Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “input estimation”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

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

At least 1,027 records · Page 57
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Smart thermostat data-driven U.S. residential occupancy schedules and development of a U.S. residential occupancy schedule simulator

Occupancy schedule is one of the key inputs in Building Energy Modeling (BEM) to reflect the interaction between buildings and occupants. Over the past decades, standardized occupancy schedules, developed mainly by engineering rule-of-thumb, have been widely used in BEM due to its simplicity and lack of real measured occupancy data. However, the BEM community has recognized their association with uncertainty and reliability in simulation results from BEM. This study introduces representative occupancy schedules in the U.S. residential buildings, derived from a large smart thermostat dataset and time-series K-means clustering, and an open-source tool to generate a stochastic residential occupancy schedule. Over 90,000 residential occupancy schedules were estimated from the ecobee Donate Your Data dataset. Then, the representative occupancy schedules were identified through clustering. This study further investigated the impacts of three parameters (day, house type, and state) on residential occupancy schedules. Then, a tool, the Residential Occupancy Schedule Simulator (ROSS), is developed using the representative occupancy schedules derived in this study. Details of this tool are presented in this paper. In conclusion, the derived representative occupancy schedules and the ROSS tool can help improve the energy modeling of residential buildings.

42 ENGINEERING↗
DOE OSTI2022

Survival analysis of localized prostate cancer with deep learning

In recent years, data-driven, deep-learning-based models have shown great promise in medical risk prediction. By utilizing the large-scale Electronic Health Record data found in the U.S. Department of Veterans Affairs, the largest integrated healthcare system in the United States, we have developed an automated, personalized risk prediction model to support the clinical decision-making process for localized prostate cancer patients. This method combines the representative power of deep learning and the analytical interpretability of parametric regression models and can implement both time-dependent and static input data. To collect a comprehensive evaluation of model performances, we calculate time-dependent C-statistics C td over 2-, 5-, and 10-year time horizons using either a composite outcome or prostate cancer mortality as the target event. The composite outcome combines the Prostate-Specific Antigen (PSA) test, metastasis, and prostate cancer mortality. Our longitudinal model Recurrent Deep Survival Machine (RDSM) achieved C td 0.85 (0.83), 0.80 (0.83), and 0.76 (0.81), while the cross-sectional model Deep Survival Machine (DSM) attained C td 0.85 (0.82), 0.80 (0.82), and 0.76 (0.79) for the 2-, 5-, and 10-year composite (mortality) outcomes, respectively. In addition to estimating the survival probability, our method can quantify the uncertainty associated with the prediction. The uncertainty scores show a consistent correlation with the prediction accuracy. We find PSA and prostate cancer stage information are the most important indicators in risk prediction. Our work demonstrates the utility of the data-driven machine learning model in prostate cancer risk prediction, which can play a critical role in the clinical decision system.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2024

Impact of the Li 6 asymptotic normalization constant onto α -induced reactions of astrophysical interest

Indirect methods have become the predominant approach in experimental nuclear astrophysics for studying several low-energy nuclear reactions occurring in stars, as direct measurements of many of these relevant reactions are rendered infeasible due to their low reaction probability. Such indirect methods, however, require theoretical input that in turn can have significant poorly quantified uncertainties, which can then be propagated to the reaction rates and have a large effect on our quantitative understanding of stellar evolution and nucleosynthesis processes. Here we present two such examples involving α-induced reactions, 13 C (α,n)⁢ 16 O and 12 C (α,γ)⁢ 16 O, for which the low-energy cross sections have been constrained with ( 6 Li,d) transfer data. In this Letter, we discuss how a first-principle calculation of 6 Li leads to a 21% reduction of the 12 C⁡(α,γ) ⁢ 16 O cross sections with respect to a previous estimation. This calculation further resolves the discrepancy between recent measurements of the 13 C (α,n)⁢ 16 O reaction and points to the need for improved theoretical formulations of nuclear reactions.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2022

Liquid Salt Combined-Cycle Pilot Plant Design

The work described in this report is responsive to the Office of Fossil Energy program ‘Energy Storage for Fossil Power Generation.’ This Phase I report has been prepared by Pintail Power LLC, with support from Nexant ECA, Electric Power Research Institute (EPRI) and Southern Company Services as a deliverable for the U.S. Department of Energy for NETL Award DE-FE-00320016. The Liquid Salt Combined Cycle™ (LSCC™) technology provides large-scale energy storage integrated with Fossil Electric Generating Units (FEGUs) to meet critical needs in the energy transition by providing: • the lowest cost large-scale storage for time-shifting of renewable energy, • superior fuel efficiency to reduce GHGs from dispatchable resources, • flexible capacity and ramping to balance variability of wind and solar resources, • essential grid stability services to assure reliability of a low-carbon grid. The LSCC approach: • employs equipment that has already been proven in utility service, • uses safe, non-toxic, non-degrading, perpetual-life storage medium, • leverages and repurposes existing FEGU assets, • expands the value stack of energy storage to reduce market, financing, and commodity risks. Pintail Power has developed the LSCC technology to meet the need for reliable, efficient, and cost-effective integration of Variable Renewable Energy (VRE) into a low-carbon electric grid by coupling proven thermal energy storage with proven gas turbines, steam turbines, and heat transfer equipment. This novel approach is intended to address the key issues facing the grid and operators of renewable and fossil generating units including: • Overgeneration and curtailment of renewables, • Need for fast ramping dispatchable resources, • Improved efficiency and flexibility of fossil units, • Additional peaking capacity to support electrification of transportation and heating, • Provision of reliability services to support high penetration of VRE, especially synchronous inertia and fast frequency response. A Technology Readiness assessment by EPRI confirmed that LSCC technology consists of commercially proven hardware used in industrial and utility applications. Although the novel LSCC approach has not yet been demonstrated as a complete system, interfaces between major components have been conservatively specified. A Phase III pilot is planned to demonstrate equipment integration and operation. The patented innovation is removal of the evaporator section from the exhaust heat recovery system, with the evaporation performed by stored energy in a separate steam generator. This arrangement couples renewable and fossil power generation via long-duration energy storage to deliver cost, performance, and operational synergies, including superior charging and discharging flexibility, reduced fuel consumption and lower CO 2 emissions compared to conventional Combined Cycle Power Plants, and low-cost, large-scale energy storage. The LSCC technology is composed of proven equipment integrated with gas turbine exhaust heat in a novel system. During charging, electric heaters raise the salt temperature as it flows from the Cold Salt Tank to the Hot Salt Tank. During discharging, hot salt produces steam from feedwater that is heated with gas turbine exhaust, which also superheats steam to drive a steam turbine. LSCC technology can be added to any combustion-turbine to integrate renewable energy, provide needed grid services, and increase the value of fossil electric generating units based on the technology’s following attributes: • Long-duration storage enables time-shifting of VRE to avoid curtailment and impairment of renewable assets. • Long storage duration combined with fast-charging capability increases arbitrage opportunities by storing more energy when the price is low and discharging more hours when the price is high. • Long storage duration allows resource adequacy to be supplied across multiple days to increase reliability and reduce risk. • The stored energy reduces fuel heat rate and GHG emissions, and increases merit, so the LSCC dispatches earlier and longer to increase the plant’s capacity factor and asset value. • The stored energy enables pre-heating and startup of the steam cycle, without operating the gas turbine, to enable fast startup and ramping when dispatched for discharge. • The steam turbine can operate without the gas turbine so it can provide valuable synchronous inertia during charging without consuming fuel. • Fast frequency response and regulation services can be provided during charging using solid-state heater and pump controls to vary the charge power input in response to grid signals. • The LSCC system can be configured for resilience including black start, islanded/micro-grid operation, and even self-recharging of storage using either gas turbine power or gas turbine exhaust heat. The commercialization plan is to add LSCC technology to existing simple cycle gas turbine power plants with the 50MW GE LM6000 aero-derivative gas turbine as the reference design basis. A Techno-economic assessment of the reference design evaluated the benefits (Levelized Avoided Cost of Energy) and costs (Levelized Cost of Energy). The plant definition included all major systems and budgetary vendor quotes. Pintail Power and NexantECA developed the overall cost estimate for the LSCC plant up to the total plant cost level, following the DOE-NETL cost estimate guidelines at AACE Class 3 (-20%/+30%). This includes the equipment cost, bulk material, direct and indirect labor costs to arrive at the bare erected cost. Engineering costs are factored from the BEC and added to it to arrive at the EPC cost. Process and project contingencies were then factored from the EPC cost and rolled-up to yield the total plant cost of $\$$184 million for 1746 MWh of discharge electricity. • At $\$$105/kWh, the reference plant costs less than any of the Energy Storage Systems evaluated by PNNL in 2020 for the Energy Storage Grand Challenge. Operations and Maintenance cost estimates were scaled from combined cycle practice, assuming that the LSCC unit was co-located with and sharing some labor expense with other units, to arrive at $\$$2.2 million per year. Plant economics were evaluated using prices from the ERCOT Day-Ahead Market for calendar year 2019 (excluding the market disruptions from the COVID pandemic and the February 2020 deep freeze event). Assuming economic dispatch in the ERCOT Day-Ahead market, the reference plant capacity factor would have discharged for 2777 hours at 91.9 MW, a 31.66% capacity factor, with a marginal cost of $\$$25.59/MWh, and a LACE of $\$$82.41/MWh. Fixed charges were calculated according to EIA guidelines to arrive at an LCOE of $\$$83.48. The benefit-to-cost ratio of 0.99 suggests that the reference plant would have been cost-effective and competitive in the market. EPRI interviewed selected utilities to gauge the need for, applicability of and interest in the LSCC system. Several utilities are currently managing increased load growth along with the inclusion of increasing levels of renewable generation, putting pressure on conventional generation by requiring increased turndown requirements and ultimately lower capacity factors. All of the utilities interviewed have CO 2 reduction targets in the 2030-2050 timeframe that will severely limit the participation of fossil generation and require better utilization of carbon free generation. While there is limited opportunity for storage in the current markets, the utilities interviewed stated that there will be a substantial need for long duration energy storage in the future given the expected trends. Utilizing an energy storage system will generally be preferred over new gas capacity in some cases, with the capabilities of the LSCC system being a potential option for retrofit to existing simple cycle gas turbine units, allowing them to deliver greater participation in the market with lower carbon intensity. A technology gap assessment and technology maturation plan identified a pilot-scale demonstration as the final step before commercialization. Key gaps to be addressed during the Phase II FEED (Front-End Engineering Design) are component selection and design, commissioning procedures, and operational procedures and the control system for LSCC charging and discharging. The project team has been expanded to include Wood Group PLC as EPC. The proposed Phase II work leads to a pilot-scale engineering demonstration (TRL 6) to be conducted at Southern Company’s Plant Rowan, where the prototype system will perform “all the functions that will be required of the operational system.” The proposed pilot will facilitate commercialization (TRL-9) by scale-up to utility-scale systems integrated with peaking GTs or directly to facility scale systems using industrial GTs. The conceptual design for the pilot plant focuses on the novel integration aspects of LSCC technology. A slipstream of gas turbine exhaust will feed a waste heat recovery unit coupled to a molten salt steam generator heated by stored energy. The pilot is intended to demonstrate all key operating modes of the LSCC technology during charging, discharging and standby. The pilot equipment will be approximately one-seventh scale of the LM6000 commercial target and is expected to have commercial off-ramp potential for facility-scale applications.

01 COAL, LIGNITE, AND PEAT↗
DOE OSTI2025

The Design of a Multiphase CO2 Turbine Test Facility

As the energy market continues to evolve with increased renewables, such as solar and wind energy, the need to balance energy demand with its availability has become increasingly more challenging. The country’s aggressive decarbonization goals in the upcoming decades will increase renewable energy market penetration, thus exacerbating the disparity between energy demand and renewable resource availability. Long-duration energy storage technologies, such as pumped thermal energy storage (PTES), are potential solutions to this challenge with high potential round-trip efficiency (RTE), no geological or geographical constraints, and low safety risks. The charge mode of a PTES system operates a heat pump cycle, converting input power to thermal energy in hot and cold reservoirs. Fundamentally, heat pumps are commonly used in daily life, but the operating conditions required by high RTE PTES systems are outside the design experience of these systems. The low-pressure side of the cycle for a supercritical CO2 (sCO2) PTES operates just above the vapor-dome and potentially drives the turbine to expand into the dome, resulting in multiphase operation in the rear stages of the turbine. This paper presents the design of the test facility for an advanced multiphase-tolerant sCO2 turbine for the heat pump of a PTES system. Heat rejection and pressure loss estimations for the multiphase sections of the test loop presented unforeseen challenges.

25 ENERGY STORAGE↗
DOE OSTI2025

Contributions of vegetation heterogeneity within tower footprint to CO 2 flux estimations through graph neural network modeling

Net ecosystem exchange of CO 2 (Fc) measured directly by eddy covariance towers is based on various assumptions, including large, flat and homogenous land cover type. In reality, often a tower site is not large enough for flux measurements, and landscapes consist of patches of different land cover types within the flux footprint. In addition, some portions of fluxes are contributed by different cover types when a footprint exceeds the size of the target ecosystem. The contributions of non-dominant patches to Fc are often ignored. Here, in this study, we propose a novel integrated modeling framework that combines random forest (RF) and XGBoost with a residual correction module based on a deep graph convolutional network (DeeperGCN) to simulate Fc for seven flux measurement sites in southwest Michigan. High-resolution remote sensing vegetation indices, soil properties, meteorological variables, and footprint-weighted spatial features were used as model inputs at three spatial resolutions (10 m, 20 m, 30 m), and their importance in predicting Fc with DeeperGCN was assessed. We found that residual correction using DeeperGCN significantly improved prediction accuracy, with the R 2 increasing from 0.9098 to 0.9479 for RF and from 0.9235 to 0.9433 for XGBoost. At site level, the maximum improvement in R 2 reached 0.1617. Paired t-tests confirmed that these improvements were statistically significant (p < 0.05). Among all predictors, leaf area index and incoming shortwave radiation emerged as the dominant drivers of spatial residual variation, followed by precipitation, relative humidity, and selected vegetation indices. The 20 m resolution yielded the best balance between model performance and computational efficiency. In conclusion, our modeling framework effectively captures both spatial heterogeneity and nonlinear interactions, offering a robust solution for spatially explicit flux modeling in structurally diverse ecosystems beyond the study sites.

footprint model↗
DOE OSTI2021

Effects of digital fringe projection operational parameters on detecting powder bed defects in additive manufacturing

Additive manufacturing is a technology transforming traditional production timelines. Specifically, metal additive manufacturing (MAM) has been increasingly adopted by a variety of industries, not only to prototype, but also to fulfill full production scale applications with much lower lead times. Like any maturing manufacturing technology, developments in verifying and validating processes are necessary to support continuous growth. Due to the complex nature of MAM, part quality and repeatability remain integral challenges that inhibit further adoption of MAM for critical component production. In this study, we present data taken from a developing in-process monitoring system designed to measure and detect powder bed defects (PBDs) in powder bed fusion MAM systems using surface height maps created with structured light illumination. We showcase the feasibility of the monitoring technique for in-process implementation by detecting streak PBDs with varying severities (height, width) created in a lab environment. We present results of powder bed measurements for varying experimental parameters of the structured light system such as illumination angle, illumination pattern, and number of illuminations. We also present an expression used to determine experimental height noise based on input parameters for PBD detection based on the instrument transfer function of the structured light monitoring system for arbitrary pixel intensity noise contributions. In conclusion, with the results of PBD detection across multiple experimental measurement parameters, we provide a best practices approach to in-process implementation of the monitoring system in powder bed fusion manufacturing.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Advancing Measurements and Understanding of the Rate and Structure of Atmospheric Warming

The rate and geographic pattern of atmospheric warming are key indicators of historical climate change and play a role in modulating future changes in Earth’s climate. The past and future evolution of atmospheric temperature is simulated using global climate models while observational estimates of past warming are derived from satellite microwave measurements. Individual climate models simulate widely varying rates of past and future atmospheric warming and most model simulations exhibit greater tropospheric (lowest ~10 km of atmosphere) temperature change than satellite observations between 1979 and 2020. This project examined intermodel differences in the pattern of atmospheric warming and how these differences influence climate feedbacks that can amplify or damp the rate of global surface warming. A key result is that climate model representation of the current climate can influence climate feedbacks and the simulation of future changes in climate. Another focus of this project was to analyze how models respond to different input datasets, such as volcanic aerosols or sea surface temperature (in atmosphere-only simulations). We find that different prescribed inputs can affect simulated changes in atmospheric temperature, even though “input uncertainty” is often unconsidered in model-observational comparisons. A final focus of this research project was to consider the influence of natural internal climate variability on satellite era changes in climate. Although the Earth is warming substantially due to anthropogenic emissions of greenhouse gases, the observed warming rate can be modulated by natural variations in the Earth’s climate. We find that natural climate variability has slowed the rate of tropical tropospheric warming, which explains model-satellite differences in the rate of tropospheric warming.

58 GEOSCIENCES↗
DOE OSTI2023

Robust deep learning framework for constitutive relations modeling

Modeling the full-range deformation behaviors of materials under complex loading and materials conditions is a significant challenge for constitutive relations (CRs) modeling. Here, we propose a general encoder-decoder deep learning framework that can model high-dimensional stress-strain data and complex loading histories with robustness and universal capability. The framework employs an encoder to project high-dimensional input information (e.g., loading history, loading conditions, and materials information) to a lower-dimensional hidden space and a decoder to map the hidden representation to the stress of interest. We evaluated various encoder architectures, including gated recurrent unit (GRU), GRU with attention, temporal convolutional network (TCN), and the Transformer encoder, on two complex stress-strain datasets that were designed to include a wide range of complex loading histories and loading conditions. All architectures achieved excellent test results with an root-mean-square error (RMSE) below 1 MPa. Additionally, we analyzed the capability of the different architectures to make predictions on out-of-domain applications, with an uncertainty estimation based on deep ensembles. The proposed approach provides a robust alternative to empirical/semi-empirical models for CRs modeling, offering the potential for more accurate and efficient materials design and optimization.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Spatially Averaged Ice Contents of Ice-Wedge Polygon Cross-Sections to 3-m Depth, July 2013, Utqiagvik, Alaska

This dataset contains spatially averaged estimates of ice content (volume percent) for two-dimensional cross-sectional profiles (from trough center to trough center to a depth of 3 meters) of low- and flat-centered ice-wedge polygons (three of each type) located near Utqiagvik, Alaska. A combination of soil pits, trenches, and cores were used to describe, sample, and map the cross-section stratigraphy of soil horizons and ice wedges for each polygon at 6 depth intervals. Observed soil horizons were assigned to four types with increasing amounts of organic components (mineral, mineral/organic, organic/mineral, and organic). The average ice contents of each soil horizon type below the permafrost boundary and wedge ice were weighted by their cross-sectional area fractions to calculate spatially averaged estimates of ice content for each polygon. In the active layer, spatially averaged estimates of volumetric water contents were similarly determined and reported here as “ice” content to enable estimates of the soil’s structurally competent porosity and excess ice in permafrost layers. In this dataset, the file AK13_ice_contents.csv includes cross-sectional area fractions and spatially averaged ice contents for soil layers and horizon types at the six depth intervals. The file AK13_permafrost_ice_fractions.csv contains the calculated partitioning of ice contents into pore ice and excess ice fractions for the three permafrost-dominated depth intervals. In addition, there is a data dictionary file for each of these data files and a file-level metadata file. These data were generated by the Department of Energy’s Soil Carbon Response to Environmental Change Scientific Focus Area and were used as inputs to model simulations examining the consequences of thaw-affected subsidence and microtopography change on active layer thickness of low-relief polygonal tundra landscapes in a warming Arctic.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Emulators for Scarce and Noisy Data: Application to Auxiliary-Field Diffusion Monte Carlo for Neutron Matter

Understanding the equation of state (EOS) of pure neutron matter is necessary for interpreting multimessenger observations of neutron stars. Reliable data analyses of these observations require well-quantified uncertainties for the EOS input, ideally propagating uncertainties from nuclear interactions directly to the EOS. This, however, requires calculations of the EOS for a prohibitively larger number of nuclear Hamiltonians, solving the nuclear many-body problem for each one. Quantum Monte Carlo methods, such as auxiliary-field diffusion Monte Carlo (AFDMC), provide precise and accurate results for the neutron matter EOS, but they are very computationally expensive, making them unsuitable for the fast evaluations necessary for uncertainty propagation. Here, we employ parametric matrix models to develop fast emulators for AFDMC calculations of neutron matter and use them to directly propagate uncertainties of coupling constants in the Hamiltonian to the EOS. As these uncertainties include estimates of the effective field theory truncation uncertainty, this approach provides robust uncertainty estimates for use in astrophysical data analyses. In conclusion, this Letter will enable novel applications such as using astrophysical observations to put constraints on coupling constants for nuclear interactions.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

Quapprox: A Framework for Benchmarking the Approximability of Variational Quantum Circuit

Most of the existing quantum neural network models, such as variational quantum circuits (VQCs), are limited in their ability to explore the non-linear relationships in input data. This gradually becomes the main obstacle for it to tackle realistic applications, such as natural language processing, medical image processing, and wireless communications. Recently, there have emerged research efforts that enable VQCs to perform non-linear operations. However, it is still unclear on the approximability of a given VQC (i.e., the order of non-linearity that can be handled by a specified design). In response to this issue, we developed an automated tool designed to benchmark the approximation of a given VQC. The proposed tool will generate a set of synthetic datasets with different orders of non-linearity and train the given VQC on these datasets to estimate their approximability. Our experiments benchmark VQCs with different designs, where we know their theoretic approximability. We then show that the proposed tool can precisely estimate the approximability, which is consistent with the theoretic value, indicating that the proposed tool can be used for benchmarking the approximability of a given quantum circuit for learning tasks.

artificial intelligence↗
DOE OSTI2020

Development of new baseline models for U.S. medium office buildings based on commercial buildings energy consumption survey data

Building energy estimation for the building sector under various scenarios are needed for building energy regulation and policy making. This often starts with representative baselines (either empirical baseline or modeled baseline). Commercial Buildings Energy Consumption Survey (CBECS) data is a widely used empirical baseline for U.S. commercial buildings, but none of the existing baseline model are developed to represent the CBECS data. This paper aims to develop new baseline models for the U.S. medium office buildings, which can produce modeled baselines consistent with the CBECS data. Here, we introduced the methodology to create baseline models and the criteria to evaluate the performance of baseline models. The methodology consists of three phases: (1) identification of model inputs, (2) model calibration, and (3) model validation with uncertainty analysis. The evaluation index is the coefficient of variation of the root-mean-square deviation (CV(RMSD)) of site energy use intensities (EUIs) between the modeled baseline and empirical baseline. Then 30 new baseline models for two vintages (pre- and post-1980) and 15 climate zones were created. The evaluation shows that the CV(RMSD) is lower than 0.05 for the modeled baselines produced by the new baseline models. As a comparison, the CV(RMSD) is higher than 0.1 for the existing modeled baselines generated by DOE Commercial Reference Building Models. Further analysis shows that the new baseline models are able to capture the uncertainties of the representative features of existing buildings.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2020

GridOPTICS/GridPACK

GridPACK is a software framework consisting of a set of modules designed to simplify the development of programs that model the power grid and run on parallel, high performance computing platforms. It also contains several fully developed applications, including powerflow, dynamic simulation, state estimation, Kalman filter analysis (dynamic state estimation), contingency analysis and real time path rating. These applications can be used either standalone or as components in more complicated workflows that combine several different types of application together. The framework modules are available as a combination of libraries and software templates and consist of components for setting up and distributing power grid networks, support for modeling the behavior of individual buses and branches in the network, converting the network models to the corresponding algebraic equations, and parallel routines for manipulating and solving large algebraic systems. The framework also contains a module for distributing tasks evenly amongst computing resources, even if individual tasks vary widely in their execution times. Additional modules support input and output, basic statistical analysis of contingency based calculations, distributed data structures, as well as basic profiling and error management.

Palmer, Bruce↗
DOE OSTI2025

Robustness of Deep Learning Classification to Adversarial Input on GPUs: Asynchronous Parallel Accumulation Is a Source of Vulnerability

The ability of machine learning (ML) classification models to resist small, targeted input perturbations—known as adversarial attacks—is a key measure of their safety and reliability. We show that floating-point non associativity (FPNA) coupled with asynchronous parallel programming on GPUs is sufficient to result in misclassification, without any perturbation to the input. Additionally, we show that this misclassification is particularly significant for inputs close to the decision boundary and that standard adversarial robustness results may be overestimated up to 4.6 when not considering machine-level details. We first study a linear classifier, before focusing on standard Graph Neural Network (GNN) architectures and datasets used in robustness assessments. We develop a novel black-box attack using Bayesian optimization to discover external workloads that can change the instruction scheduling which bias the output of reductions on GPUs and reliably lead to misclassification. Motivated by these results, we present a new learnable permutation (LP) gradient-based approach to learning floating-point operation orderings that lead to misclassifications. The LP approach provides a worst-case estimate in a computationally efficient manner, avoiding the need to run identical experiments tens of thousands of times over a potentially large set of possible GPU states or architectures. Finally, using instrumentation-based testing, we investigate parallel reduction ordering across different GPU architectures under external background workloads, when utilizing multi-GPU virtualization, and when applying power capping. Our results demonstrate that parallel reduction ordering varies significantly across architectures under the first two conditions, substantially increasing the search space required to fully test the effects of this parallel scheduler-based vulnerability. These results and the methods developed here can help to include machine-level considerations into adversarial robustness assessments, which can make a difference in safety and mission critical applications.

Shanmugavelu, Sanjif [Maxeler Technologies, a Groq↗
DOE OSTI2021

KIC 12602250: A Low-amplitude Double-mode Delta Scuti Star with Amplitude Modulation

We report a detailed light-curve analysis of the Kepler target Kepler Input Catalog (KIC) 12602250. The results show that KIC 12602250 is a low-amplitude radial double-mode δ-Scuti star with amplitude modulation. The Fourier analysis of the long cadence data (i.e., Q0—Q17, spanning 1471 days) reveals that the variations of the light curve are dominated by the strongest mode with frequency F0 = 11.6141 d{sup -1}, suggesting that KIC 12602250 is a δ-Scuti star. The other independent mode F1 = 14.9741 d{sup -1} is newly detected. The amplitude of the light variations of KIC 12602250 is ~0.06 mag, which indicates that this is a low-amplitude δ-Scuti star; but the ratio of F0/F1 is estimated as 0.7756, which is typical of HADS, and a slow amplitude growth is detected in F1 and f {sub 3}, which could be due to stellar evolution, suggesting that KIC 12602250 could be a post-main-sequence δ Scuti that is crossing the instability strip for the first time.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Variability in Diurnal and Seasonal Ambient Conditions on Geothermal Plant Performance and Cost

Geothermal plant performance is bounded by the second law efficiency, which accounts for the quantity of exergy that can be converted into useful work. This, in turn, is dependent on the geothermal resource temperature and the temperature of the heat sink (i.e., the ambient temperature). In this study, we show that ambient temperature variability on a diurnal and seasonal basis can affect performance and cost estimations for geothermal plants. We have utilized the updated System Advisor Model (SAM) to assess nine geothermal sites with existing resource capacities across three climate zones. Our analysis shows that both evaporatively-cooled flash and air-cooled binary cycle plants are affected by temperature, with a slightly higher effect in enhanced geothermal system binary sites. By assuming an ambient (wet bulb) temperature baseline of 15.6 degrees C (60 degrees F) and comparing baseline results to those from site-specific data, we observe up to 15% underestimation of plant performance and up to 20% overestimation of cost. These results make a case for the inclusion of location-based weather data as inputs to supply curves that are used in capacity expansion models for the prediction of future geothermal deployment scenarios.

ambient temperature↗
DOE OSTI2023

Feasibility Analysis for the Use of Retrofitted Air-Conditioners Using Thermal Energy Storage (TES) for High Ambient Temperature (HAT) Countries

In high ambient temperature (HAT) countries, summer temperatures exceed 35℃, degrading the performance of air-conditioning systems and straining power grids. In this paper, a feasibility analysis was conducted to investigate potential savings during the peak using latent Thermal Energy Storage (TES) at near room phase-change temperatures, replacing condensers, thus, minimizing temperature lifts. Weather data, building loads, and baseline air-conditioning systems data were gathered for Dubai. A transient vapor-compression model in Modelica was used to compare the COP, total power input and cooling capacity of the air-conditioner at peak hours when ambient temperatures range between 35 – 45 ⁰C versus the TES-Phase-Change Material (PCM) melting temperatures from 22 – 28⁰C. The results indicate that the TES-PCM can enhance the system COP at the peak by a factor 1.4 and 2 for during for outdoor temperatures of 35 – 40⁰C, and 40 – 45⁰C, respectively. Lower melting temperature PCMs were able to reduce the required power input by 30-50%, with more savings occurring at higher temperature days. On the other hand, higher temperature PCMs enhancements were minimal especially at outdoor ambient temperatures ranging between 35 – 40⁰C. Improvements to the cooling capacity range from 8 – 18 % for the outdoor temperature range of 35 – 45 ⁰C. An economic analysis was conducted to find the potential saving in utility costs for 30%, 60%, and 90% of the space cooling demands of Dubai, and find the trade-off points between utility savings and cost of PCM-TES implementation. If the peak loads are to be shifted by 6 hours daily, the percentage utility savings for the city is 18%. Using estimated costs of the PCM-TES, ranging from $200-500/kWh, the daily load shifting hours were estimated to range from 4 hours at the lowest cost systems to 2.5 hours at the highest costs.

25 ENERGY STORAGE↗