusa2’s first touchdown (chamber A) and five from a small artificial crater (chamber C). We received three particles from Bennu of 0.07, 0.89 and 0.94 mg mass (OREX-800032-102/3/4). All particles were processed, allocated, weighed, and transferred into our UHV sample chamber within pure N2, to prevent any contact with air. We examined all He-Xe isotopes with our custom-built mass spectrometer “Albatros” ([3,4]). Gases were extracted individually from each particle by melting induced by a 1064 μm Nd:YAG IR laser in 2–4 heating steps. Large amounts of water and other reactive volatiles were released, consistent with the presence of abundant volatiles from both Ryugu and Bennu. Residues of Ryugu particles remaining after lasering, subsequently exposed to air, were additionally extracted in a crucible to verify complete gas extraction. Only the particle that still showed a “crystalline” structure after lasering contained detectable Ryugu noble gases, whereas all other particles turned immediately into glassy spheres by laser heating and lacked any gas. Results & Discussion: Ryugu and Bennu matter contains noble gases similar to the most pristine, unheated aqueously altered meteorites, e.g., CM or CI chondrites [5,6]. Remarkably, all four chamber A samples that originated from Ryugu’s surface show solar wind (SW) Ne (trapped (20Ne/22Ne)tr ~12–13), whereas all five chamber C samples do not ((20Ne/22Ne)tr ~9.0–10.7 consistent with the presence of non-solar trapped Ne components such as Q and HL). This demonstrates that JAXA’s sampling strategy succeeded: Chamber C samples analyzed in this work originated from areas shielded from SW, slightly below the surface, that was sampled after the impactor excavated material. The three Bennu particles populate the same range as the Ryugu particles in Ne isotope space with one particle containing SW suggesting that it was part of the uppermost surface layer of Bennu during sampling, while Ne in the smallest sample can be explained best by mixing Ne from presolar diamonds and SiC or graphite, and that in the third one by mixing of, e.g., Q and HL. Using cosmic ray production rates given by [7] yields preliminary exposure ages for Ryugu particles in the same range as determined in [7]. There is, as expected, also no systematic difference between samples from chambers A and C because both sample sets were likely collected within the uppermost layer of Ryugu, which were not shielded from cosmic rays. The concentration of cosmogenic (cos) Ne in the Bennu particle, where Necos was resolvable, is in the same range as found for Ryugu, suggesting roughly similar exposure time to cosmic rays for 5–8 Ma for all particles examined here. The Xe isotopic compositions of all particles are consistent with those of phase Q, with small additions of Xe-HL from presolar diamonds and excess 129Xe* from short-lived 129I decay. The comparably high 129Xe*/129XeQ ratio is similar in all samples, suggesting similar initial I concentrations and closure to Xe loss –or the incorporation of a well-mixed Xe reservoir. The Bennu material was in brief contact with air in the capsule, after atmospheric entry and before storage in pure N2 in a clean room. Both, the –most diagnostic– Xe isotopes and the Ar/Xe vs. Kr/Xe systematics prove that the noble gases are not affected by terrestrial contamination, in contrast to many CI, CY, and CM chondrites found on Earth [5,6], illustrating the importance of sample return. He-Xe concentrations in all particles unaffected bySW are comparable to but at the upper end of ranges reported [5-7]. In accordance with the strong aqueous alteration experienced on the parent bodies of Ryugu and Bennu [1,2], all samples lack the Ar-rich and water-susceptible noble gas components found in less aqueously altered CMs or CRs [e.g., 5,8] and all COs.

Henner Busemann↗
DOE OSTI2021

Automating Discovery of Physics-Informed Neural State Space Models via Learning and Evolution

Recent works exploring deep learning application to dynamical systems modeling have demonstrated that embedding physical priors into neural networks can yield more effective, physically-realistic, and data-efficient models. However, in the absence of complete prior knowledge of a dynamical system's physical characteristics, determining the optimal structure and optimization strategy for these models can be difficult. In this work, we explore methods for discovering neural state space dynamics models for system identification. Starting with a design space of block-oriented state space models and structured linear maps with strong physical priors, we encode these components into a model genome alongside network structure, penalty constraints, and optimization hyperparameters. Demonstrating the overall utility of the design space, we employ an asynchronous genetic search algorithm that alternates between model selection and optimization and obtains accurate physically consistent models of three physical systems: an aerodynamics body, a continuous stirred tank reactor, and a two tank interacting system.

genetic algorithms, neural architecture search, ne↗
DOE OSTI2013

Probing sea quarks and gluons: the Electron-Ion Collider project

The 21st century holds great promise for reaching a new era for unlocking the mysteries of the structure of the atomic nucleus and the nucleons inside it governed by the theory of strong interactions (QCD). In particular, much remains to be learned about the dynamical basis of the structure of hadrons and nuclei in terms of the fundamental quarks and gluons. One of the main goals of existing and nearly completed facilities is to map out the spin flavor structure of the nucleons in the valence region. A future Electron-Ion Collider (EIC) would be the world's first polarized electron-proton collider, and the world's first e-A collider, and would seek the QCD foundation of nucleons and nuclei in terms of the sea quarks and gluons, matching to these valence quark studies. The EIC will provide a versatile range of kinematics and beam polarization, as well as beam species, to allow for mapping the spin and spatial structure of the quark sea and gluons, to discover the collective effects of gluons in atomic nuclei, and to understand the emergence of hadronic matter from color charge. I will summarize the physics goals and present an overview of the proposed EIC.

Horn, Tanja↗
DOE OSTI2026

Frameworks, Algorithms, and Scalable Technologies for Mathematics (FASTMath) SciDAC Institute

As computational models scale to larger computers, the rate at which they produce data has far outstripped the same computers ability to write that data and further the file systems ability to store that data. Almost all of the SciDAC applications, but especially those related to fusion solve very large scale PDEs whose scientific output his impacted by this problem. To gain access to dynamics in an exascale simulation that are not identifiable a priori and to make that dynamical data available to machine learning requires fundamental research in the area of in situ data data analytics. Here data analytics includes compression, visualization, uncertainty quantification, and machine learning. This in situ data analytics will enable on-the-fly spatial and temporal compression of solution dynamics, expose that space-time compressed field to machine learning algorithms that have been specialized to work with dynamically evolving data (existing machine learning algorithms treat data sets as static), greatly improving the opportunity for machine learning to provide feedback to the compression, all within an ongoing simulation, without the need to write data to files. The same concepts are also being applied to uncertainty quantification and multi-fidelity modeling which have similar needs for spatial and temporal compression of the ongoing exascale simulation to perform either without the typical, unacceptable writing of data to files.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1998

Mars Atmospheric Dynamics

The Martian atmosphere is dynamically similar to the Earth's. Its spin-axis rotation rate is only minutes longer than Earth's so the Coriolois force is nearly identical to Earth's. The inclination of its spin axis is also similar to Earth's giving it similarity in seasonal change. And the Martian atmosphere is nearly transparent to solar radiation (except during dust periods) such that it is heated primarily by upwelling infrared radiation from the surface. These characteristics make Mars an ideal laboratory for studying the dynamics of rapidly rotating differentially heated atmospheres. This talk reviews what we have learned about Mars atmospheric dynamics and how if compares with Earth. The source of information to make such a comparison comes from observations and models. The former are sparse and that the latter have played a major role in shaping our thinking about the general circulation on Mars. However, the models need validation. Fortunately, the first two orbiters in NASA's Mars Surveyor Program have instrumentation to address many of the issues related to the general circulation and climate of Mars. The first, Mars Global Surveyor, is already at Mars gathering data. The second, the Mars 98 Orbiter to be launched later this year, carries a dedicated atmospheric sounder. Thus, much will be learned about Mars' atmosphere in the next few years.

Haberle, Robert↗
DOE OSTI2020

Physics-constrained Deep Recurrent Neural Models of Building Thermal Dynamics

We develop physics-constrained and control-oriented predictive deep learning models for the thermal dynamics of a real-world commercial office building. The proposed method is based on the systematic encoding of physics-based prior knowledge into a structured recurrent neural architecture. Specifically, our model mimics the structure of the building thermal dynamics model and leverages penalty methods to model inequality constraints. Additionally, we use constrained matrix parameterization based on the Perron-Frobenius theorem to bound the eigenvalues of the learned network weights. We interpret the stable eigenvalues as dissipativeness of the learned building thermal model. We demonstrate the effectiveness of the proposed approach on a dataset obtained from an office building with $20$ thermal zones.

Building Energy, structured neural networks↗
DOE OSTI2025

Non-intrusive reduced-order modeling for dynamical systems with spatially localized features

This work presents a non-intrusive reduced-order modeling framework for dynamical systems with spatially localized features characterized by slow singular value decay. The proposed approach builds upon two existing methodologies for reduced and full-order non-intrusive modeling, namely Operator Inference (OpInf) and sparse Full-Order Model (sFOM) inference. We decompose the domain into two complementary subdomains that exhibit fast and slow singular value decay. The dynamics of the subdomain exhibiting slow singular value decay are learned with sFOM while the dynamics with intrinsically low dimensionality on the complementary subdomain are learned with OpInf. The resulting, coupled OpInf-sFOM formulation leverages the computational efficiency of OpInf and the high resolution of sFOM, and thus enables fast non-intrusive predictions for conditions beyond those sampled in the training data set. A novel regularization technique with a closed-form solution based on the Gershgorin disk theorem is introduced to promote stable sFOM and OpInf models. We also provide a data-driven indicator for subdomain selection and ensure solution smoothness over the interface via a post-processing interpolation step. We evaluate the efficiency of the approach in terms of offline and online speedup through a quantitative, parametric computational cost analysis. We demonstrate the coupled OpInf-sFOM formulation for two test cases: a one-dimensional Burgers’ model for which accurate predictions beyond the span of the training snapshots are presented, and a two-dimensional parametric model for the Pine Island Glacier ice thickness dynamics, for which the OpInf-sFOM model achieves an average prediction error on the order of 1% with an online speedup factor of approximately 8$\times$ compared to the numerical simulation.

42 ENGINEERING↗
DOE OSTI2020

Single Episode Policy Transfer in Reinforcement Learning

Transfer and adaptation to new unknown environmental dynamics is a key challenge for reinforcement learning (RL). An even greater challenge is performing near-optimally in a single attempt at test time, possibly without access to dense rewards, which is not addressed by current methods that require multiple experience rollouts for adaptation. To achieve single episode transfer in a family of environments with related dynamics, we propose a general algorithm that optimizes a probe and an inference model to rapidly estimate underlying latent variables of test dynamics, which are then immediately used as input to a universal control policy. This modular approach enables integration of state-of-the-art algorithms for variational inference or RL. Moreover, our approach does not require access to rewards at test time, allowing it to perform in settings where existing adaptive approaches cannot. In diverse experimental domains with a single episode test constraint, we report our method significantly outperforms existing adaptive approaches and shows favorable performance against baselines for robust transfer.

42 ENGINEERING↗
DOE OSTI2025

Machine Learning Enabled Advanced Electron Tomography for Resolving Chemical Inhomogeneity and Materials Dynamics in Lithium-Ion Battery Electrodes

The objective of this project is to develop machine learning-assisted electron microscopy, together with three-dimensional, cryogenic, and in-situ imaging techniques, for resolving chemical inhomogeneity and materials dynamics in lithium battery electrodes and interfaces. The project aims to expand the spatial, temporal, and dimensional resolution of transmission electron microscopy and to enable quantitative analysis of beam-sensitive battery materials.

25 ENERGY STORAGE↗
DOE OSTI2023

DeePMD-kit v2: A software package for deep potential models

DeePMD-kit is a powerful open-source software package that facilitates molecular dynamics simulations using machine learning potentials known as Deep Potential (DP) models. This package, which was released in 2017, has been widely used in the fields of physics, chemistry, biology, and material science for studying atomistic systems. The current version of DeePMD-kit offers numerous advanced features, such as DeepPot-SE, attention-based and hybrid descriptors, the ability to fit tensile properties, type embedding, model deviation, DP-range correction, DP long range, graphics processing unit support for customized operators, model compression, non-von Neumann molecular dynamics, and improved usability, including documentation, compiled binary packages, graphical user interfaces, and application programming interfaces. This article presents an overview of the current major version of the DeePMD-kit package, highlighting its features and technical details. Additionally, this article presents a comprehensive procedure for conducting molecular dynamics as a representative application, benchmarks the accuracy and efficiency of different models, and discusses ongoing developments.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Transfer learning of memory kernels for transferable coarse-graining of polymer dynamics

The present work concerns the transferability of coarse-grained (CG) modeling in reproducing the dynamic properties of the reference atomistic systems across a range of parameters. In particular, we focus on implicit-solvent CG modeling of polymer solutions. The CG model is based on the generalized Langevin equation, where the memory kernel plays the critical role in determining the dynamics in all time scales. Thus, we propose methods for transfer learning of memory kernels. The key ingredient of our methods is Gaussian process regression. By integration with the model order reduction via proper orthogonal decomposition and the active learning technique, the transfer learning can be practically efficient and requires minimum training data. Through two example polymer solution systems, we demonstrate the accuracy and efficiency of the proposed transfer learning methods in the construction of transferable memory kernels. The transferability allows for out-of-sample predictions, even in the extrapolated domain of parameters. Built on the transferable memory kernels, the CG models can reproduce the dynamic properties of polymers in all time scales at different thermodynamic conditions (such as temperature and solvent viscosity) and for different systems with varying concentrations and lengths of polymers.

Ma, Zhan↗
DOE OSTI2024

Molecular simulation using transfer-learned potentials for the disordered nanoscale structure of nitrogen-doped nanoporous carbons

Machine learning (ML)-based molecular dynamics (MD) simulations of the formation of a class of N-doped nanoporous carbons are performed to assess their disordered partially graphitized nanoscale structure. The study is motivated by the effectiveness of so-called nitrogen assembly carbons (NACs) for catalysis applications. Benchmark simulations for pure-C disordered graphitic systems reveal the importance of reliably capturing the vdW component of the potentials in order to accurately describe the tendency for layering of disordered graphene-like sheets. In our modeling, this is achieved by a transfer learning strategy incorporating features of the energetics from the optB88-vdW DFT functional into potentials initially trained with a less expensive functional, thereby providing a superior description of the pure-C systems. Generation from MD simulations of realistic partially graphitized structures is significantly more challenging for N-doped versus for pure C systems. However, such structures are achieved by a tailored MD simulation protocol mimicking the experimental synthesis process and in particular incorporating an annealing and subsequent quenching stages. Simulated PXRD patterns effectively reproduce the features of experimental observations for NACs, including the appearance of a prominent but broad (002) peak at around 25, and the development of another weaker feature associated with in-layer ordering of mixed C-N graphene-like sheets.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

Enforcing constraints for time series prediction in supervised, unsupervised and reinforcement learning

We assume that we are given a time series of data from a dynamical system and our task is to learn the flow map of the dynamical system. We present a collection of results on how to enforce constraints coming from the dynamical system in order to accelerate the training of deep neural networks to represent the flow map of the system as well as increase their predictive ability. In particular, we provide ways to enforce constraints during training for all three major modes of learning, namely supervised, unsupervised and reinforcement learning. In general, the dynamic constraints need to include terms which are analogous to memory terms in model reduction formalisms. Such memory terms act as a restoring force which corrects the errors committed by the learned flow map during prediction. For supervised learning, the constraints are added to the objective function. For the case of unsupervised learning, in particular generative adversarial networks, the constraints are introduced by augmenting the input of the discriminator. Finally, for the case of reinforcement learning and in particular actor-critic methods, the constraints are added to the reward function. In addition, for the reinforcement learning case, we present a novel approach based on homotopy of the action-value function in order to stabilize and accelerate training. We use numerical results for the Lorenz system to illustrate the various constructions.

Stinis, Panagiotis↗
DOE OSTI2024

Chaotic neural dynamics facilitate probabilistic computations through sampling

Cortical neurons exhibit highly variable responses over trials and time. Theoretical works posit that this variability arises potentially from chaotic network dynamics of recurrently connected neurons. Here, we demonstrate that chaotic neural dynamics, formed through synaptic learning, allow networks to perform sensory cue integration in a sampling-based implementation. We show that the emergent chaotic dynamics provide neural substrates for generating samples not only of a static variable but also of a dynamical trajectory, where generic recurrent networks acquire these abilities with a biologically plausible learning rule through trial and error. Furthermore, the networks generalize their experience in the stimulus-evoked samples to the inference without partial or all sensory information, which suggests a computational role of spontaneous activity as a representation of the priors as well as a tractable biological computation for marginal distributions. These findings suggest that chaotic neural dynamics may serve for the brain function as a Bayesian generative model.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2020

Machine Learning-Based Upscaling of Finite-Size Molecular Dynamics Diffusion Simulations for Binary Fluids

Molecular diffusion coefficients calculated using molecular dynamics (MD) simulations suffer from finite-size (i.e., finite box size and finite particle number) effects. Results from finite-sized MD simulations can be upscaled to infinite simulation size by applying a correction factor. For self-diffusion of single-component fluids, this correction has been well-studied by many researchers including Yeh and Hummer (YH); for binary fluid mixtures, a modified YH correction was recently proposed for correcting MD-predicted Maxwell–Stephan (MS) diffusion rates. In this study we use both empirical and machine learning methods to identify improvements to the finite-size correction factors for both self-diffusion and MS diffusion of binary Lennard-Jones (LJ) fluid mixtures. Using artificial neural networks (ANNs), the error in the corrected LJ fluid diffusion is reduced by an order of magnitude versus existing YH corrections, and the ANN models perform well for mixtures with large dissimilarities in size and interaction energies where the YH correction proves insufficient.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗