Common Engineering Machine Learning Environments
Overview of 9321's Computing Capabilities related to Machine Learning
Hughes, Bryan↗
SEARCH · Search NASA
Results for “computational capability”
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.
Integrative computational investigation of the spectroscopy, dynamics, and controlling of molecular excitons in complex environments (Final Technical Report)
The major goal of this research project was to advance theoretical and computational capability to describe the dynamics and the spectroscopy of molecular excitons as reliably and efficiently as possible. Outcomes of the project contributed to new understanding of the interaction between light and molecular materials and ensuing dynamics of excited electronic states. These can assist experimentalists in developing novel methods to utilize quantum properties of molecules and molecular excitons for new energy harvesting and quantum information processing.
37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
Large Language Model for Validation, Optical Calibration, and Learning (VOCAL) Distributed Temperature Sensing Interface
Distributed temperature sensing (DTS) using fiber optic sensors (FOS) offers a promising method for temperature measurements in advanced reactors, such as sodium fast reactors and molten salt cooled reactors. To support the calibration and validation of DTS measurements, Argonne National Laboratory developed the Validation, Optical Calibration, and Learning (VOCAL) software package. This report describes the integration of a local large language model (LLM) with a retrieval-augmented generation (RAG) system into the VOCAL interface to serve as an interactive user assistant. The LLM framework enhances the VOCAL platform’s accessibility to users by explaining interface components, clarifying inputs and outputs, and answering user queries dynamically in real-time. The accuracy of the LLM assistant performance was evaluated with 20 queries regarding the interface and its parameters using experimental data from the Thermal Hydraulic Experimental Test Article (THETA) facility. Results demonstrate that the LLM achieved a 95% accuracy rate, with a BERTScore of 0.8816 and SBERT value of 0.7417. Furthermore, validation of the RAG system within the LLM framework showed optimal accuracy with k-values between 1 and 2 using the k-refinement convergence test. The prompt perturbation analysis demonstrated good initial consistency for the RAG system, exhibiting the highest accuracy under punctuation variations and the greatest sensitivity under query reordering. Notably, the model’s errors were limited to data retrieval failures rather than factual hallucinations, reinforcing its baseline reliability. The integration of LLM provides a highly accurate, userfriendly enhancement to the VOCAL platform without disrupting its core computational capabilities for FOS calibration and validation.
Hong, Evan↗
Scale-up Unlearnable Examples Learning with High-performance Computing
Recent advancements in AI models, like ChatGPT, are structured to retain user interactions, which could inadvertently include sensitive healthcare data. In the healthcare field, particularly when radiologists use AI-driven diagnostic tools hosted on online platforms, there is a risk that medical imaging data may be repurposed for future AI training without explicit consent, spotlighting critical privacy and intellectual property concerns around healthcare data usage. Addressing these privacy challenges, a novel approach known as Unlearnable Examples (UEs) has been introduced, aiming to make data unlearnable to deep learning models. A prominent method within this area, called Unlearnable Clustering (UC), has shown improved UE performance with larger batch sizes but was previously limited by computational resources (e.g., a single workstation). To push the boundaries of UE performance with theoretically unlimited resources, we scaled up UC learning across various datasets using Distributed Data Parallel (DDP) training on the Summit supercomputer. Our goal was to examine UE efficacy at high-performance computing (HPC) levels to prevent unauthorized learning and enhance data security, particularly exploring the impact of batch size on UE’s unlearnability. Utilizing the robust computational capabilities of the Summit, extensive experiments were conducted on diverse datasets such as Pets, MedMNist, Flowers, and Flowers102. Our findings reveal that both overly large and overly small batch sizes can lead to performance instability and affect accuracy. However, the relationship between batch size and unlearnability varied across datasets, highlighting the necessity for tailored batch size strategies to achieve optimal data protection. The use of Summit’s high-performance GPUs, along with the efficiency of the DDP framework, facilitated rapid updates of model parameters and consistent training across nodes. Our results underscore the critical role of selecting appropriate batch sizes based on the specific characteristics of each dataset to prevent learning and ensure data security in deep learning applications. The source code is publicly available at https: // github. com/ hrlblab/ UE_ HPC .
Zhu, Yanfan [Vanderbilt University, Nashville, TN,↗
Network Security Challenges and Countermeasures for Software-Defined Smart Grids: A Survey
The rise of grid modernization has been prompted by the escalating demand for power, the deteriorating state of infrastructure, and the growing concern regarding the reliability of electric utilities. The smart grid encompasses recent advancements in electronics, technology, telecommunications, and computer capabilities. Smart grid telecommunication frameworks provide bidirectional communication to facilitate grid operations. Software-defined networking (SDN) is a proposed approach for monitoring and regulating telecommunication networks, which allows for enhanced visibility, control, and security in smart grid systems. Nevertheless, the integration of telecommunications infrastructure exposes smart grid networks to potential cyberattacks. Unauthorized individuals may exploit unauthorized access to intercept communications, introduce fabricated data into system measurements, overwhelm communication channels with false data packets, or attack centralized controllers to disable network control. An ongoing, thorough examination of cyber attacks and protection strategies for smart grid networks is essential due to the ever-changing nature of these threats. Previous surveys on smart grid security lack modern methodologies and, to the best of our knowledge, most, if not all, focus on only one sort of attack or protection. This survey examines the most recent security techniques, simultaneous multi-pronged cyber attacks, and defense utilities in order to address the challenges of future SDN smart grid research. The objective is to identify future research requirements, describe the existing security challenges, and highlight emerging threats and their potential impact on the deployment of software-defined smart grid (SD-SG).
24 POWER TRANSMISSION AND DISTRIBUTION↗
Choice of Tuning Parameters on 3D IC Engine Simulations Using G-Equation
3D CFD spark-ignition IC engine simulations are extremely complex for the regular user. Truly-predictive CFD simulations for the turbulent flame combustion that solve fully coupled transport/chemistry equations may require large computational capabilities unavailable to regular CFD users. A solution is to use a simpler phenomenological model such as the G-equation that decouples transport/chemistry result. Such simulation can still provide acceptable and faster results at the expense of predictive capabilities. While the G-equation is well understood within the experienced modeling community, the goal of this paper is to document some of them for a novice or less experienced CFD user who may not be aware that phenomenological models of turbulent flame combustion usually require heavy tuning and calibration from the user to mimic experimental observations. This study used ANSYS® Forte, Version 17.2, and the built-in G-equation model, to investigate two tuning constants that influence flame propagation in 3D CFD SI engine simulations: the stretch factor coefficient, Cms and the flame development coefficient, Cm2. After identifying several Cm2-Cms pairs that matched experimental data at one operating conditions, simulation results showed that engine models that used different Cm2-Cms sets predicted similar combustion performance, when the spark timing, engine load, and engine speed were changed from the operating condition used to validate the CFD simulation. A dramatic shift was observed when engine speed was doubled, which suggested that the flame stretch coefficient, Cms, had a much larger influence at higher engine speeds compared to the flame development coefficient, Cm2. Therefore, the Cm2-Cms sets that predicted a higher turbulent flame under higher in-cylinder pressure and temperature increased the peak pressure and efficiency. This suggest that the choice of the Cm2-Cms will affect the G-equation-based simulation accuracy when engine speed increases from the one used to validate the model. As a result, for the less-experienced CFD user and in the absence of enough experimental data that would help retune the tuning parameters at various operating conditions, the purpose of a good G-equation-based 3D engine simulation is to guide and/or complement experimental investigations, not the other way around. Only a truly-predictive simulation that fully couples the turbulence/chemistry equations can help reduce the amount of experimental work.
42 ENGINEERING↗
Scientific challenges to characterizing the wind resource in the marine atmospheric boundary layer
Abstract. With the increasing level of offshore wind energy investment, it is correspondingly important to be able to accurately characterize the wind resource in terms of energy potential as well as operating conditions affecting wind plant performance, maintenance, and lifespan. Accurate resource assessment at a particular site supports investment decisions. Following construction, accurate wind forecasts are needed to support efficient power markets and integration of wind power with the electrical grid. To optimize the design of wind turbines, it is necessary to accurately describe the environmental characteristics, such as precipitation and waves, that erode turbine surfaces and generate structural loads as a complicated response to the combined impact of shear, atmospheric turbulence, and wave stresses. Despite recent considerable progress both in improvements to numerical weather prediction models and in coupling these models to turbulent flows within wind plants, major challenges remain, especially in the offshore environment. Accurately simulating the interactions among winds, waves, wakes, and their structural interactions with offshore wind turbines requires accounting for spatial (and associated temporal) scales from O(1 m) to O(100 km). Computing capabilities for the foreseeable future will not be able to resolve all of these scales simultaneously, necessitating continuing improvement in subgrid-scale parameterizations within highly nonlinear models. In addition, observations to constrain and validate these models, especially in the rotor-swept area of turbines over the ocean, remains largely absent. Thus, gaining sufficient understanding of the physics of atmospheric flow within and around wind plants remains one of the grand challenges of wind energy, particularly in the offshore environment. This paper provides a review of prominent scientific challenges to characterizing the offshore wind resource using as examples phenomena that occur in the rapidly developing wind energy areas off the United States. Such phenomena include horizontal temperature gradients that lead to strong vertical stratification; consequent features such as low-level jets and internal boundary layers; highly nonstationary conditions, which occur with both extratropical storms (e.g., nor'easters) and tropical storms; air–sea interaction, including deformation of conventional wind profiles by the wave boundary layer; and precipitation with its contributions to leading-edge erosion of wind turbine blades. The paper also describes the current state of modeling and observations in the marine atmospheric boundary layer and provides specific recommendations for filling key current knowledge gaps.
17 WIND ENERGY↗
gpCAM v8
gpCAM is a Python software for large-scale Gaussian-Process driven uncertainty quantification, Bayesian Optimization, and Autonomous Experimentation. It is designed with maximum flexibility and customizability while offering record-breaking computing capabilities. In 2023 gpCAM broke the world record for Gaussian processes on large datasets set in 2019. Through clever programming and math, Gaussian Process building blocks, such as prior mean, kernel, and noise functions maintain their mathematical flexibility while supporting acceleration and scaling. This flexibility means that the approximation and uncertainty quantification can be domain-aware and extended over exotic input and output spaces.
Noack, Marcus↗
Bacterial Identification Using Light Scattering Measurements: a Preliminary Report
The light scattering properties of single bacterial cells were examined as a possible means of identification. Three species were studied with streptococcus faecalis exhibiting a unique pattern; the light-scattering traces for staphylococcus aureus and escherichia coli were quite similar although differences existed. Based on preliminary investigations, the light scattering approach appeared promising with additional research needed to include a wide variety of bacterial species, computer capability to handle and analyze data, and expansion of light scattering theory to include bacterial cells.
Wilkins, J. R.↗
Study of the dynamics of orbital assemblies including interactions with geometrical appendages. Unified flexible spacecraft load program (LOAD): Final report and user's and operation manual
The addition of a dynamic loads computation capability to the Unified Flexible Spacecraft Simulation (UFSS) program is discussed. The added capability provides a means for determining the internal member loads due to the time-varying external loading conditions experienced by an orbiting spacecraft/cluster. The flexible bodies are modeled as a system of joints or nodes which are interconnected by weightless finite element members. All masses are lumped at the joints. The orthogonal functions used to describe the spatial deformation of the bodies are normally taken to be the orthonormal cantilever modes produced by a standard structural dynamics program.
Ness, D. J.↗
Modular digital computer system design
Automatically-Reconfigurable Modular Multiprocessor System (ARMMS) provides redundant processing with dynamic mode switching in real time. Design will provide higher computer capability than that presently available for same amount of hardware and will furnish modular system which is responsive to diverse problems effectively.
Source record↗
Study to design and develop remote manipulator system
Modeling of human performance in remote manipulation tasks is reported by automated procedures using computers to analyze and count motions during a manipulation task. Performance is monitored by an on-line computer capable of measuring the joint angles of both master and slave and in some cases the trajectory and velocity of the hand itself. In this way the operator's strategies with different transmission delays, displays, tasks, and manipulators can be analyzed in detail for comparison. Some progress is described in obtaining a set of standard tasks and difficulty measures for evaluating manipulator performance.
Hill, J. W.↗
Sediment concentration mapping in tidal estuaries
An analytical procedure has been developed that considers the ERTS-1 multispectral scanner as a reflectance spectrophotometer. ADP techniques requiring only very limited computer capability are utilized to search the data defining the spectral reflectance characteristics of a scene on a pixel-by-pixel basis, identify each pixel whose spectral reflectance matches a reference spectrum, and generate maps that identify pixel locations where spectrum matches occur and identify the spectrum that was matched. If the reference spectra are known to represent a specific condition on the ground, a map of the distribution of that condition can be output as a dimensionally accurate overlay to maps of any selected scale. Two applications are described: (1) mapping the distribution of water masses exhibiting specific suspended sediment concentrations; and (2) determining the location and delineation of surface water bodies in the southeastern U. S. The techniques described are being successfully used to map the land area inundated by the 1973 spring flood in the Lower Mississippi River Valley, map sediment distributions in Lake Pontchartrain (in Louisiana) as a result of opening the Bonnet Carre Floodway during the spring flood, and inventory lakes and reservoirs.
Williamson, A. N.↗
Preentry communication design elements for outer planets atmospheric entry probe
Four related tasks are discussed for data transmission from a probe prior to entering the atmosphere of Jupiter to an orbiting spacecraft in a trajectory past the planet: (1) link analysis and design; (2) system conceptual design; (3) Doppler measurement analysis; and (4) an electronically despun antenna. For tasks 1, 3, and 4, an analytical approach was developed and combined with computational capability available to produce quantitative results corresponding to requirements and constraints given by NASA, ARC. One constraint having a major impact on the numerical results of the link analysis was the assumption of a nonsteerable antenna on a spinning orbiter. Other constraints included the interplanetary trajectory and the approach trajectory. Because the Jupiter Orbiter Probe (JOP) program is currently in a state of evolution, all requirements and constraints applied during this study are subject to change. However, the relationships of parameters as developed will remain valid and will aid in planning Jupiter missions.
Source record↗
Full employment maintenance in the private sector
Operationally, full employment can be accomplished by applying modern computer capabilities, game and decision concepts, and communication feedback possibilities, rather than accepted economic tools, to the problem of assuring invariant full employment. The government must provide positive direction to individual firms concerning the net number of employees that each firm must hire or refrain from hiring to assure national full employment. To preserve free enterprise and the decision making power of the individual manager, this direction must be based on each private firm's own numerical employment projections.
Young, G. A.↗
Cell-level battery charge/discharge protection system
The paper describes three design approaches to individual cell monitoring and control for sealed secondary battery cells. One technique involves a modular strap-on single cell protector which contains all the electronics required for monitoring cell voltage, responding to external commands, and forming a bypass circuit for the cell. A second technique, the multiplexed cell protector, uses common circuitry to monitor and control each cell in a battery pack. The third technique, the computerized cell protector, by replacing the hard-wired logic of the multiplexed cell protector with a microprocessor, achieves greatest control flexibility and inherent computational capability with a minimum parts count implementation.
Donovan, R. L.↗
Gigaflop architecture, a hardware perspective
Any super computer built in the early 1980s will use components that are available by fall 1978. The architecture of such a system cannot depart radically from current super computers if the software experience painfully acquired from these computers in the 70's is to apply. Given the above constraints, 10 billion floating point operations per second (BFLOPS) are attainable and a problem memory of 512 million (64 bit) words could be supported by the technology of the time. In contrast to this, industry is likely to respond with commercially available machines with a performance of less than 150 MFLOPS. This is due to self-imposed constraints on the manufacturers to provide upward compatible architectures (same instruction set) and systems which can be sold in significant volumes. Since this computing speed is inadequate to meet the demands of computational fluid dynamics, a special processor is required. Issues which are felt to be significant in the pursuit of maximum compute capability in this special processor are discussed.
Feierbach, G. F.↗
Interactive data line study and experiment
The paper considers the utility of having remote terminals on a Landsat image processing system and the development of efficient data transfer strategies such as data compression. The approach is to link an image processing terminal (located in a mobile van) to a multiuser host computer via the Communications Technology Satellite and a leased phone line. The host computer will share time between the remote state users and local users. The terminal includes a minicomputer, a color video display, and a communications interface, and the user has access to all of the resources of the host facility via the data link. While at present the terminal acts in the display-only mode, the computing capability of the terminal can be varied from display-only up to a stand-alone image processing system. An analysis of trade-offs between data link and remote terminal capacities is described with attention to the evaluation of the system by state users and the selection of a system architecture suitable to unique needs.
Rasmussen, D. N.↗