Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “generalized additive models”

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 505 records · Page 28

FY20 Improvements to CTF Code Verification and Unit Testing

In 2010, the U.S. Department of Energy created its first Energy Innovation Hub, which focuses on improving Light Water Reactors (LWRs) through Modeling and Simulation. This hub, named the Consortium for the Advanced Simulation of LWRs (CASL), attempts to characterize and understand LWR behavior under normal operating conditions and use any gained insights to improve their efficiency. In collaboration with North Carolina State University (NCSU), CASL has worked extensively on the thermal-hydraulic subchannel code Coolant Boiling in Rod Arrays–Three Field (COBRA-TF). The NCSU/CASL version of COBRA-TF has been rebranded as CTF. This document focuses on code verification test problems that ensure CTF converges to the correct answer for the intended application. The suite of code verification tests are mapped to the underlying conservation equations of CTF, and significant gaps are addressed. Convergence behavior and numerical errors are quantified for each of the tests. Tests that converge at the correct rate to the corresponding analytic solution are incorporated into the CTF automated regression suite. A new verification utility is created for this purpose, which enables code verification by generalizing the process. For problems that do not behave correctly, the results are reported but the problem is not included in the regression suite. In addition to verification studies, this document also quantifies the existing tests of constitutive models. A few existing gaps are addressed by adding new unit tests.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

A DGTD method for the numerical modeling of the interaction of light with nanometer scale metallic structures taking into account non-local dispersion effects

The interaction of light with metallic nanostructures is increasingly attracting interest because of numerous potential applications. Sub-wavelength metallic structures, when illuminated with a frequency close to the plasma frequency of the metal, present resonances that cause extreme local field enhancements. Exploiting the latter in applications of interest requires a detailed knowledge about the occurring fields which can actually not be obtained analytically. For the latter mentioned reason, numerical tools are thus an absolute necessity. The insight they provide is very often the only way to get a deep enough understanding of the very rich physics at play. For the numerical modeling of light-structure interaction on the nanoscale, the choice of an appropriate material model is a crucial point. Approaches that are adopted in a first instance are based on local (i.e. with no interaction between electrons) dispersive models, e.g. Drude or Drude–Lorentz models. From the mathematical point of view, when a time-domain modeling is considered, these models lead to an additional system of ordinary differential equations coupled to Maxwell's equations. However, recent experiments have shown that the repulsive interaction between electrons inside the metal makes the response of metals intrinsically non-local and that this effect cannot generally be overlooked. Technological achievements have enabled the consideration of metallic structures in a regime where such non-localities have a significant influence on the structures' optical response. This leads to an additional, in general non-linear, system of partial differential equations which is, when coupled to Maxwell's equations, significantly more difficult to treat. Nevertheless, dealing with a linearized non-local dispersion model already opens the route to numerous practical applications of plasmonics. In this work, we present a Discontinuous Galerkin Time-Domain (DGTD) method able to solve the system of Maxwell's equations coupled to a linearized non-local dispersion model relevant to plasmonics. While the method is presented in the general 3D case, numerical results are given for 2D simulation settings.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Stratospheric Temperature Changes: Observations and Model Simulations

This paper reviews observations of stratospheric temperatures that have been made over a period of several decades. Those observed temperatures have been used to assess variations and trends in stratospheric temperatures. A wide range of observation datasets have been used, comprising measurements by radiosonde (1940s to the present), satellite (1979 - present), lidar (1979 - present) and rocketsonde (periods varying with location, but most terminating by about the mid-1990s). In addition, trends have also been assessed from meteorological analyses, based on radiosonde and/or satellite data, and products based on assimilating observations into a general circulation model. Radiosonde and satellite data indicate a cooling trend of the annual-mean lower stratosphere since about 1980. Over the period 1979-1994, the trend is 0.6K/decade. For the period prior to 1980, the radiosonde data exhibit a substantially weaker long-term cooling trend. In the northern hemisphere, the cooling trend is about 0.75K/decade in the lower stratosphere, with a reduction in the cooling in mid-stratosphere (near 35 km), and increased cooling in the upper stratosphere (approximately 2 K per decade at 50 km). Model simulations indicate that the depletion of lower stratospheric ozone is the dominant factor in the observed lower stratospheric cooling. In the middle and upper stratosphere both the well-mixed greenhouse gases (such as CO) and ozone changes contribute in an important manner to the cooling.

Ramaswamy, V.↗

Process-based and observation-constrained SOA simulations in China: the role of semivolatile and intermediate-volatility organic compounds and OH levels

Organic aerosol (OA) is a major component of tropospheric submicron aerosol that contributes to air pollution and causes adverse effects on human health. Chemical transport models have difficulties in reproducing the variability in OA concentrations in polluted areas, hindering understanding of the OA budget and sources. Herein, we apply both process-based and observation-constrained schemes to simulate OA in GEOS-Chem. Comprehensive data sets of surface OA, OA components, secondary organic aerosol (SOA) precursors, and oxidants were used for model–observation comparisons. The base models generally underestimate the SOA concentrations in China. In the revised schemes, updates were made on the emissions, volatility distributions, and SOA yields of semivolatile and intermediate-volatility organic compounds (SVOCs and IVOCs) and additional nitrous acid sources. With all the model improvements, both the process-based and observation-constrained SOA schemes can reproduce the observed mass concentrations of SOA and show spatial and seasonal consistency with each other. Our best model simulations suggest that anthropogenic SVOCs and IVOCs are the dominant source of SOA, with a contribution of over 50% in most of China, which should be considered for pollution mitigation in the future. The residential sector may be the predominant source of SVOCs and IVOCs in winter, despite large uncertainty remaining in the emissions of IVOCs from the residential sector in northern China. The industry sector is also an important source of IVOCs, especially in summer. More SVOC and IVOC measurements are needed to constrain their emissions. Besides, the results highlight the sensitivity of SOA to hydroxyl radical (OH) levels in winter in polluted environments. The addition of nitrous acid sources can lead to over 30% greater SOA mass concentrations in winter in northern China. It is important to have good OH simulations in air quality models.

54 ENVIRONMENTAL SCIENCES↗

Adapting Wave-front Algorithms to Efficiently Utilize Systems with Deep Communication Hierarchies

Large-scale systems increasingly exhibit a differential between intra-chip and inter-chip communication performance especially in hybrid systems using accelerators. Processorcores on the same socket are able to communicate at lower latencies, and with higher bandwidths, than cores on different sockets either within the same node or between nodes. A key challenge is to efficiently use this communication hierarchy and hence optimize performance. We consider here the class of applications that contains wavefront processing. In these applications data can only be processed after their upstream neighbors have been processed. Similar dependencies result between processors in which communication is required to pass boundary data downstream and whose cost is typically impacted by the slowest communication channel in use. In this work we develop a novel hierarchical wave-front approach that reduces the use of slower communications in the hierarchy but at the cost of additional steps in the parallel computation and higher use of on-chip communications. This tradeoff is explored using a performance model. An implementation using the Reverse-acceleration programming model on the petascale Roadrunner system demonstrates a 27% performance improvement at full system-scale on a kernel application. The approach is generally applicable to large-scale multi-core and accelerated systems where a differential in system communication performance exists.

43 PARTICLE ACCELERATORS↗

RESULTS OF THE FY09 ENHANCED DOE HIGH LEVEL WASTE MELTER THROUGHPUT STUDIES AT SRNL

High-level waste (HLW) throughput (i.e., the amount of waste processed per unit time) is a function of two critical parameters: waste loading (WL) and melt rate. For the Waste Treatment and Immobilization Plant (WTP) at the Hanford Site and the Defense Waste Processing Facility (DWPF) at the Savannah River Site (SRS), increasing HLW throughput would significantly reduce the overall mission life cycle costs for the Department of Energy (DOE). The objective of this task is to develop data, assess property models, and refine or develop the necessary models to support increased WL of HLW at SRS. It is a continuation of the studies initiated in FY07, but is under the specific guidance of a Task Change Request (TCR)/Work Authorization received from DOE headquarters (Project Number RV071301). Using the data generated in FY07, FY08 and historical data, two test matrices (60 glasses total) were developed at the Savannah River National Laboratory (SRNL) in order to generate data in broader compositional regions. These glasses were fabricated and characterized using chemical composition analysis, X-ray Diffraction (XRD), viscosity, liquidus temperature (TL) measurement and durability as defined by the Product Consistency Test (PCT). The results of this study are summarized below: (1) In general, the current durability model predicts the durabilities of higher waste loading glasses quite well. A few of the glasses exhibited poorer durability than predicted. (2) Some of the glasses exhibited anomalous behavior with respect to durability (normalized leachate for boron (NL [B])). The quenched samples of FY09EM21-02, -07 and -21 contained no nepheline or other wasteform affecting crystals, but have unacceptable NL [B] values (> 10 g/L). The ccc sample of FY09EM21-07 has a NL [B] value that is more than one half the value of the quenched sample. These glasses also have lower concentrations of Al{sub 2}O{sub 3} and SiO{sub 2}. (3) Five of the ccc samples (EM-13, -14, -15, -29 and -30) completely crystallized with both magnetite and nepheline, and still had extremely low NL [B] values. These particular glasses have more CaO present than any of the other glasses in the matrix. It appears that while all of the glasses contain nepheline, the NL [B] values decrease as the CaO concentration increases from 2.3 wt% to 4.3 wt%. A different form of nepheline may be created at higher concentrations of CaO that does not significantly reduce glass durability. (4) The T{sub L} model appears to be under-predicting the measured values of higher waste loading glasses. Trends in T{sub L} with composition are not evident in the data from these studies. (5) A small number of glasses in the FY09 matrix have measured viscosities that are much lower than the viscosity range over which the current model was developed. The decrease in viscosity is due to a higher concentration of non-bridging oxygens (NBO). A high iron concentration is the cause of the increase in NBO. Durability, viscosity and T{sub L} data collected during FY07 and FY09 that specifically targeted higher waste loading glasses was compiled and assessed. It appears that additional data may be required to expand the coverage of the T{sub L} and viscosity models for higher waste loading glasses. In general, the compositional regions of the higher waste loading glasses are very different than those used to develop these models. On the other hand, the current durability model seems to be applicable to the new data. At this time, there is no evidence to modify this model; however additional experimental studies should be conducted to determine the cause of the anomalous durability data.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Trend Estimates of AERONET-Observed and Model-Simulated AOTs Between 1993 and 2013

Recently, temporal changes in Aerosol Optical Thickness (AOT) have been investigated based on model simulations, satellite and ground-based observations. Most AOT trend studies used monthly or annual arithmetic means that discard details of the generally right-skewed AOT distributions. Potentially, such results can be biased by extreme values (including outliers). This study additionally uses percentiles (i.e., the lowest 5%, 25%, 50%, 75% and 95% of the monthly cumulative distributions fitted to Aerosol Robotic Network (AERONET)-observed and ECHAM/MESSy Atmospheric Chemistry (EMAC)-model simulated AOTs) that are less affected by outliers caused by measurement error, cloud contamination and occasional extreme aerosol events. Since the limited statistical representativeness of monthly percentiles and means can lead to bias, this study adopts the number of observations as a weighting factor, which improves the statistical robustness of trend estimates. By analyzing the aerosol composition of AERONET-observed and EMAC-simulated AOTs in selected regions of interest, we distinguish the dominant aerosol types and investigate the causes of regional AOT trends. The simulated and observed trends are generally consistent with a high correlation coefficient (R = 0.89) and small bias (slope+/-2(sigma) = 0.75 +/- 0.19). A significant decrease in EMAC-decomposed AOTs by water-soluble compounds and black carbon is found over the USA and the EU due to environmental regulation. In particular, a clear reversal in the AERONET AOT trend percentiles is found over the USA, probably related to the AOT diurnal cycle and the frequency of wildfires. In most of the selected regions of interest, EMAC-simulated trends are mainly attributed to the significant changes of the dominant aerosols; e.g., significant decrease in sea salt and water soluble compounds over Central America, increase in dust over Northern Africa and Middle East, and decrease in black carbon and organic carbon over Australia.

AERONET↗

Considering value of information when using CFD in design

This thesis presents an approach to find lower resolution CFD models that can accurately lead a designer to a correct decision at a lower computational cost. High-fidelity CFD models often contain too much information and come at a higher computational cost, limiting the designs a designer can test and how much optimization can be performed on the design. Lower model resolution is commonly used to reduce computational time. However there are no clear guidelines on how much model accuracy is required. Instead experience and intuition are used to select an appropriate lower resolution model. This thesis presents an alternative to this ad hoc method by considering the added value of the addition information provided by increasing accurate and more computationally expensive models.

99 GENERAL AND MISCELLANEOUS↗

Accounting for multi-dimensional dependencies among decision-makers within a generalized model framework: An application to understanding shared mobility service usage levels

Activity-travel choices of individuals are influenced by spatial dependency effects. As individuals interact and exchange information with, or observe the behaviors of, those in close proximity of themselves, they are likely to shape their behavioral choices accordingly. For this reason, econometric choice models that account for spatial dependency effects have been developed and applied in a number of fields, including transportation. However, spatial dependence models to date have largely defined the strength of association across behavioral units based on spatial or geographic proximity. In the current context of social media platforms and ubiquitous internet and mobile connectivity, the strength of associations among individuals is no longer solely dependent on spatial proximity. Rather, the strength of associations among individuals may be based on shared attitudes and preferences as well. In other words, behavioral choice models may benefit from defining dependency effects based on attitudinal constructs in addition to geographical constructs. In this paper, frequency of usage of car-sharing and ride-hailing services is modeled using a generalized heterogeneous data model (GHDM) framework that incorporates multi-dimensional dependencies among decision-makers. The model system is estimated on the 2014-2015 Puget Sound Regional Travel Study survey sample, with proximity in latent attitudinal constructs defined by a number of personality trait variables. Model estimation results show that social dependency effects arising from similarities in attitudes and preferences are significant in explaining shared mobility service usage. Ignoring such effects may lead to erroneous estimates of the adoption and usage of future transportation technologies and mobility services.

33 ADVANCED PROPULSION SYSTEMS↗

Thin Interface Asymptotics for an Energy/Entropy Approach to Phase-Field Models with Unequal Conductivities

Karma and Rapped recently developed a new sharp interface asymptotic analysis of the phase-field equations that is especially appropriate for modeling dendritic growth at low undercoolings. Their approach relieves a stringent restriction on the interface thickness that applies in the conventional asymptotic analysis, and has the added advantage that interfacial kinetic effects can also be eliminated. However, their analysis focussed on the case of equal thermal conductivities in the solid and liquid phases; when applied to a standard phase-field model with unequal conductivities, anomalous terms arise in the limiting forms of the boundary conditions for the interfacial temperature that are not present in conventional sharp-interface solidification models, as discussed further by Almgren. In this paper we apply their asymptotic methodology to a generalized phase-field model which is derived using a thermodynamically consistent approach that is based on independent entropy and internal energy gradient functionals that include double wells in both the entropy and internal energy densities. The additional degrees of freedom associated with the generalized phased-field equations can be chosen to eliminate the anomalous terms that arise for unequal conductivities.

McFadden, G. B.↗

Transonic Aerodynamic Characteristics of a 45 deg Swept Wing Fuselage Model with a Finned and Unfinned Body Pylon Mounted Beneath the Fuselage or Wing, Including Measurements of Body Loads

An investigation of a model of a standard size body in combination with a representative 45 deg swept-wing-fuselage model has been conducted in the Langley 8-foot transonic pressure tunnel over a Mach number range from 0.80 to 1.43. The body, with a fineness ratio of 8.5, was tested with and without fins, and was pylon-mounted beneath the fuselage or wing. Force measurements were obtained on the wing-fuselage model with and without the body, for an angle-of-attack range from -2 deg to approximately 12 deg and an angle-of-sideslip range from -8 deg to 8 deg. In addition, body loads were measured over the same angle-of-attack and angle-of-sideslip range. The Reynolds number for the investigation, based on the wing mean aerodynamic chord, varied from 1.85 x 10(exp 6) to 2.85 x 10(exp 6). The addition of the body beneath the fuselage or the wing increased the drag coefficient of the complete model over the Mach number range tested. On the basis of the drag increase per body, the under-fuselage position was the more favorable. Furthermore, the bodies tended to increase the lateral stability of the complete model. The variation of body loads with angle of attack for the unfinned bodies was generally small and linear over the Mach number range tested with the addition of fins causing large increases in the rates of change of normal-force coefficient and nose-down pitching-moment coefficient. The variation of body side-force coefficient with sideslip for the unfinned body beneath the fuselage was at least twice as large as the variation of this load for the unfinned body beneath the wing. The addition of fins to the body beneath either the fuselage or the wing approximately doubled the rate of change of body side-force coefficient with sideslip. Furthermore, the variation of body side-force coefficient with sideslip for the body beneath the wing was at least twice as large as the variation of this load with angle of attack.

Wornom, Dewey E.↗

A Novel use of Direct Simulation Monte-Carlo to Model Dynamics of COVID-19 Pandemic Spread

In this report, we evaluate a novel method for modeling the spread of COVID-19 pandemic. In this new approach we leverage methods and algorithms developed for fully-kinetic plasma physics simulations using Particle-In-Cell (PIC) Direct Simulation Monte-Carlo (DSMC) models. This approach then leverages Sandia-unique simulation capabilities, and High-Performance Computer (HPC) resources and expertise in particle-particle interactions using stochastic processes. Our hypothesis is that this approach would provide a more efficient platform with assumptions based on physical data that would then enable the user to assess the impact of mitigation strategies and forecast different phases of infection. This work addresses key scientific questions related to the assumptions this new approach must make to model the interactions of people using algorithms typically used for modeling particle interactions in physics codes (kinetic plasma, gas dynamics). The model developed uses rational/physical inputs while also providing critical insight; the results could serve as inputs to, or alternatives for, existing models. The model work presented was developed over a four-week time frame, thus far showing promising results and many ways in which this model/approach could be improved. This work is aimed at providing a proof-of-concept for this new pandemic modeling approach, which could have an immediate impact on the COVID-19 pandemic modeling, while laying a basis to model future pandemic scenarios in a manner that is timely and efficient. Additionally, this new approach provides new visualization tools to help epidemiologists comprehend and articulate the spread of this and other pandemics as well as a more general tool to determine key parameters needed in order to better predict pandemic modeling in the future. In the report we describe our model for pandemic modeling, apply this model to COVID-19 data for New York City (NYC), assess model sensitivities to different inputs and parameters and , finally, propagate the model forward under different conditions to assess the effects of mitigation and associated timing. Finally, our approach will help understand the role of asymptomatic cases, and could be extended to elucidate the role of recovered individuals in the second round of the infection, which is currently being ignored.

59 BASIC BIOLOGICAL SCIENCES↗

Multifidelity Neural Network Formulations for Prediction of Reactive Molecular Potential Energy Surfaces

Here, this paper focuses on the development of multifidelity modeling approaches using neural network surrogates, where training data arising from multiple model forms and resolutions are integrated to predict high-fidelity response quantities of interest at lower cost. We focus on the context of quantum chemistry and the integration of information from multiple levels of theory. Important foundations include the use of symmetry function-based atomic energy vector constructions as feature vectors for representing structures across families of molecules and single-fidelity neural network training capabilities that learn the relationships needed to map feature vectors to potential energy predictions. These foundations are embedded within several multifidelity topologies that decompose the high-fidelity mapping into model-based components, including sequential formulations that admit a general nonlinear mapping across fidelities and discrepancy-based formulations that presume an additive decomposition. Methodologies are first explored and demonstrated on a pair of simple analytical test problems and then deployed for potential energy prediction for C 5 H 5 using B2PLYP-D3/6-311++G(d,p) for high-fidelity simulation data and Hartree–Fock 6-31G for low-fidelity data. For the common case of limited access to high-fidelity data, our computational results demonstrate that multifidelity neural network potential energy surface constructions achieve roughly an order of magnitude improvement, either in terms of test error reduction for equivalent total simulation cost or reduction in total cost for equivalent error.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Enabling Advanced Modeling and Simulations for Fuel-Flexible Combustors

The overall goal of the present project is to enable advanced modeling and simulations for the design and optimization of fuel-flexible turbine combustors. For this purpose we use a high-fidelity, extensively-tested large-eddy simulation (LES) code and state-of-the-art models for premixed/partially-premixed turbulent combustion developed in the PI's group. In the frame of the present project, these techniques are applied, assessed, and improved for hydrogen enriched premixed and partially premixed gas-turbine combustion. Our innovative approaches include a completely consistent description of flame propagation, a coupled progress variable/level set method to resolve the detailed flame structure, and incorporation of thermal-diffusion (non-unity Lewis number) effects. In addition, we have developed a general flamelet-type transformation holding in the limits of both non-premixed and premixed burning. As a result, a model for partially premixed combustion has been derived. The coupled progress variable/level method and the general flamelet tranformation were validated by LES of a lean-premixed low-swirl burner that has been studied experimentally at Lawrence Berkeley National Laboratory. The model is extended to include the non-unity Lewis number effects, which play a critical role in fuel-flexible combustor with high hydrogen content fuel. More specifically, a two-scalar model for lean hydrogen and hydrogen-enriched combustion is developed and validated against experimental and direct numerical simulation (DNS) data. Results are presented to emphasize the importance of non-unity Lewis number effects in the lean-premixed low-swirl burner of interest in this project. The proposed model gives improved results, which shows that the inclusion of the non-unity Lewis number effects is essential for accurate prediction of the lean-premixed low-swirl flame.

08 HYDROGEN↗

Enabling Advanced Modeling and Simulations for Fuel-Flexible Combustors

The overall goal of the present project is to enable advanced modeling and simulations for the design and optimization of fuel-flexible turbine combustors. For this purpose we use a high fidelity, extensively-tested large-eddy simulation (LES) code and state-of-the-art models for premixed/partially-premixed turbulent combustion developed in the PI's group. In the frame of the present project, these techniques are applied, assessed, and improved for hydrogen enriched premixed and partially premixed gas-turbine combustion. Our innovative approaches include a completely consistent description of flame propagation; a coupled progress variable/level set method to resolve the detailed flame structure, and incorporation of thermal-diffusion (non-unity Lewis number) effects. In addition, we have developed a general flamelet-type transformation holding in the limits of both non-premixed and premixed burning. As a result, a model for partially premixed combustion has been derived. The coupled progress variable/level method and the general flamelet transformation were validated by LES of a lean-premixed low-swirl burner that has been studied experimentally at Lawrence Berkeley National Laboratory. The model is extended to include the non-unity Lewis number effects, which play a critical role in fuel-flexible combustor with high hydrogen content fuel. More specifically, a two-scalar model for lean hydrogen and hydrogen-enriched combustion is developed and validated against experimental and direct numerical simulation (DNS) data. Results are presented to emphasize the importance of non-unity Lewis number effects in the lean-premixed low-swirl burner of interest in this project. The proposed model gives improved results, which shows that the inclusion of the non-unity Lewis number effects is essential for accurate prediction of the lean-premixed low-swirl flame.

01 COAL, LIGNITE, AND PEAT↗

Water Impact Test and Simulation of a Composite Energy Absorbing Fuselage Section

In March 2002, a 25-ft/s vertical drop test of a composite fuselage section was conducted onto water. The purpose of the test was to obtain experimental data characterizing the structural response of the fuselage section during water impact for comparison with two previous drop tests that were performed onto a rigid surface and soft soil. For the drop test, the fuselage section was configured with ten 100-lb. lead masses, five per side, that were attached to seat rails mounted to the floor. The fuselage section was raised to a height of 10-ft. and dropped vertically into a 15-ft. diameter pool filled to a depth of 3.5-ft. with water. Approximately 70 channels of data were collected during the drop test at a 10-kHz sampling rate. The test data were used to validate crash simulations of the water impact that were developed using the nonlinear, explicit transient dynamic codes, MSC.Dytran and LS-DYNA. The fuselage structure was modeled using shell and solid elements with a Lagrangian mesh, and the water was modeled with both Eulerian and Lagrangian techniques. The fluid-structure interactions were executed using the fast general coupling in MSC.Dytran and the Arbitrary Lagrange-Euler (ALE) coupling in LS-DYNA. Additionally, the smooth particle hydrodynamics (SPH) meshless Lagrangian technique was used in LS-DYNA to represent the fluid. The simulation results were correlated with the test data to validate the modeling approach. Additional simulation studies were performed to determine how changes in mesh density, mesh uniformity, fluid viscosity, and failure strain influence the test-analysis correlation.

Fasanella, Edwin L.↗

Intelligent Manufacturing Support: Specialized LLMs for Composite Material Processing and Equipment Operation

Engineering educational curriculum and standards cover many material and manufacturing options. However, engineers and designers are often unfamiliar with certain composite materials or manufacturing techniques. Large language models (LLMs) could potentially bridge the gap. Their capacity to store and retrieve data from large databases provides them with a breadth of knowledge across disciplines. However, their generalized knowledge base can lack targeted, industry-specific knowledge. To this end, we present two LLM-based applications based on the GPT-4 architecture: (1) The Composites Guide: a system that provides expert knowledge on composites material and connects users with research and industry professionals who can provide additional support and (2) The Equipment Assistant: a system that provides guidance for manufacturing tool operation and material characterization. By combining the knowledge of general AI models with industry-specific knowledge, both applications are intended to provide more meaningful information for engineers. In this paper, we discuss the development of the applications and evaluate it through a benchmark and two informal user studies. The benchmark analysis uses the Rouge and Bertscore metrics to evaluate our models’ performance against GPT-4o. The results show that GPT-4o and the proposed models perform similarly or better on the ROUGE and BERTScore metrics. The two user studies supplement this quantitative evaluation by asking experts to provide qualitative and open-ended feedback about our model’s performance on a set of domain-specific questions. The results of both studies highlight a potential for more detailed and specific responses with the Composites Guide and the Equipment Assistant.

Kapoor, Gunnika [Oak Ridge National Laboratory (OR↗

HADRONIC MODELS FOR THE EXTRA SPECTRAL COMPONENT IN THE SHORT GRB 090510

A short gamma-ray burst GRB 090510 detected by Fermi shows an extra spectral component between 10 MeV and 30 GeV, an addition to a more usual low-energy (<10 MeV) Band component. In general, such an extra component could originate from accelerated protons. In particular, inverse Compton emission from secondary electron-positron pairs and proton synchrotron emission are competitive models for reproducing the hard spectrum of the extra component in GRB 090510. Here, using Monte Carlo simulations, we test the hadronic scenarios against the observed properties. To reproduce the extra component around GeV with these models, the proton injection isotropic-equivalent luminosity is required to be larger than 10{sup 55} erg s{sup -1}. Such large proton luminosities are a challenge for the hadronic models.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗