Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “distributed computing methodologies”

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 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
DOE OSTI2020

ICME and In-Situ Process Monitoring for Rapid Qualification of Components Made by Laser-based Powder Bed Additive Manufacturing Processes for Nuclear Structural Applications

Additive manufacturing (AM) through selective powder bed melting of successive layers is being considered as disruptive technology for rapid production of low-cost nuclear reactor internal components with complex geometries. However, there is a potential for non-uniform distribution of physical features, such as porosity or microstructural differences, due to variations of temperature across a component build. These heterogeneities make the deployment of traditional material qualification and non-destructive evaluation of AM components difficult. This research project explored the feasibility of using in-situ process monitoring methods and integrated computational materials engineering (ICME) principles as an alternate qualification methodology and approach. This project included six individual tasks: (i) design of artifacts relevant to the nuclear power industry, (ii) evaluation of laser processing and in-situ measurements, (iii) computational modeling, (iv) ex-situ microstructural characterization, (v) evaluation of scaling the methodology for large-scale structures, and (vi) development of a data package to codes and standards organizations. As a part of this research, laser powder bed fusion of metals (316L, Alloy 718, and Ti6Al4V) was explored. This document provides an overview of the research performed over three years and serves as the final U.S. Department of Energy report for this project. In the third year, the project focused on the following activities: (a) in-situ infrared and optical imaging of each layer during processing of representative stainless steel geometries; (b) development of methodologies to analyze the data; and (c) application of ICME methodologies and other experiments to estimate the relevance of defects and microstructure to tensile properties.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗
DOE OSTI2020

ICME and In-Situ Process Monitoring for Rapid Qualification of Components Made by Laser-based Powder Bed Additive Manufacturing Processes for Nuclear Structural Applications

Additive manufacturing (AM) through selective powder bed melting of successive layers is being considered as disruptive technology for rapid production of low-cost nuclear reactor internal components with complex geometries. However, there is a potential for non-uniform distribution of physical features, such as porosity or microstructural differences, due to variations of temperature across a component build. These heterogenities make the deployment of traditional material qualification and non-destructive evaluation of AM components difficult. This research project explored the feasibility of using in-situ process monitoring methods and integrated computational materials engineering (ICME) principles as an alternate qualification methodology and approach. This project included six individual tasks: (i) design of artifacts relevant to the nuclear power industry, (ii) evaluation of laser processing and in-situ measurements, (iii) computational modeling, (iv) ex-situ microstructural characterization, (v) evaluation of scaling the methodology for large-scale structures, and (vi) development of a data package to codes and standards organizations. As a part of this research, laser powder bed fusion of metals (316L, Alloy 718, and Ti6Al4V) was explored. This document provides an overview of the research performed over three years and serves as the final U.S. Department of Energy report for this project. In the third year, the project focused on the following activities: (a) in-situ infrared and optical imaging of each layer during processing of representative stainless steel geometries; (b) development of methodologies to analyze the data; and (c) application of ICME methodologies and other experiments to estimate the relevance of defects and microstructure to tensile properties.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗
DOE OSTI2020

ICME and In-Situ Process Monitoring for Rapid Qualification of Components Made by Laser-based Powder Bed Additive Manufacturing Processes for Nuclear Structural Applications

Additive manufacturing (AM) through selective powder bed melting of successive layers is being considered as disruptive technology for rapid production of low-cost nuclear reactor internal components with complex geometries. However, there is a potential for non-uniform distribution of physical features, such as porosity or microstructural differences, due to variations of temperature across a component build. These heterogenities make the deployment of traditional material qualification and non-destructive evaluation of AM components difficult. This research project explored the feasibility of using in-situ process monitoring methods and integrated computational materials engineering (ICME) principles as an alternate qualification methodology and approach. This project included six individual tasks: (i) design of artifacts relevant to the nuclear power industry, (ii) evaluation of laser processing and in-situ measurements, (iii) computational modeling, (iv) ex-situ microstructural characterization, (v) evaluation of scaling the methodology for large-scale structures, and (vi) development of a data package to codes and standards organizations. As a part of this research, laser powder bed fusion of metals (316L, Alloy 718, and Ti6Al4V) was explored. This document provides an overview of the research performed over three years and serves as the final U.S. Department of Energy report for this project. In the third year, the project focused on the following activities: (a) in-situ infrared and optical imaging of each layer during processing of representative stainless steel geometries; (b) development of methodologies to analyze the data; and (c) application of ICME methodologies and other experiments to estimate the relevance of defects and microstructure to tensile properties.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2024

Assessment of the Impact of Realistic Sensor Physics and the Integration of Ex-Core Sensors on Reactor Power Synthesis

In the work documented in this report, a weighting function–based core power synthesis method was applied to multiple Monte Carlo N-Particle (MCNP) reactor models, which are informed based on simulated self-powered neutron detector (SPND) responses. The weighting function method used has been coined the point-based iterative (PBI) method. The goal of this application is to assess the impact of considering realistic sensor physics in the generation of the simulated SPND outputs as well as to consider how the synthesis is impacted based on the inclusion of ex-core detectors in the model. The NuScale small modular reactor (SMR) and Westinghouse AP1000 pressurized water reactor (PWR) are the models that served as the testbeds for the assessment of realistic sensor physics; this was achieved by using Geant4 SPND models in comparison with analytical models, such that the effect of electron transport in realistic SPND geometries in the Geant4 model can be understood in terms of synthesis error and convergence time. The comparison was considered for fuel burnup–induced perturbations, for a range of sensor string densities and synthesized power distribution axial fidelities. The Texas A&M Testing, Research, Isotopes, General Atomics Reactor (TAMU TRIGA) reactor MCNP model was used to assess the impact of ex-core sensors; this was done by performing synthesis with and without the ex-core detectors and by quantifying the synthesis error and number of iterations associated with Gaussian-type perturbations in many locations in the core. The TAMU TRIGA model was particularly pertinent for this study because of the interest in future experimental tests with SPNDs in this reactor, as well as the ease of modifying the MCNP model to include ex-core detectors with heterogeneously described response functions. Results from the comparison between the Geant4 and analytical SPND models indicate that similar average and maximum synthesis errors were obtained for burnup-induced perturbations in both the NuScale SMR and the AP1000. This was true for a range of sensor string densities and axial fidelities. However, there were marked differences between both the Geant4 and analytically informed models in terms of the iterations required to converge on the synthesized power distribution. Namely, the Geant4-informed models tended to lead to fewer iterations, except for a few sensor–core configurations that had particularly numerous iterations. Results from the ex-core sensor assessment with the TAMU TRIGA model indicate that the inclusion of ex-core sensors drastically reduces the synthesis error of Gaussian-type perturbations close to the edge of the core, and it slightly reduces synthesis errors for perturbations closer to the center of the core. This was achieved with a minimal increase in computational cost—that is, the number of iterations required for convergence. The errors were identified to be in the same location as the perturbation in the core, indicating that the methodology remains robust for unperturbed regions of the core. A secondary result from this study with the TAMU TRIGA was yielded by analysis of the neutron flux levels in the in-core and ex-core sensor locations of the core; these flux levels indicate that SPNDs could be used as both in-core and ex-core sensors, so long as the emitter material is sensitive to thermal neutrons. The results from these studies provide a quantitative understanding of the importance of considering realistic sensor physics and including ex-core sensors to perform accurate and timely power distribution synthesis of a reactor core.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2026

Boron-Based Neutron Scintillator Screen Characterization with X-Rays and Neutrons

Recent work on boron-based neutron scintillator screens suggests these screens can offer superior performance when compared to commonly used screens. Borated neutron scintillator screens perform well in terms of light output (5-6 times greater than a standard Gadox screen) and detection effi-ciency (larger than standard LiF+ZnS screens). However, previously manu-factured boron-based screens have exhibited non-uniform surface coating and a poor mixture between phosphor and converter particles. The objective of this work was to evaluate newly fabricated scintillator screens to deter-mine if enhanced fabrication methods produced a more homogeneous distribution between neutron converter and scintillation phosphor particles. Uniformity of scintillator material deposition was also inspected. This new iteration of screens appeared more uniform than previous generations with the new coating method improving surface chemistry and scintillator material homogeneity. Additionally, a new methodology for screen characterization, involving the correlation of a neutron image taken with a borated scintillator screen to X-ray computed tomography of that same screen, was demonstrated to elucidate a relationship between scintillator screen thickness and relative light output of the screen under neutron exposure. This method suggested that the ideal thickness of scintillator material was ~150 µm to maximize light output of the screen.

36 - MATERIALS SCIENCE↗
DOE OSTI2023

Validation and Verification of TEDS Facility HYBRID Modeling

The HYBRID modeling repository is an in-house developed library of models for selected integrated energy systems (IES) modelling. HYBRID models have been developed since 2015 to describe the physical operation of tightly coupled thermal systems including power generators, thermal transport systems, thermal storage, thermal-to-electric conversion systems, and other thermal applications. Here, validation and verification (V&V) capabilities are demonstrated using the Thermal Energy Distribution System (TEDS) at INL. Building upon prior work, the TEDS model has been updated and verified so that it better represents the installed system configuration and the operating control system. The model control system was changed to allow replication of actual experimental procedures. Experimental operations focusing primarily on thermocline tank performance were devised and performed. Several anomalies were found in the operation data of the experiment facility. V&V activities calibrating a selected input parameter are demonstrated on a single component as well as with a single parameter within the thermocline. Calibrating is then demonstrated on multiple components and a multi-parameter metric for the entire system. The validation methodology is successfully applied to validate the model with experimental data. It is also used to confirm a hypothesis behind one of the anomalies in experimental performance.

25 ENERGY STORAGE↗
DOE OSTI2022

Modeling electrokinetic flows with the discrete ion stochastic continuum overdamped solvent algorithm

In this article we develop an algorithm for the efficient simulation of electrolytes in the presence of physical boundaries. In previous work the discrete ion stochastic continuum overdamped solvent (DISCOS) algorithm was derived for triply periodic domains, and was validated through ion-ion pair correlation functions and Debye-Hückel-Onsager theory for conductivity, including the Wien effect for strong electric fields. In extending this approach to include an accurate treatment of physical boundaries we must address several important issues. First, the modifications to the spreading and interpolation operators necessary to incorporate interactions of the ions with the boundary are described. Next we discuss the modifications to the electrostatic solver to handle the influence of charges near either a fixed potential or dielectric boundary. An additional short-ranged potential is also introduced to represent interaction of the ions with a solid wall. Finally, the dry diffusion term is modified to account for the reduced mobility of ions near a boundary, which introduces an additional stochastic drift correction. Herein, several validation tests are presented confirming the correct equilibrium distribution of ions in a channel. Additionally, the methodology is demonstrated using electro-osmosis and induced-charge electro-osmosis, with comparison made to theory and other numerical methods. Notably, the DISCOS approach achieves greater accuracy than a continuum electrostatic simulation method. We also examine the effect of under-resolving hydrodynamic effects using a “dry diffusion” approach, and find that considerable computational speedup can be achieved with a negligible impact on accuracy.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Selecting Critical Scenarios of DER Adoption in Distribution Grids Using Bayesian Optimization

We develop a new methodology to select scenarios of DER adoption most critical for distribution grids. Anticipating risks of future voltage and line flow violations due to additional PV adopters is central for utility investment planning but continues to rely on deterministic or ad hoc scenario selection. We propose a highly efficient search framework based on multi-objective Bayesian Optimization. We treat underlying grid stress metrics as computationally expensive black-box functions, approximated via Gaussian Process surrogates and design an acquisition function based on probability of scenarios being Pareto-critical across a collection of line- and bus-based violation objectives. Our approach provides a statistical guarantee and offers an order of magnitude speed-up relative to a conservative exhaustive search. Case studies on realistic feeders with 200-400 buses demonstrate the effectiveness and accuracy of our approach.

Mulkin, Olivier↗
DOE OSTI2023

An Integrated Approach to Predicting Ash Deposition and Heat Transfer in Coal-Fired Boilers

The overall goal of this project is to develop via measurements and simulations an advanced online technology to predict, monitor and manage fireside ash deposition in a coal-fired boiler allowing for more efficient operations under a range of load conditions and fuel property variability. With this in place fuel sorting and blending can be done upstream and operations can be optimized to compensate for load and fuel properties. In support of this objective, three experimental campaigns were undertaken during the course of the project to measure ash deposition rates within the boiler at different fuel flow rates and its ash composition variability. Simulations of the experimental conditions representing actual geometry, operational scenarios in terms of air flow rates, coal flow rates as well as coal compositions, heating values, and particle size distributions were also carried out. Deposition rates were predicted using a unique particle kinetic energy and viscosity based ash deposition methodology whose validity was ascertained by comparing against deposition rate measurements for widely varying operating conditions and ash compositions in a lab-scale furnace. With a unique end-to-end combustion modeling methodology established and different simulation scenarios carried out, the results from our computational fluid dynamic (CFD) simulations in conjunction with the plant data summarized in this report were used to refine Microbeam Technology Incorporated’s MTI CSPI-CT Tool to predict and monitor fire-side ash deposition under a range of load conditions and fuel property variability in real time.

01 COAL, LIGNITE, AND PEAT↗
DOE OSTI2025

Heterogeneous fatigue damage in a nickel-based single-crystal superalloy unraveled using correlative 3D X-ray technology

Nickel-based single-crystal (Ni-SX) superalloys under cyclic stress are susceptible to cracking at stress-concentration sites, eventually leading to low-cycle fatigue (LCF) failure. LCF cracks typically originate from intrinsic defects (e.g., voids and carbides) within solidified dendrites. However, systematic quantitative experimental analyses of defect-mediated local damage remain limited. To thoroughly understand the microscopic origins and evolution of LCF damage, correlated 3D mapping of dendrites across various regions is essential. Here, in this study, macroscale micro-computed tomography (μ-CT) was initially used to capture internal interdendritic secondary cracks within bulk DD413 superalloy after LCF testing at 760 °C. Subsequently, a multimodal methodology combining synchrotron 3D microdiffraction (3D-μXRD), high-resolution μ-CT, and electron microscopy was established. This approach allowed precise localization of internal damage zones near interdendritic secondary cracks and detailed mapping of the 3D correlated distributions of dendrites, defects, and residual stress/strain fields within these zones at submicron spatial resolution. Finally, the same approach was applied to specimens subjected to interrupted loading at approximately 40 % of the fatigue life to uncover the early damage states of dendrites. The dendrite cores (DCs) and interdendritic regions (IDs) exhibit microscale heterogeneous mechanical responses: nearly defect-free DCs accumulate local irreversible slip along specific slip systems to generate slip bands, while the IDs containing various defects accommodate local microplasticity through the activation of multiple slip systems around these defects. The local tensile stress near defects in the IDs exceeds that in the DC slip band regions by more than threefold, leading to the generation of local damage zones within the IDs. Chain-like defect distributions facilitate the interconnection of these local zones into a continuous damage region, further elevating the overall tensile stress in the IDs. Additionally, geometrically necessary dislocations alone are insufficient as indicators of LCF damage; both the internal stress state and its magnitude must be considered. These experimental results provide critical data and insights for the development of multi-physics fatigue models.

Localized deformation↗
DOE OSTI2020

Non-destructive simulation of node defects in additively manufactured lattice structures

Additive Manufacturing (AM), commonly referred to as 3D printing, offers the ability to not only fabricate geometrically complex lattice structures but parts in which lattice topologies in-fill volumes bounded by complex surface geometries. However, current AM processes produce defects on the strut and node elements which make up the lattice structure. This creates an inherent difference between the as-designed and as-fabricated geometries, which negatively affects predictions (via numerical simulation) of the lattice’s mechanical performance. Although experimental and numerical analysis of an AM lattice’s bulk structure, unit cell and struts have been performed, there exists almost no research data on the mechanical response of the individual as-manufactured lattice node elements. Here we propose a methodology that, for the first time, allows non-destructive quantification of the mechanical response of node elements within an as-manufactured lattice structure. A custom-developed tool is used to extract and classify each individual node geometry from micro-computed tomography scans of an AM fabricated lattice. Voxel-based finite element meshes are generated for numerical simulation and the mechanical response distribution is compared to that of the idealised computer-aided design model. The method demonstrates compatibility with Uncertainty Quantification methods that provide opportunities for efficient prediction of a population of nodal responses from sampled data. Overall, the non-destructive and automated nature of the node extraction and response evaluation is promising for its application in qualification and certification of additively manufactured lattice structures.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Evaluation of Converter Performance Considering Static and Dynamic Device Part-to-Part Variability

This paper presents a methodology to incorporate and analyze the impact of semiconductor device part-to-part variation on power converter performance. By integrating extensive static and dynamic device characterization data with an automated compact model generation process that reflects manufacturing variability, device models with inherent variability features are utilized in converter simulations for a comprehensive assessment of performance impacts. The traditional converter performance evaluation process typically yields fixed efficiency values, often dismissing the inherent part-to-part variability caused by the manufacturing process of semiconductor devices. To address this limitation, a large population of devices was characterized to capture variations in static parameters-such as transfer, output, and capacitance characteristics-as well as dynamic behaviors, including switching losses. This data-driven approach enables the development of individual compact models, which were then integrated into converter simulations to evaluate efficiency ranges rather than single point estimated values. The converter simulation results show that part-to-part component variation can lead to significant efficiency deviations, exceeding several percentage points in high-power conversion applications. By offering a more accurate representation of converter behavior under real-world manufacturing conditions, this methodology enables designers to anticipate performance variability, improving the robustness of power converter designs.

device characterization↗
DOE OSTI2022

Verification of the DIF3D Software to Support Fast Reactor Analysis

Ongoing design activities at Argonne National Laboratory are requiring a thorough verification of the Argonne Reactor Computation codes be performed. DIF3D is central to this system. The driver for this effort requires the 3D Cartesian, triangular-Z, and hexagonal-Z core geometry options of DIF3D be verified. Previous work identified the DIF3D features required to be verified to support current design activities, features of which are generally applicable to hexagonal-Z fast reactor designs. The scope of this verification effort includes verifying DIF3D’s ability to correctly translate the user’s model in to DIF3D’s preferred format, verifying that options planned for use have the desired effect, and verifying the correctness of the eigenvalue, fixed-source, forward, and adjoint solvers in DIF3D-FD and DIF3D-VARIANT. This manuscript provides the verification tasks and their results with respect to the features needed for current design activities. Since analytic solutions of the neutron diffusion and transport equations are either limited in scope or not possible, multiple tiers of problems unique to each solver and geometry type were implemented. Each of these tiers tests features independent and complementary arguments for why the separate testing of functionalities is acceptable. Finally, this separate testing was also supplemented with a high-level integral check of each the diffusion and transport capabilities and applicable geometries. To accommodate cases which an analytic solution is not feasible, MCNP6.2 was relied upon to provide a higher-order reference solution. This therefore required that the capabilities within MCNP6.2 which were relied upon for this work are also verified in this work. No MCNP discrepancies were noted in this effort. Note that the MCNP6.2 verification included in this work does not stand as a full verification of MCNP6.2, but merely verifies the features used in verifying DIF3D. The verification effort identified no issues that are debilitating or otherwise impactful to design usage of DIF3D, and thus DIF3D version 11.0, release 3012 is considered verified. As some additional changes have been made to the ARC software since this point all versions between release 3012 and 3266 can be considered verified as version 3253 was used for all updates in this revision. The types of issues that were identified were predominantly in the areas of: unclear documentation, software bugs which were inconsequential to final results, editing options which were ignored in favor of printing more information than requested, bugs in the outputs of intermediate results, or secondary output binary file information which was not present. While not a bug, this verification report also identified that the algorithm used to evaluate the peak fast flux in a nodal transport solution can be quite unreliable due to the methodology used and the location of the peak within the mesh. The authors of the report therefore recommend the usage of the EvaluateFlux software (distributed with ARC) as a more robust alternative noting that DIF3D will properly notify the user when the peaking values it is providing are potentially incorrect.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

An extended methodology for automated calculations of non-Boltzmann kinetic sequences: H + C 2 H 2 + X and combustion impact

It is generally assumed in phenomenological kinetic models that bimolecular reactions only occur between species whose rovibrational energy follows a Boltzmann (thermal) distribution. That is, any complexes initially formed in non-Boltzmann distributions are assumed to be thermalized by energy-transferring collisions prior to bimolecular reactions. Given the high mole fractions of reactive species, X, in combustion environments, reactive collisions of the complexes with X often occur on the same timescale as energy-transferring collisions – yielding sequences proceeding through non-Boltzmann intermediates across multiple potential energy surfaces. Recent studies have shown that such non-Boltzmann kinetic sequences can have substantial impact on the global reactivity in combustion systems. Simulations of these non-Boltzmann reaction sequences, which can be described in phenomenological kinetic models via chemically termolecular reactions, require that rovibrational excitation from one potential energy surface be carried over to the next. Here this paper presents an extended theoretical and computational methodology that couples multiple master equations and derives rate constants for phenomenological reactions describing the conversion of thermal reactants to thermal products for use in phenomenological kinetic schemes. The methodology is then implemented using in-house scripts for non-Boltzmann sequences involving C2H$^*_3$ + X (with X = O 2 , H, and OH) where C 2 H$^*_3$ is formed via H + C 2 H 2 association – which were identified as having strong potential for influencing combustion predictions in a recent study. The results reveal that non-Boltzmann reaction sequences for X = O 2 (the primary focus of this paper) significantly alters the total conversion rate from H + C 2 H 2 to products and product branching fractions from those of thermal sequential pathways. Furthermore, the present results demonstrate that non-Boltzmann reaction sequences have significant impact – as high as an order of magnitude – on predicted ignition delay times. Similarly, they yield significantly different dependence of ignition delay times with temperature and O 2 mole fraction – yielding signatures that are likely observable experimentally.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Data-driven surrogate modeling of hPIC ion energy-angle distributions for high-dimensional sensitivity analysis of plasma parameters' uncertainty

In this work, we present a data-driven strategy for effective construction of a surrogate model in high-dimensional parameter space for the ion energy-angle distribution (IEAD) output of hPIC simulations of plasma-surface interactions. The methodology is based on a bin-by-bin least-squares fitting of the IEAD in the parameter space. The fitting is performed in a transformed coordinate system to normalize the IEAD, and it employs sparse grids for sampling the parameter space to overcome sampling challenges in high dimensions. The surrogate model is significantly cheaper computationally than direct hPIC simulations yet maintains high fidelity to them, providing a fast emulator for hPIC simulations. Sensitivity analysis based on the surrogate model is utilized to characterize the dependence of the ion impact angle and energy moments on the physical parameters.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

Robust Machine Learning Inference from X-ray Absorption Near Edge Spectra through Featurization

X-ray absorption spectroscopy (XAS) is a commonly employed technique for characterizing functional materials. In particular, X-ray absorption near edge spectra (XANES) encode local coordination and electronic information, and machine learning approaches to extract this information are of significant interest. To date, most ML approaches for XANES have primarily focused on using the raw spectral intensities as input, overlooking the potential benefits of incorporating spectral transformations and dimensionality reduction techniques into ML predictions. Here, in this work, we focused on systematically comparing the impact of different featurization methods on the performance of ML models for XAS analysis. We evaluated the classification and regression capabilities of these models on computed data sets and validated their performance on previously unseen experimental data sets. Our analysis revealed an intriguing discovery: the cumulative distribution function feature achieves both high prediction accuracy and exceptional transferability. This remarkably robust performance can be attributed to its tolerance to horizontal shifts in the spectra, which is crucial when validating models using experimental data. While this work exclusively focuses on XANES analysis, we anticipate that the methodology presented here will hold promise as a versatile asset to the broader spectroscopy community.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Hvac: Removing I/O Bottleneck for Large-Scale Deep Learning Applications

Scientific communities are increasingly adopting deep learning (DL) models in their applications to accelerate scientific discovery processes. However, with rapid growth in the computing capabilities of HPC supercomputers, large-scale DL applications have to spend a significant portion of training time performing I/O to a parallel storage system. Previous research works have investigated optimization techniques such as prefetching and caching. Unfortunately, there exist non-trivial challenges to adopting the existing solutions on HPC supercomputers for large-scale DL training applications, which include non-performance and/or failures at extreme scale, lack of portability and generality in design, complex deployment methodology, and being limited to a specific application or dataset. To address these challenges, we propose High-Velocity AI Cache (HVAC), a distributed read-cache layer that targets and fully exploits the node-local storage or near node-local storage technology. HVAC seamlessly accelerates read I/O by aggregating node-local or near node-local storage, avoiding metadata lookups and file locking while preserving portability in the application code. We deploy and evaluate HVAC on 1,024 nodes (with over 6000 NVIDIA V100 GPUS) of the Summit supercomputer. In particular, we evaluate the scalability, efficiency, accuracy, and load distribution of HVAC compared to GPFS and XFS-on-NVMe. With four different DL applications, we observe an average 25 % performance improvement atop GPFS and 9% drop against XFS-on-NVMe, which scale linearly and are considered the performance upper bound. We envision HVAC as an important caching library for upcoming HPC supercomputers such as Frontier.

Khan, Awais↗
DOE OSTI2020

Quantifying leakage fields at ionic grain boundaries using off-axis electron holography

The electrical properties of interfaces in semiconductors and ionic conductors are immensely important in a wide range of applications. Electron holography is ideally suited for the direct measurement of the electrostatic potential of such interfaces. A key challenge with this approach is the contribution of the leakage field from the sample to the observed electron phase shift. This leakage field cannot be a priori independently determined and can cause an overestimation of the phase shift. In this work, we use finite element simulations to compute the three-dimensional electrostatic potential in the vicinity of an interface associated with a given interfacial charge density distribution. We then evaluate the predicted phase shift and demonstrate that the leakage field strongly affects the recovery of the projected interface potential. From the difference between the true potential and uncorrected, recovered potential, we propose a method to correct for this effect. We then demonstrate the application of this methodology to the analysis of experimental off-axis electron holography data acquired from the grain boundaries in lightly doped ceria.

36 MATERIALS SCIENCE↗