Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “training data”

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 1,315 records · Page 73
All sourcesNASA NTRSDOE OSTI
NASA NTRSDate not supplied

GT2024-128885: Flow Reconstruction in a Transonic Turbine Cascade using Physics-Informed Neural Networks (PINNs)

This presentation investigates the application of Physics-Informed Neural Networks (PINNs) for the analysis of turbine blades in a transonic cascade. PINNs are a machine learning method trained on losses calculated from reconstructed governing equations, assigned boundary/initial conditions, and measured data. We reconstruct the 2-D flow field in a transonic turbine cascade in two ways: the traditional forward approach (without training/experimental data) and by training the PINN using experimental data. We then compare the PINN solutions to measured data. This is repeated for three different turbine blades with distinct loading characteristics. The experimental data used for training is the static pressure measurements along the suction and pressure sides of each blade. The PINN is trained utilizing all available data, half the available data, data from only the leading edge region, and data from only the trailing edge region. It's shown that the PINN can reconstruct the flow field in all cases with acceptable errors. Cases where the PINN is trained on all the data, and even half the data, resulted in the lowest errors. The exit Mach number is inferred for each case and compared to the experimentally calculated value.

Machine Learning↗
NASA NTRS1979

The influence of false color infrared display on training field identification

The overall success of large-scale crop inventories of agricultural regions using Landsat multispectral scanner data is highly dependent upon the labeling of training data by analyst/photointerpreters. The principal analyst tool in labeling training data is a false color infrared composite of Landsat bands 4, 5, and 7. In this paper, this color display is investigated and its influence upon classification errors is partially determined.

Coberly, W. A.↗
NASA NTRSDate not supplied

Ice-Over-Water Cloud Identification in an Artificial Neural Network Approach

An artificial neural network (ANN) algorithm, employing several Aqua MODIS channels, the retrieved cloud phase and total cloud visible optical depth, and temperature and humidity vertical profiles is trained to detect multilayer (ML) ice-over-water cloud systems identified by matched 2008 CloudSat and CALIPSO (CC) data. The trained MLANN was applied to 2009 MODIS data resulting in combined ML and single layer detection accuracies of 87% (89%) and 86% (89%) for snow-free (snow-covered) regions during the day and night, respectively. When corrected for the viewing-zenith-angle dependence of each parameter, the ML fraction detected is relatively invariant across the swath. Compared to the CC ML variability, the MLANN is robust seasonally and interannually, and produces similar distribution patterns over the globe, except in the polar regions. Additional research is needed to conclusively evaluate the VZA dependence and further improve the MLANN accuracy. This approach should greatly improve the monitoring of cloud vertical structure using operational passive sensors.

MODIS↗
DOE OSTI2021

Enriching Load Data Using Micro-PMUs and Smart Meters

In modern distribution systems, load uncertainty can be fully captured by micro-PMUs, which can record high-resolution data; however, in practice, micro-PMUs are installed at limited locations in distribution networks due to budgetary constraints. In contrast, smart meters are widely deployed but can only measure relatively low-resolution energy consumption, which cannot sufficiently reflect the actual instantaneous load volatility within each sampling interval. In this paper, we have proposed a novel approach for enriching load data for service transformers that only have low-resolution smart meters. The key to our approach is to statistically recover the high-resolution load data, which is masked by the low-resolution data, using trained probabilistic models of service transformers that have both high- and low-resolution data sources, i.e., micro-PMUs and smart meters. The overall framework consists of two steps: first, for the transformers with micro-PMUs, a Gaussian Process is leveraged to capture the relationship between the maximum/minimum load and average load within each low-resolution sampling interval of smart meters; a Markov chain model is employed to characterize the transition probability of known high-resolution load. Next, the trained models are used as teachers for the transformers with only smart meters to decompose known low-resolution load data into targeted high-resolution load data. The enriched data can recover instantaneous load uncertainty and significantly enhance distribution system observability and situational awareness. Here, we have verified the proposed approach using real high- and low-resolution load data.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2018

Efficient Generalizable Deep Learning

Stochastic optimization is a fundamental field of research for machine learning. Stochastic gradient descent (SGD) and related methods provide a feasible means to train complicated prediction models over large datasets. SGD, however, does not explicitly address the problem of overfitting, which can lead to predictions that perform poorly on new data. This difference between loss performance on unseen testing data verses that of training data defines the generalization gap of a model. We introduce a new computational kernel called Stochastic Hessian Projection (SHP) that uses a maximum likelihood framework to simultaneously estimate gradient noise covariance and local curvature of the loss function. Our analysis illustrates that these quantities affect the evolution of parameter uncertainty and therefore generalizability. We show how these computations allow us to predict the generalization gap without requiring holdout data. Explicitly assessing this metric for generalizability during training may improve machine learning predictions when data is scarce and understanding prediction variability is critical.

42 ENGINEERING↗
DOE OSTI2022

Use of machine learning for a helium line intensity ratio method in Magnum-PSI

Optical emission spectroscopy (OES) of helium (He) line intensities has been used to measure the electron density, ne, and temperature, Te, in various plasma devices. In this study, a neural network with five hidden layers is introduced to model the relation between the OES data and ne/Te from laser Thomson scattering in the linear plasma device Magnum-PSI and compared to multiple regression analysis. It is shown that the neural network reduces the residual errors of prediction values (ne and Te) less than half those of the multiple regression analysis. We checked two different data splitting methods for training and validation data, i.e., with and without considering the unit of discharge. A comparison of the splitting methods suggests that the residual error will decrease to ~10% even for a new discharge data when accumulating a sufficient data set.

Kajita, Shin↗
DOE OSTI2022

Equation‐Free Surrogate Modeling of Geophysical Flows at the Intersection of Machine Learning and Data Assimilation

Abstract There is a growing interest in developing data‐driven reduced‐order models for atmospheric and oceanic flows that are trained on data obtained either from high‐resolution simulations or satellite observations. The data‐driven models are non‐intrusive in nature and offer significant computational savings compared to large‐scale numerical models. These low‐dimensional models can be utilized to reduce the computational burden of generating forecasts and estimating model uncertainty without losing the key information needed for data assimilation (DA) to produce accurate state estimates. This paper aims at exploring an equation‐free surrogate modeling approach at the intersection of machine learning and DA in Earth system modeling. With this objective, we introduce an end‐to‐end non‐intrusive reduced‐order modeling (NIROM) framework equipped with contributions in modal decomposition, time series prediction, optimal sensor placement, and sequential DA. Specifically, we use proper orthogonal decomposition (POD) to identify the dominant structures of the flow, and a long short‐term memory network to model the dynamics of the POD modes. The NIROM is integrated within the deterministic ensemble Kalman filter (DEnKF) to incorporate sparse and noisy observations at optimal sensor locations obtained through QR pivoting. The feasibility and the benefit of the proposed framework are demonstrated for the NOAA Optimum Interpolation Sea Surface Temperature (SST) V2 data set. Our results indicate that the NIROM is stable for long‐term forecasting and can model dynamics of SST with a reasonable level of accuracy. Furthermore, the prediction accuracy of the NIROM gets improved by almost one order of magnitude by the DEnKF algorithm.

Pawar, Suraj↗
DOE OSTI2025

A Comparative Study of Physics‐Informed and Data‐Driven Neural Networks for Compound Flood Simulation at River‐Ocean Interfaces: A Case Study of Hurricane Irene

Simulating compound flooding (CF) at the river-ocean interface within large-scale Earth System Models (ESMs) presents significant challenges due to complex interactions between river discharge, storm surge, and tides. This study assesses the comparative advantages of physics-informed and data-driven machine learning (ML) approaches for enhancing local ESM performance. We systematically compare data-driven neural network models (i.e., CNNs, U-Net, Long Short-Term Memory (LSTM), Gated Recurrent Unit), and physics-informed neural network (PINN) models, including vanilla PINN and a finite-difference-based PINN (FD-PINN). Specifically, FD-PINN is introduced to enhance computational efficiency, accelerating vanilla PINNs by ∼6.5 times while improving accuracy. To enhance data-driven model training, a new data-generation approach is developed to sample historical fluvial and coastal flood events, which ensures a robust data set for extreme event prediction. The models are evaluated using a realistic one-dimensional river domain extracted from an ESM's river mesh and the Hurricane Irene event as an independent test case. Results show that FD-PINN achieves accurate predictions with significantly reduced computational costs relative to vanilla PINNs. Among data-driven models, the best overall performance is achieved by a CNN-LSTM hybrid, which balances accuracy and efficiency. While a fully connected CNN (CNN-FC) provides the best accuracy, it incurs high computational cost. Architectures lacking strong temporal modeling tend to underperform on unseen events. These findings highlight the importance of sequence-aware designs for robust generalization. This study reveals the trade-offs between physics-informed and data-driven models and proposes an adaptive hybrid framework for integrating ML into ESMs to enhance local flood simulations.

Earth Systems Modeling↗
DOE OSTI2023

Harnessing Uncertainty through Functional Data Analysis in Gas Breakthrough Data

Detecting subsurface explosions from radionuclide gas migration through rock fractures is an effective way to identify nuclear activity. Los Alamos National Laboratory (LANL) has developed simulation methods, based on data from the 1962 Hardhat underground nuclear test, to predict gas breakthrough times at the surface. However, these methods rely on an imperfect understanding of the relationship between rock damage and fracture permeability. Our clinic project studies methods for predicting breakthrough curves that characterize total mass produced as a function of time, as well as quantifying the uncertainty associated with these predictions. The model that is currently employed to relate damage to permeability uses an empirically motivated power-law expression, with a range of parameter values that are compatible with the experimental Hardhat data. We develop emulators, built from functional data analysis techniques and trained on simulation data, that rapidly predict the gas breakthrough curve given a damage field and given the parameter values of the power-law equation. Using Bayesian regression, we address the problem of uncertainty quantification in our emulators. Finally, in order to test the robustness of the model, we further validate it on a damage field representing different physical conditions.

58 GEOSCIENCES↗
NASA NTRS2006

Comparison of Bystander Cardiopulmonary Resuscitation (BCPR) Performance in the Absence and Presence of Timing Devices for Coordinating Delivery of Ventilatory Breaths and Cardiac Compressions in a Model of Adult Cardiopulmonary Arrest

Astronaut crew medical officers (CMO) aboard the International Space Station (ISS) receive 40 hours of medical training during the 18 months preceding each mission. Part of this training ilncludes twoperson cardiopulmonary resuscitation (CPR) per training guidelines from the American Heart Association (AHA). Recent studies concluded that the use of metronomic tones improves the coordination of CPR by trained clinicians. Similar data for bystander or "trained lay people" (e.g. CMO) performance of CPR (BCPR) have been limited. The purpose of this study was to evailuate whether use of timing devices, such as audible metronomic tones, would improve BCPR perfomance by trained bystanders. Twenty pairs of bystanders trained in two-person BCPR performled BCPR for 4 minutes on a simulated cardiopulmonary arrest patient using three interventions: 1) BCPR with no timing devices, 2) BCPR plus metronomic tones for coordinating compression rate only, 3) BCPR with a timing device and metronome for coordinating ventilation and compression rates, respectively. Bystanders were evaluated on their ability to meet international and AHA CPR guidelines. Bystanders failed to provide the recommended number of breaths and number of compressions in the absence of a timing device and in the presence of audible metronomic tones for only coordinating compression rate. Bystanders using timing devices to coordinate both components of BCPR provided the reco number of breaths and were closer to providing the recommended number of compressions compared with the other interventions. Survey results indicated that bystanders preferred to use a metronome for delivery of compressions during BCPR. BCPR performance is improved by timing devices that coordinate both compressions and breaths.

Hurst, Victor, IV↗
DOE OSTI2024

Neural Network Prediction of Strong Lensing Systems with Domain Adaptation and Uncertainty Quantification

Modeling strong gravitational lenses is computationally expensive for the complex data from modern and next-generation cosmic surveys. Deep learning has emerged as a promising approach for finding lenses and predicting lensing parameters, such as the Einstein radius. Mean-variance Estimators (MVEs) are a common approach for obtaining aleatoric (data) uncertainties from a neural network prediction. However, neural networks have not been demonstrated to perform well on out-of-domain target data successfully - e.g., when trained on simulated data and applied to real, observational data. In this work, we perform the first study of the efficacy of MVEs in combination with unsupervised domain adaptation (UDA) on strong lensing data. The source domain data is noiseless, and the target domain data has noise mimicking modern cosmology surveys. We find that adding UDA to MVE increases the accuracy on the target data by a factor of about two over an MVE model without UDA. Including UDA also permits much more well-calibrated aleatoric uncertainty predictions. Advancements in this approach may enable future applications of MVE models to real observational data.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2025

Finding Real Uncertainties From Physical Simulations

Modeling strong gravitational lenses is computationally expensive for the complex data from modern and next-generation cosmic surveys. Deep learning has emerged as a promising approach for finding lenses and predicting lensing parameters, such as the Einstein radius. Mean-variance Estimators (MVEs) are a common approach for obtaining aleatoric (data) uncertainties from a neural network prediction. However, neural networks have not been demonstrated to perform well on out-of-domain target data successfully - e.g., when trained on simulated data and applied to real, observational data. In this work, we perform the first study of the efficacy of MVEs in combination with unsupervised domain adaptation (UDA) on strong lensing data. The source domain data is noiseless, and the target domain data has noise mimicking modern cosmology surveys. We find that adding UDA to MVE increases the accuracy on the target data by a factor of about two over an MVE model without UDA. Including UDA also permits much more well-calibrated aleatoric uncertainty predictions. Advancements in this approach may enable future applications of MVE models to real observational data.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Data Efficiency Assessment of Generative Adversarial Networks for Critical Heat Flux Synthetic Data Generation

This study investigates the application of generative artificial intelligence techniques, particularly conditional generative adversarial networks (cGAN), in real-world engineering contexts, with a specific focus on synthetic data generation for critical heat flux (CHF). Utilizing a dataset comprising more than 20,000 real experimental CHF measurements, we conduct a series of experiments to examine cGAN’s behavior. These experiments encompass varying sizes of the training dataset, training cGAN on data from diverse experimental sources to generate new data on unseen experimental setups, and assessing the impact of excluding various input features on cGAN’s data generation accuracy. Our findings underscore the pronounced data dependency of cGAN for reliable performance, with decreased efficacy observed with smaller training dataset sizes. Notably, cGAN exhibits varying performance when trained on data from different experiments, with superior predictive capabilities observed for certain experiment sources compared to others. For instance, when cGAN was trained on data from Smolin et al.’s experiments or Zenkevich et al., it exhibited relatively good performance in generating the data from Becker et al., Kirillov et al., and Alekseev et al. experiments. In contrast, when trained with Alekseev et al.’s data and tasked with generating other experimental setups, cGAN showed notably poor performance. In both scenarios, cGAN’s performance was inferior compared to training on samples from all experiments concurrently. A feature importance analysis highlights the significant influence of parameters such as mass flux and heated length on accurate CHF generation, while other parameters like diameter and pressure have less impact. Inlet temperature is identified as a moderating factor by cGAN.

22 - GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2024

An Autonomous Critical Data Extrapolator for the AGN-201m

Nuclear nonproliferation serves as a key goal, being undertaken by the International Atomic Energy Agency (IAEA). To recognize proliferation there are two pathways that states, who intend to use nuclear material for malicious purposes can take, diversion can misuse. Diversion is when fissile nuclear material is declared to the IAEA for non-weapon purposes, but then covertly removed. If the source of nuclear material, that is not declared and not fissionable, is placed inside the reactor core to create fissile material used to create weapons then the state is using the second pathway of proliferation, misuse. With the emerging development in areas of simulation and machine learning the creation of virtual models of reactor systems, digital twins, serve as a potential method to identify proliferation through detecting anomalous behavior in the reactor. A digital twin for a physical nuclear reactor has never been developed, as digital twins serve as an emerging technology. To investigate the process for development and use of a digital twin for a nuclear reactor Idaho State University’s AGN-201m serves as the nuclear reactor used for development of this digital twin. A data acquisition system has been installed to the reactor system allowing for the transfer of collected data from a reactor operation to Idaho National Laboratory’s Deeplynx data warehouse. When utilizing data to train reactor physics and machine learning models, a significant challenge encountered is the initial state of the data. Nuclear proliferation will have the capacity to be detected when the reactor immediately starts up, nor will it occur after the reactor shuts down. Generally, it will be detected when the reactor is operating at some desired power over a sufficient period for that specific reactor design. For the AGN-201m this will be when the reactor is critical (generally 1 mW or above) for a timespan that is within or less than the range of a regular business day. Datasets sent to Deeplynx have had to be manually cut to when the reactor is critical based on plots of power levels. This method is inefficient and laborious, especially when using multiple datasets at once to train a model. To provide a more streamlined approach an automated critical data extrapolator is developed, with capabilities of recognizing when the reactor operation first reaches criticality, and when the reactor undergoes a SCRAM and is shutdown.

99 GENERAL AND MISCELLANEOUS↗
DOE OSTI2019

Single-Net Continual Learning with Progressive Segmented Training

There is an increasing need of continual learning in dynamic systems, such as the self-driving vehicle, the surveillance drone, and the robotic system. Such a system requires learning from the data stream, training the model to preserve previous information and adapt to a new task, and generating a single-headed vector for future inference. Different from previous approaches with dynamic structures, this work focuses on a single network and model segmentation to prevent catastrophic forgetting. Leveraging the redundant capacity of a single network, model parameters for each task are separated into two groups: one important group which is frozen to preserve current knowledge, and secondary group to be saved (not pruned) for a future learning. A fixed-size memory containing a small amount of previously seen data is further adopted to assist the training. Without additional regularization, the simple yet effective approach of Progressive Segmented Training (PST) successfully incorporates multiple tasks and achieves the state-of-the-art accuracy in the single-head evaluation on CIFAR-10 and CIFAR-100 datasets. Moreover, the segmented training significantly improves computation efficiency in continual learning at the edge.

Du, Xiaocong↗
NASA NTRSDate not supplied

Earth-Independent Medical Operations (EIMO) Concept of Operations

In contrast to the current crew health paradigm for low-Earth orbit and Lunar missions, which depends on real-time communication with Mission Control, deep-space exploration missions will require a significant shift in medical operations. This shift is driven by the constraints of operating at a considerable distance from Earth, such as resource limitations—lack of resupply, restricted mass, power, volume, and data—as well as communication delays and the inability to evacuate back to Earth during emergencies. To move toward a more self-reliant medical model, a strategy is needed to gradually increase space-based crew autonomy and reduce risks to mission success in the challenging environment of deep space. This transformative change, known as "Earth-Independent Medical Operations" (EIMO), explores the gradual transfer of medical care and decision-making from Earth-based support to space-based systems. The goal of this transition is to enhance astronaut health and performance while minimizing mission risks. EIMO requires the development of a medical system that integrates seamlessly with mission planning, vehicle and spacesuit design, and data architecture. This integration is crucial for building a robust medical infrastructure that not only safeguards astronaut well-being but also ensures overall mission success. The Human Research Program (HRP) Exploration Medical Capability (ExMC) Element has revised the EIMO model-based Concept of Operations (ConOps) which outlines an initial vision for EIMO. The ConOps, which is built on the stakeholders’ need, system goals, and objectives (NGOs), presents an array of in-mission scenarios that span a wide range of medical conditions demonstrating the system’s capabilities from basic to complex events. Developed by a multidisciplinary team of systems engineers, scientists, and clinicians within ExMC, the ConOps revision includes two new scenarios(Barotrauma and Self-Medical Management and Behavioral Health and Chronic Medical Care), and implementation of findings from EIMO technical interchange meetings that focused on data and training. The envisioned EIMO Medical System (MS) operates as a system of systems, gathering data from various sources such as reference databases, real-time wearable sensors, point-of-care diagnostics, and environmental controls. The MS also incorporates advanced training tools to support autonomous medical care, assisting the Crew Medical Officer (CMO) during medical events where Ground Support is either unavailable or communication-delayed beyond practicality. Furthermore, MS functions and capabilities were decomposed from the scenarios to establish foundational requirements for EIMO and traced to the NASA Spaceflight Human-System Standard(NASA-STD-3001, Volumes 1 and 2). These traces were performed to gain insights on the alignment of EIMO requirements with the NASA standard. This work serves as an initial recommendation to increase crew autonomy gradually and safely for Mars missions and future deep-space exploration.

medical system↗
DOE OSTI2016

An analysis of image storage systems for scalable training of deep neural networks

This study presents a principled empirical evaluation of image storage systems for training deep neural networks. We employ the Caffe deep learning framework to train neural network models for three different data sets, MNIST, CIFAR-10, and ImageNet. While training the models, we evaluate five different options to retrieve training image data: (1) PNG-formatted image files on local file system; (2) pushing pixel arrays from image files into a single HDF5 file on local file system; (3) in-memory arrays to hold the pixel arrays in Python and C++; (4) loading the training data into LevelDB, a log-structured merge tree based key-value storage; and (5) loading the training data into LMDB, a B+tree based key-value storage. The experimental results quantitatively highlight the disadvantage of using normal image files on local file systems to train deep neural networks and demonstrate reliable performance with key-value storage based storage systems. When training a model on the ImageNet dataset, the image file option was more than 17 times slower than the key-value storage option. Along with measurements on training time, this study provides in-depth analysis on the cause of performance advantages/disadvantages of each back-end to train deep neural networks. We envision the provided measurements and analysis will shed light on the optimal way to architect systems for training neural networks in a scalable manner.

Lim, Seung-Hwan↗
DOE OSTI2019

Identification of Plasma Current Center by Neural Network Inference in EAST

For the efficient and safe operation of experimental advanced superconducting tokamak (EAST), it is necessary to accurately identify and control the plasma current and its central position. In this article, neural network is used to identify the position of the plasma current center. The model trained by the basic back-propagation neural network can well match the relationship between the electromagnetic diagnostic signals and plasma current center positions. Both noisy simulation data and experimental data are applied to train and test the neural network inference model. Adding 0.1% noise to the training data is proven to improve the noise immunity of the inference model. Basic neural networks trained with both noisy simulation data and actual experimental data show good results with sufficient inputs; however, in both cases, the performance degrades significantly when only the poloidal field coil currents are given as inputs. For this kind of time-series problem, the dynamic neural network containing delay and feedback architecture is introduced, and an improved model requiring much fewer inputs is trained and tested for current center inference. Furthermore, some parameters of this model are compared and analyzed in this article. With suitable neural network architecture, the mapping between the controlled variables (poloidal field coil currents) and response variables (plasma current center) can be well-established.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗