Search NASA⌕ Search

SEARCH · Search NASA

Results for “Augmented”

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 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

A Scalable Reduced‐Order Model for the Steady Navier–Stokes Equations

Scaling up new scientific technologies from laboratory to industry often involves demonstrating performance on a larger scale. Computer simulations can accelerate design and predictions in the deployment process, though traditional numerical methods are computationally intractable even for intermediate pilot plant scales. Recently, the component reduced order modeling method has been developed to tackle this challenge by combining projection reduced order modeling and discontinuous Galerkin domain decomposition. However, while many scientific or engineering applications involve nonlinear physics, this method has only been demonstrated for various linear systems. In this work, the component reduced order modeling method is extended to steady Navier–Stokes flow, with application to general nonlinear physics in view. The large‐scale, global domain is decomposed into a combination of small‐scale unit component. Linear subspaces for flow velocity and pressure are identified via proper orthogonal decomposition over sample snapshots collected from each small‐scale unit component. Velocity bases are augmented with a pressure supremizer to satisfy the inf–sup condition for stable pressure prediction. Two different nonlinear reduced order modeling methods are employed and compared for efficient evaluation of nonlinear advection: A third‐order tensor projection operator and the empirical quadrature procedure. The proposed method is demonstrated on the flow over arrays of five different unit objects, achieving a 23‐fold speedup with less than 4% relative error in domains up to 256 times larger than the unit components. Furthermore, a numerical experiment with the pressure supremizer strongly indicates the need for a supremizer for stable pressure prediction. A comparison between the tensorial approach and the empirical quadrature procedure revealed a slight advantage of the empirical quadrature procedure. The framework is compared with an alternating Schwarz‐based reduced‐order approach, demonstrating improved efficiency and robustness for the DG‐based global solver while retaining flexibility for sub‐scale iterative solvers. The method is further extended to a coupled advection–diffusion and Navier–Stokes system, illustrating its applicability to multi‐physics problems and its potential for more general, inter‐coupled nonlinear systems.

42 ENGINEERING↗
DOE OSTI2025

Uncertainty Quantification Enabled by Automatic Differentiation for Hydrodynamic Simulation of Shock‐to‐Detonation Transition in High Explosives

Quantifying the effects of uncertainty in a reactive burn model on the run-to-detonation time in high explosives (HEs) provides a robust methodology for assessing the probability of an HE failing the IHE qualification standard. Moreover, uncertainty quantification helps evaluate whether the model calibration accurately represents data outside the calibration set. This study uses a specialized hydrodynamic simulation code for modeling detonation to determine the run-to-detonation time of the HE PBX 9502 for various impact velocities. To quickly approximate uncertainties in the model, a surrogate was constructed using a Taylor series expansion centered at the mean of the input parameters. To obtain the sensitivities required for constructing the Taylor series, HYP-percomplex Automatic Differentiation (HYPAD) was implemented. HYPAD is a methodology for infusing existing codes with automatic differentiation capabilities by augmenting variables with one or more imaginary units to compute step-size independent partial derivatives. These derivatives are accurate to machine precision with respect to the implemented numerical algorithm, meaning their accuracy reflects that of the underlying method (e.g., integration or discretization schemes). Using reduced order modeling techniques, the mean and standard deviation of the run-to-detonation time of a shock within PBX 9502 were computed for a number of initial impact velocities. A weighted least squares regression was then performed to obtain a best fit curve and prediction interval for the computed statistics. Historical data points from explosively driven wedge tests were utilized to validate the prediction interval, ensuring its reliability in predicting future outcomes. With this prediction interval and a known safety constraint curve, the most probable point of failure and the probability of failure for the HE PBX 9502 were determined.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Data‐Efficient Generation of Synthetic Microstructures of Polymer‐Bonded Energetic Material With Fine‐Tuned Stable Diffusion

Among current deep learning approaches for synthetic image generation, diffusion-based models stand out in terms of algorithmic stability and ability to retain high-fidelity image features with detailed resolution. Here, in this work, we employ Dreambooth, a method for fine-tuning Stable Diffusion, on X-ray CT images of microstructure of the polymer-bonded form (PBX) of a commonly used high explosive, Pentaerythritol tetranitrate (PETN), which yields generative models for creating synthetic PBX images. The models developed here represent five classes (or ‘lots’) of microstructures and demonstrate successful generation of images of each class with high fidelity, as verified by computed classification accuracy of ∼ 94% or higher. Data augmentation afforded by such image synthesis can be used to more reliably decipher underlying statistics, build processing-structure correlations, recognize off-normal structural anomalies, and identify age-related changes. Ideas related to converting image data into appropriate density mapping and performing mesoscale simulation or surrogate modeling of detonation are also discussed.

Dreambooth↗
DOE OSTI2024

Enhanced Interfacial Electron Transfer in Photocatalyst-Natural Enzyme Coupled Artificial Photosynthesis System: Tuning Strategies and Molecular Simulations

Laccase is capable of catalyzing a vast array of reactions, but its low redox potential limits its potential applications. The use of photocatalytic materials offers a solution to this problem by converting absorbed visible light into electrons to facilitate enzyme catalysis. Herein, MIL-53(Fe) and NH 2 -MIL-53(Fe) serve as both light absorbers and enzyme immobilization carriers, and laccase is employed for solar-driven chemical conversion. Electron spin resonance spectroscopy results confirm that visible light irradiation causes rapid transfer of photogenerated electrons from MOF excitation to T1 Cu(II) of laccase, significantly increasing the degradation rate constant of tetracycline (TC) from 0.0062 to 0.0127 min -1 . Conversely, there is only minimal or no electron transfer between MOF and laccase in the physical mixture state. Theoretical calculations demonstrate that the immobilization of laccase's active site and its covalent binding to the metal-organic framework surface augment the coupled system's activity, reducing the active site accessible from 27.8 to 18.1 Å. In conclusion, the constructed photo-enzyme coupled system successfully combines enzyme catalysis’ selectivity with photocatalysis's high reactivity, providing a promising solution for solar energy use.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Advances and Challenges of SCAN and r 2 SCAN Density Functionals in Transition‐Metal Compounds

ABSTRACT Transition‐metal compounds (TMCs) with open‐shell d ‐electrons are characterized by a complex interplay of lattice, charge, orbital, and spin degrees of freedom, giving rise to various fascinating applications. Often exhibiting exotic properties, these compounds are commonly classified as correlated systems due to strong inter‐electronic interactions called Hubbard U . This inherent complexity presents significant challenges to Kohn‐Sham density functional theory (KS‐DFT), the most widely used electronic structure method in condensed matter physics and materials science. While KS‐DFT is, in principle, exact for the ground‐state total energy, its exchange‐correlation energy must be approximated in practice. The mean‐field nature of KS implementations, combined with the limitations of current exchange‐correlation density functional approximations, has led to the perception that DFT is inadequate for correlated systems, particularly TMCs. Consequently, a common workaround involves augmenting DFT with an on‐site Hubbard‐like U correction. In recent years, the strongly constrained and appropriately normed (SCAN) density functional, along with its refined variant r 2 SCAN, has achieved remarkable progress in accurately describing the structural, energetic, electronic, magnetic, and vibrational properties of TMCs, challenging the traditional perception of DFT's limitations. This review explores the design principles of SCAN and r 2 SCAN, highlights their key advancements in studying TMCs, explains the mechanisms driving these improvements, and addresses the remaining challenges in this evolving field.

Zhang, Yubo [Minjiang Collaborative Center for The↗
DOE OSTI2025

A Comparison of Pre‐Construction and Operational Wake Loss Estimates for Land‐Based Wind Plants

The overall bias between pre‐construction energy yield assessment (EYA) estimates of wind plant energy production and the achieved operational production is improving in the wind industry, but uncertainty remains high for individual wind plants. Wake effects within wind plants are one of the largest sources of energy loss considered in the EYA process, and previous work shows wake loss estimates to be a major source of disagreement among wind energy consultants who perform EYAs. To better understand the accuracy of wake loss predictions, we compare overall operational wake loss estimates based on supervisory control and data acquisition data to pre‐construction estimates provided by six wind energy consultants for five land‐based wind plants in North America. By augmenting existing approaches for quantifying operational wake losses, we estimate wake losses during the period of record for which operational data are available as well as the expected long‐term wake losses, based on historical reanalysis weather data, to which the EYA estimates are compared. To account for power variations at different turbine locations caused by terrain‐induced wind resource heterogeneity, we correct the operational wake loss estimates using predicted freestream wind speed variations from the Wind Systems Engineering Reynolds‐averaged Navier–Stokes (RANS) tool. We identify long‐term corrected operational wake losses between 1.9% and 6.4% for the five plants, with a mean loss of 4%. For the project deemed most acceptable for operational wake loss assessment, which is located in the simplest terrain and isolated from neighboring plants, the mean EYA wake loss estimate is within 0.7 percentage points of the operational value of 6.4%. For most of the remaining plants, results suggest that wake losses are generally overpredicted by 2.6–6.3 percentage points. However, operational wake losses may be underestimated for many of these projects because of spatial wind resource variations not captured by the RANS model, external wake effects that are unaccounted for in the estimation process, and wind plant blockage effects. To better understand factors that contribute to the observed wake losses, we investigate operational wake losses as a function of wind direction and wind speed. As expected, wake losses are generally concentrated near wind directions that are aligned with rows of closely spaced turbines and at below‐rated wind speeds; however, for some projects, the energy produced by the wind plant exceeds the estimated potential energy of the plant without wake interactions for certain wind directions and wind speeds, suggesting inaccurate assumptions in the wake loss estimation method for those plants. Lastly, we compare predicted and operational wake losses for individual wind turbines, finding that even when overall wake losses are predicted accurately, large uncertainty exists at the turbine level.

17 WIND ENERGY↗
DOE OSTI2025

Use of Frit‐Disc Crucible Sets to Make Solution Growth More Quantitative and Versatile

The recent availability of step‐edge, frit‐disc crucible sets (generally sold as Canfield Crucible Sets or CCS) has led to multiple innovations associated with the group's use of solution growth. The use of CCS allows for the clean separation of liquid from solid phases during the growth process. This clean separation enables the reuse of the decanted liquid, either allowing for simple, economic, savings associated with recycling expensive precursor elements or allowing for the fractionation of a growth into multiple, small steps, revealing the progression of multiple solidifications. Clean separation of liquid from solid phases also allows for the determination of the liquidus line (or surface) and the creation, or correction, of composition–temperature phase diagrams. The reuse of clean decanted liquid has also allowed to prepare liquids ideally suited for the growth of large single crystals of specific phases by tuning the composition of the melt to the optimal composition for growth of the desired phase, often with reduced nucleation sites. Finally, it is discussed how solution growth and CCS use can be harnessed to provide a plethora of composition–temperature data points defining liquidus lines or surfaces with differing degrees of precision to either test or anchor artificial intelligence and/or machine‐learning‐based attempts to augment and extend the limited experimentally determined database.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

NuGraph2 with context-aware inputs: physics-inspired improvements in semantic segmentation

Graph neural networks have recently shown strong promise for event reconstruction tasks in Liquid Argon Time Projection Chambers, yet their performance remains limited for underrepresented classes of particles, such as Michel electrons. In this work, we investigate physics-informed strategies to improve semantic segmentation within the NuGraph2 architecture. We explore three complementary approaches: (i) enriching the input representation with context-aware features derived from detector geometry and track continuity, (ii) introducing auxiliary decoders to capture class-level correlations, and (iii) incorporating energy-based regularization terms motivated by Michel electron energy distributions. Experiments on MicroBooNE public datasets show that physics-inspired feature augmentation yields the largest gains, particularly boosting Michel electron precision and recall by disentangling overlapping latent space regions. In contrast, auxiliary decoders and energy-regularization terms provided limited improvements, partly due to the hit-level nature of NuGraph2, which lacks explicit particle- or event-level representations. Our findings highlight that embedding physics context directly into node-level inputs is more effective than imposing task-specific auxiliary losses, and suggest that future hierarchical architectures such as NuGraph3, with explicit particle- and event-level reasoning, will provide a more natural setting for advanced decoders and physics-based regularization. The code for this work is publicly available on Github at https://github.com/vitorgrizzi/nugraph_phys/tree/main_phys.

Other Experiments↗
DOE OSTI2024

Rindler fluids from gravitational shockwaves

We study a correspondence between gravitational shockwave geometry and its fluid description near a Rindler horizon in Minkowski spacetime. Utilizing the Petrov classification that describes algebraic symmetries for Lorentzian spaces, we establish an explicit mapping between a potential fluid and the shockwave metric perturbation, where the Einstein equation for the shockwave geometry is equivalent to the incompressibility condition of the fluid, augmented by a shockwave source. Then we consider an Ansatz of a stochastic quantum source for the potential fluid, which has the physical interpretation of shockwaves created by vacuum energy fluctuations. Under such circumstance, the Einstein equation, or equivalently, the incompressibility condition for the fluid, becomes a stochastic differential equation. By smearing the quantum source on a stretched horizon in a Lorentz invariant manner with a Planckian width (similarly to the membrane paradigm), we integrate fluctuations near the Rindler horizon to find an accumulated effect of the variance in the round-trip time of a photon traversing the horizon of a causal diamond.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2025

Nonperturbative quantum gravity in a closed Lorentzian universe

We study how meaningful physical predictions can arise in nonperturbative quantum gravity in a closed Lorentzian universe. In such settings, recent developments suggest that the quantum gravitational Hilbert space is one-dimensional and real for each α-sector, as induced by spacetime wormholes. This appears to obstruct the conventional quantum-mechanical prescription of assigning probabilities via projection onto a basis of states. While previous approaches have introduced external observers or augmented the theory to resolve this issue, we argue that quantum gravity itself contains all the necessary ingredients to make physical predictions. We demonstrate that the emergence of classical observables and probabilistic outcomes can be understood as a consequence of partial observability: physical observers access only a subsystem of the universe. Tracing out the inaccessible degrees of freedom yields reduced density matrices that encode classical information, with uncertainties exponentially suppressed by the environment’s entropy. We develop this perspective using both the Lorentzian path integral and operator formalisms and support it with a simple microscopic model. Our results show that quantum gravity in a closed universe naturally gives rise to meaningful, robust predictions without recourse to external constructs.

AdS-CFT Correspondence↗
DOE OSTI2026

Toward real-time optimization through model reduction and model discrepancy sensitivities

Optimization problems arise in a range of scenarios, from optimal control to model parameter estimation. In many applications, such as the development of digital twins, it is essential to solve these optimization problems within wall-clock-time limitations. However, this is often unattainable for complex systems, such as those modeled by nonlinear partial differential equations. One strategy for mitigating this issue is to construct a reduced-order model (ROM) that enables more rapid optimization. In particular, the use of nonintrusive ROMs—those that do not require access to the full-order model at evaluation time—is popular because they facilitate the computation of optimization solutions within the wall-clock time requirements. However, the optimization solution will be unreliable if the iterates move outside the ROM training data. This article proposes the use of hyper-differential sensitivity analysis with respect to model discrepancy (HDSA-MD) as a computationally efficient tool to augment ROM-constrained optimization and improve its reliability. The proposed approach consists of two phases: (i) an offline phase where several full-order model evaluations are computed to train the ROM, and (ii) an online phase where a ROM-constrained optimization problem is solved, a limited number of full-order model evaluations are computed, and HDSA-MD is used to enhance the optimization solution. Numerical results are demonstrated for two examples, atmospheric contaminant control and wildfire ignition location estimation, in which a ROM is trained offline using inaccurate atmospheric data. In conclusion, the HDSA-MD update yields a significant improvement in the ROM-constrained optimization solution using only one full-order model evaluation online with corrected atmospheric data.

PDE-constrained optimization↗
DOE OSTI2025

Unraveling the Hsp70-ROS-autophagy axis in pentachlorophenol-challenged lung and liver epithelial cells

Pentachlorophenol (PCP) was extensively utilized as an organochlorine pesticide and wood preservative in the United States from the 1930s until the Environmental Protection Agency (EPA) imposed restrictions due to concerns about its toxicity and potential carcinogenic properties. Although it is no longer widely used, PCP remains a concern due to its environmental persistence and potential for long-term health effects. Significant occupational and environmental exposures have likely occurred, with the health and economic costs of PCP exposure potentially being substantial given its known toxicity. Notably, PCP exhibits rapid absorption through both the skin and respiratory system and has been shown to cause hepatotoxicity, developmental toxicity, immunotoxicity, irritation, and carcinogenicity in laboratory animal studies. PCP exposure induces oxidative stress, a key mechanism underlying its inflammatory and toxic effects, which can activate cellular stress responses including upregulation of heat shock protein 70 (Hsp70). Previous studies in lung and liver epithelial cells have shown that Hsp70 and oxidative stress play pivotal roles in triggering autophagy. This study establishes the critical role of the Hsp70-reactive oxygen species (ROS)-autophagy axis in regulating cellular responses to PCP exposure in human alveolar (A549) and liver carcinoma (HepG2) epithelial cells. Our research elucidated the molecular mechanisms underlying PCP's cellular effects, demonstrating that its exposure resulted in increased expression of autophagy-related proteins (Beclin-1, LC3B, ATG12, and ATG16), subunits of NADPH oxidase (NCF-1, NCF-2, NOX2, and Rac), and antioxidant proteins (SOD and GPx) in both lung and liver cell types. Notably, PCP augmented the interaction between Hsp70 and the autophagy regulator Beclin-1. Pretreatment with the ROS inhibitor N-acetylcysteine or Hsp70 knockdown markedly reversed PCP-induced responses. Our in-silico protein–protein docking analysis and molecular dynamics simulation studies revealed enhanced interactions and/or stable confirmations maintained throughout the simulations for TLR4-Hsp70 and Hsp70-Beclin-1 complexes in the presence of PCP. These findings provide a strong foundation for future studies, employing in vivo experimental models and human populations to identify promising targets for PCP-induced toxicity and cellular injury. As a result, these findings may have far-reaching implications for public health and environmental policy, ultimately leading to the identification of biomarkers and the development of more effective interventions for environmentally induced toxicity and diseases.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

U-net architected deep material network training with microstructure local field information

The Deep Material Network (DMN) has recently emerged as a powerful reduced-order modeling framework for simulating the mechanical response of heterogeneous materials such as composites. Unlike most data-driven approaches that directly learn a material’s response under prescribed loading, the DMN acts as a homogenization operator, learning the kinematic constraints and mechanical interactions of the underlying microstructure. However, traditional DMN training relies exclusively on homogenized effective properties derived from Direct Numerical Simulations (DNS), discarding the rich local field data that govern microstructural interactions. In this work, we extend the DMN framework to incorporate such local field information into the offline training process. Utilizing a U-Net architecture, we augment the DMN training objective to include the first and second statistical moments of the local stress fields obtained from linear DNS. This ensures that the learned network topology not only fits the effective stiffness but also accurately reflects the internal local stress and strain partitioning of the microstructure. The results confirm that supervising the localization process during training yields a superior surrogate model, reducing local prediction errors by an order of magnitude and significantly improving generalization to unseen nonlinear constitutive behaviors compared to traditional DMNs.

36 MATERIALS SCIENCE↗
DOE OSTI2026

Analysis of an irradiated uranium sample for source attribution without chemical separation using microplasma ionization and ultrahigh resolution mass spectrometry

The use of element isotope ratios has great potential in not only determining the reactor type used to produce plutonium (Pu) but also in determining the burnup and the time since irradiation. While a powerful nuclear forensic technique, determining element isotope ratios is complicated by severe isobaric interferences when performed on typical inductively coupled plasma mass spectrometers. Such analyses require extensive chemical separations prior to analysis to alleviate the inter-elemental isobars. Ultrahigh mass resolution spectrometry provides a potential alternative, greatly reducing the complexity of sample preparation and turnaround times for these critical measurements. To demonstrate the power of the approach, a sample of irradiated, depleted uranium was analyzed with the liquid sampling—atmospheric pressure glow discharge ion source coupled to an Orbitrap mass spectrometer. The Orbitrap is augmented with an external data acquisition system, Spectroswiss’s FTMS-Booster X2T, allowing collection of extended ion transients, providing higher mass resolution. In using this approach, the 150 Sm/ 149 Sm and 152 Sm/ 149 Sm isotope ratios were found to be within 20% of predicted values without any chemical separations and without mass bias corrections. In addition, the 240 Pu/ 239 Pu isotope ratio was determined, free from the 238 UH + interferences common to the ICP-MS platforms, while at the same time allowing for the determination of U isotopic signatures. While these demonstrative results are from a single sample, the advantages of the microplasma/ultrahigh mass resolution approach to intra-element isotope ratio determinations are clear.

Fuel burnup↗
DOE OSTI2024

Calculation of machine precision second order derivatives using dual-complex numbers

It is well known that both complex and dual numbers can be employed to obtain machine precision first-order derivatives; however, neither, on their own, can compute machine precision 2nd order derivatives. To address this limitation, it is demonstrated in this paper that combined dual-complex numbers can be used to compute machine precision 1st and 2nd order derivatives. The dual-complex approach is simpler than utilizing multicomplex or hyper-dual numbers as existing dual libraries can be used as is or easily augmented to accept complex numbers, and the complexity of developing, integrating, and deploying multicomplex or hyper-dual libraries is avoided. The efficacy of this approach is demonstrated for both univariate and multivariate functions. Finally, source code examples using the Python, Julia, and Mathematica languages are provided as supplemental material.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Geothermal-integrated thermally anisotropic building envelope for energy and peak-demand reduction

Buildings consume large amounts of energy for heating and cooling, while peak electricity demand places significant stress on the power grid. This paper presents a reduced-order co-simulation framework and load-oriented supervisory control strategy for a geothermal-integrated thermally anisotropic building envelope with a ground loop (TABE+GL). In TABE+GL, a hydronic loop embedded in the building envelope is directly coupled with a geothermal ground loop, allowing for bidirectional heat exchange between the envelope, the ground, and the indoor environment. A hybrid co-simulation framework was established by coupling a reduced-order resistor–capacitor (RC) thermal network model with EnergyPlus augmented with GHEDesigner modules. The RC model generated feasible heat flux options under three operating modes, and EnergyPlus predicted sensible loads, energy use, and pump energy demand. At each simulation step, a supervisory control algorithm selected the optimal loop configuration and duty factor that maximizes useful TABE geothermal utilization without exceeding the predicted sensible load, thereby avoiding overheating or cooling. Case studies were conducted for Los Angeles, California, Charleston, South Carolina, and Denver, Colorado. Results showed that the proposed framework reduced HVAC electricity consumption by 43%–67%, natural gas use for space heating by 11%–38%, and peak electricity demand by 43%–88%. These results highlight the potential of combining reduced-order envelope modeling, direct geothermal coupling, and load-oriented supervisory control to improve whole building energy performance and reduce peak demand across diverse weather conditions.

Howard, Daniel [Southern Adventist University]↗
DOE OSTI2024

Modification of Jet Velocities in an Explosively Loaded Copper Target with a Conical Cavity

Here, in this work, the design and execution of an experiment with the goal of demonstrating control over the evolution of a copper jet is described. Simulations show that when using simple multi-material buffers placed between a copper target with a conical cavity and a cylinder of high-explosive, a variety of jetting behaviors occur based on material placement, including both jet velocity augmentation and mitigation. A parameter sweep was performed to determine optimal buffer designs in two configurations. Experiments using the optimal buffer designs verified the effectiveness of the buffers at altering jet velocities. Similar trends were shown between the experimental results and the modeling.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2024

Optimizing High-Throughput Inference on Graph Neural Networks at Shared Computing Facilities with the NVIDIA Triton Inference Server

Abstract With machine learning applications now spanning a variety of computational tasks, multi-user shared computing facilities are devoting a rapidly increasing proportion of their resources to such algorithms. Graph neural networks (GNNs), for example, have provided astounding improvements in extracting complex signatures from data and are now widely used in a variety of applications, such as particle jet classification in high energy physics (HEP). However, GNNs also come with an enormous computational penalty that requires the use of GPUs to maintain reasonable throughput. At shared computing facilities, such as those used by physicists at Fermi National Accelerator Laboratory (Fermilab), methodical resource allocation and high throughput at the many-user scale are key to ensuring that resources are being used as efficiently as possible. These facilities, however, primarily provide CPU-only nodes, which proves detrimental to time-to-insight and computational throughput for workflows that include machine learning inference. In this work, we describe how a shared computing facility can use the NVIDIA Triton Inference Server to optimize its resource allocation and computing structure, recovering high throughput while scaling out to multiple users by massively parallelizing their machine learning inference. To demonstrate the effectiveness of this system in a realistic multi-user environment, we use the Fermilab Elastic Analysis Facility augmented with the Triton Inference Server to provide scalable and high-throughput access to a HEP-specific GNN and report on the outcome.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗