Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “network performance evaluation”

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 1,675 records · Page 93
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

A Graphics Processing Unit (GPU) Approach to Large Eddy Simulation (LES) for Transport and Contaminant Dispersion

Recent advances in the development of large eddy simulation (LES) atmospheric models with corresponding atmospheric transport and dispersion (AT&D) modeling capabilities have made it possible to simulate short, time-averaged, single realizations of pollutant dispersion at the spatial and temporal resolution necessary for common atmospheric dispersion needs, such as designing air sampling networks, assessing pollutant sensor system performance, and characterizing the impact of airborne materials on human health. The high computational burden required to form an ensemble of single-realization dispersion solutions using an LES and coupled AT&D model has, until recently, limited its use to a few proof-of-concept studies. An example of an LES model that can meet the temporal and spatial resolution and computational requirements of these applications is the joint outdoor-indoor urban large eddy simulation (JOULES). A key enabling element within JOULES is the computationally efficient graphics processing unit (GPU)-based LES, which is on the order of 150 times faster than if the LES contaminant dispersion simulations were executed on a central processing unit (CPU) computing platform. JOULES is capable of resolving the turbulence components at a suitable scale for both open terrain and urban landscapes, e.g., owing to varying environmental conditions and a diverse building topology. In this paper, we describe the JOULES modeling system, prior efforts to validate the accuracy of its meteorological simulations, and current results from an evaluation that uses ensembles of dispersion solutions for unstable, neutral, and stable static stability conditions in an open terrain environment.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Improving Progressive Retrieval for HPC Scientific Data using Deep Neural Network

As the disparity between compute and I/O on high-performance computing systems has continued to widen, it has become increasingly difficult to perform post-hoc data analytics on full-resolution scientific simulation data due to the high I/O cost. Error-bounded data decomposition and progressive data retrieval framework has recently been developed to address such a challenge by performing data decomposition before storage and reading only part of the decomposed data when necessary. However, the performance of the progressive retrieval framework has been suffering from the over-pessimistic error control theory, such that the achieved maximum error of recomposed data is significantly lower than the required error. Therefore, more data than required is fetched for recomposition, incurring additional I/O overhead. In order to tackle this issue, we propose a DNN-based progressive retrieval framework that can better identify the minimum amount of data to be retrieved. Our contributions are as follows: 1) We provide an in-depth investigation of the recently developed progressive retrieval framework; 2) We propose two designs of prediction models (named D-MGARD and E-MGARD) to estimate the amount of retrieved data size based on error bounds. 3) We evaluate our proposed solutions using scientific datasets generated by real-world simulations from two domains. Evaluation results demonstrate the effectiveness of our solution in accurately predicting the amount of retrieval data size, as well as the advantages of our solution over the traditional approach to reducing the I/O overhead. Based on our evaluation, our solution is shown to read significantly less data (5% - 40% with D-MGARD, 20% - 80% with E-MGARD).

Wang, Jinzhen↗
DOE OSTI2022

Non-destructive measurement and real-time monitoring of apple hardness during ultrasonic contact drying via portable NIR spectroscopy and machine learning

Portable near-infrared spectrometer in the spectral range of 900–1700 nm was evaluated for the first time to assess and monitor apple hardness in real-time during ultrasonic drying. Calibration models were developed using PLS and ANN, and their performances were evaluated by internal leave-one-out cross-validation and an external dataset. Several pre-treatments including standard normal variate (SNV), multiplicative scatter correction (MSC), Savitzky–Golay first and second derivatives were employed to examine the effects of spectral variations in hardness prediction. Seven important wavelengths were selected using weighted regression coefficients to develop a simple MLR model to facilitate the model interpretation and circumvent noise. The models using PLS, MLR, and ANN with selected wavelengths predicted the apple hardness with R 2 p of 0.91, 0.91, 0.95, and RMSEP of 14.78, 14.85, and 12.46 N, respectively. Finally, the results indicate that portable NIR spectrometers are quite promising for real-time monitoring of apple hardness during ultrasonic drying.

47 OTHER INSTRUMENTATION↗
NASA NTRS2017

Validation of High-Resolution MAIAC Aerosol Product over South America

Multiangle Implementation of Atmospheric Correction (MAIAC) is a new Moderate Resolution Imaging Spectroradiometer (MODIS) algorithm that combines time series approach and image processing to derive surface reflectance and atmosphere products, such as aerosol optical depth (AOD) and columnar water vapor (CWV). The quality assessment of MAIAC AOD at 1 km resolution is still lacking across South America. In the present study, critical assessment of MAIAC AOD(sub 550) was performed using ground-truth data from 19 Aerosol Robotic Network (AERONET) sites over South America. Additionally, we validated the MAIAC CWV retrievals using the same AERONET sites. In general, MAIAC AOD Terra/Aqua retrievals show high agreement with ground-based measurements, with a correlation coefficient (R) close to unity (R(sub Terra):0.956 and R(sub Aqua):0.949). MAIAC accuracy depends on the surface properties and comparisons revealed high confidence retrievals over cropland, forest, savanna, and grassland covers, where more than 2/3 (approximately 66%) of retrievals are within the expected error (EE = +/-(0.05 + 0.05 × AOD)) and R exceeding 0.86. However, AOD retrievals over bright surfaces show lower correlation than those over vegetated areas. Both MAIAC Terra and Aqua retrievals are similarly comparable to AERONET AOD over the MODIS lifetime (small bias offset approximately 0.006). Additionally, MAIAC CWV presents quantitative information with R approximatley 0.97 and more than 70% of retrievals within error (+/-15%). Nonetheless, the time series validation shows an upward bias trend in CWV Terra retrievals and systematic negative bias for CWV Aqua. These results contribute to a comprehensive evaluation of MAIAC AOD retrievals as a new atmospheric product for future aerosol studies over South America.

Martins, V. S.↗
DOE OSTI2021

Deep Learning for Fish Identification from Sonar Data: CRADA 481 [Abstract only]

To help solve the challenges of hydropower energy production related to the potential for eel injury and mortality from passage through hydropower turbines, we will develop a deep learning method for identifying migrating eels from imaging sonar. This project continues with a prior project conducted by the Pacific Northwest National Laboratory (PNNL) and the Electric Power Research Institute (EPRI) in FY2018-2019. The proposed method employs Convolution Neural Network (CNN), a powerful deep learning method for image classification, to distinguish between images of eels and non-eel moving objects. We propose to collect more laboratory data and add more existing field data to train a powerful deep learning model. In addition to eels and sticks as classified in previous studies, we will add images containing several non-eel fish species and macrophyte mats to the training data. A multi-class classification model will be developed to distinguish these objects. Object detection algorithm will be explored and developed to locate and identify multiple objects in each sonar frame. Motion analysis will be performed to track the movement of objects in sonar video clips. We will also improve the data conversion algorithm so that it can read in both DIDSON and ARIS (both are imaging sonars developed by Sound Metrics Corp) data files and convert them to images with comparably high resolution, regardless of the varying detection ranges in different environments. The developed algorithms will be packaged as a software with a graphic user interface. The software will be evaluated by external collaborators in the field. The developed framework can be generalized for automatic monitoring of fish passage and migration using other imaging sonars like ARIS and will benefit the design and operation of ecologically friendly hydroelectric projects. The developed wavelet and CNN model configuration parameters can potentially be transferred to lamprey detection in similar riverine environments.

13 HYDRO ENERGY↗
DOE OSTI2024

Unpaired image translation to mitigate domain shift in liquid argon time projection chamber detector responses

Deep learning algorithms often are developed and trained on a training dataset and deployed on test datasets. Any systematic difference between the training and a test dataset may severely degrade the final algorithm performance on the test dataset—what is known as the domain shift problem . This issue is prevalent in many scientific domains where algorithms are trained on simulated data but applied to real-world datasets. Typically, the domain shift problem is solved through various domain adaptation (DA) methods. However, these methods are often tailored for a specific downstream task, such as classification or semantic segmentation, and may not easily generalize to different tasks. This work explores the feasibility of using an alternative way to solve the domain shift problem that is not specific to any downstream algorithm. The proposed approach relies on modern Unpaired Image-to-Image (UI2I) translation techniques, designed to find translations between different image domains in a fully unsupervised fashion. In this study, the approach is applied to a domain shift problem commonly encountered in Liquid Argon Time Projection Chamber (LArTPC) detector research when seeking a way to translate samples between two differently distributed LArTPC detector datasets deterministically. This translation allows for mapping real-world data into the simulated data domain where the downstream algorithms can be run with much less domain-shift-related performance degradation. Conversely, using the translation from the simulated data to a real-world domain can increase the realism of the simulated dataset and reduce the magnitude of any systematic uncertainties. To evaluate the quality of the translations, we use both pixel-wise metrics and a downstream task to measure the effectiveness of UI2I methods for mitigating the domain shift problem. We adapted several popular UI2I translation algorithms to work on scientific data and demonstrated the viability of these techniques for solving the domain shift problem with LArTPC detector data. To facilitate further development of DA techniques for scientific datasets, the ‘Simple Liquid-Argon Track Samples’ dataset used in this study is also published.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2006

Characterization of Days Based On Analysis of National Airspace System Performance Metrics

Days of operations in the National Airspace System can be described in term of traffic demand, runway conditions, equipment outages, and surface and enroute weather conditions. These causes manifest themselves in terms of departure delays, arrival delays, enroute delays and traffic flow management delays, Traffic flow management initiatives such as, ground stops, ground delay programs, miles-in-trail restrictions, rerouting and airborne holding are imposed to balance the air traffic demand with respect to the available capacity, In order to maintain operational efficiency of the National Airspace System, the Federal Aviation Administration (FAA) maintains delay sad other statistics in the Air Traffic Operations Network (OPSNET) and the Aviation System Performance Metrics (ASPM) databases. OPSNET data includes reportable delays of fifteen minutes ox more experienced by Instrument Flight Rule (IFR) flights. Numbers of aircraft affected by departure delays, enroute delays, arrival delays and traffic flow delays are recorded in the OPSNET data. ASPM data consist of number of actual departures, number of canceled departures, percentage of on time departures, percentage of on time gate arrivals, taxi-out delays. taxi-in delays, gate delays, arrival delays and block delays. Surface conditions at the major U.S. airports are classified in terms of Instrument Meteorological Condition (IMC) and Visual Meteorological Condition (VMC) as a function of the time of the day in the ASPM data. The main objective of this paper is to use OPSNET and ASPM data to classify the days in the datasets into few distinct groups, where each group is separated from the other groups in terms of a distance metric. The motivations for classifying the days are two-fold, 1) to enable selection of days of traffic with particular operational characteristics for concept evaluation using system-wide simulation systems such as the National Aeronautics and Space Administration's Airspace Concepts Evaluation Tool (ACES) and 2) to enable evaluation of a given day with respect to the characteristics of the classified groups. The first part of the paper is devoted to the analysis of major trends seen in the OPSNET and ASPM data. The second part of the paper is devoted to describing features or measures derived from the OPSNET and ASPM data that are suitable for characterizing days, and the classification algorithm used for grouping the days. Finally, the method for evaluating the characteristics of a given day with respect to the properties of the groups is described.

Chatterji, Gano B.↗
DOE OSTI2020

FPDeep: Scalable Acceleration of CNN Training on Deeply-Pipelined FPGA Clusters

In this paper, we propose a framework called FPDeep, which uses a hybrid of model and layer paral- lelism to configure distributed reconfigurable clusters to train DNNs. This approach has numerous benefits. First, the design does not suffer from batch size growth. Second, novel workload and weight partitioning leads to balanced loads of both among nodes. And third, the entire system is fine-grained pipeline. This leads to high parallelism and utilization and also minimizes the time features need to be cached while waiting for back-propagation.

Wang, Tianqi↗
DOE OSTI2023

Deployment of Dynamic Neural Network Optimization to Minimize Heat Rate During Ramping for Coal Power Plants (Final Technical Report)

Much success was achieved throughout the course of this project. A successful implementation of Dynamic Neural Network Optimization (D-NNO) was coupled with Adaptive Predictive Controls (APC) and a novel hardware installation comprised of an advanced sensor network (ASN) measuring mass-weighted averages of flue gas constituents above the horizontal superheater of a coal-fired utility boiler. From 2019 through 2023 (including an extension due to COVID delays), the team was able to prototype, evaluate, deploy, iterate, and ultimately finalize an advanced closed-loop control D-NNO system which demonstrated the ability to: •improve unit efficiency ~2.0% relative to unoptimized operation (represented as total fuel fired per MWh generated) •improve unit NOx emission rates 10%+ beyond static optimization baselines •improve unit temperature stability as much as 58% and on average 12% •improve operating load stability as much as 35% The culmination of this project has generated an advanced methodology of deploying specially designed recurrent neural networks (long short-term memory, gated recurrent unit, encoder-decoder networks, transformers, etc.), customized trajectory planning and closed-loop optimization modules capable of adapting to live electric grid responses and demands, self-tuning and adaptive expert controls constantly adjusting prediction parameters to real-time unit behavior, and a hardware/software package able to reliably calculate net unit heat rate (NUHR) in real-time using flue gas constituents, machine learning, and known combustion relationships. Through this real-time NUHR value, immediate feedback on system adjustments relative to operating efficiency was available, allowing for rapid improvements to system performance. In addition to development and deployment of the advanced D-NNO system, the approach methodology has been readily commercialized through the project platform Griffin Open Systems, LLC, the D-NNO software platform host. Similar methodologies to those developed by this project have already been deployed at 5 other units across the United States, with another 6 implementations scheduled, and more expected. Over the course of the project, multiple academic papers were submitted and accepted for publication within esteemed academic journals, and PhD students were trained and graduated, as well as undergraduate students becoming involved and participating to project objectives.

01 COAL, LIGNITE, AND PEAT↗
DOE OSTI2026

Laboratory and field assessment of mid-infrared absorption (MIRA) instrument performance for methane and ethane dry mole fractions

Concurrent measurements of methane (CH 4 ) and ethane (C 2 H 6 ) can be used to identify and separate methane sources, as ethane is present in thermogenic sources (e.g., oil and natural gas) but not in biogenic sources (e.g., agriculture). In this study, we evaluated the performance of multiple Aeris MIRA Ultra instruments (Versions 1 and 2) through controlled laboratory tests and tower-based deployments under field conditions. The systems were modified with an external pump, flow control, a Nafion dryer, and a custom-built auxiliary box to automate the system and transmit near real-time data. We determined the best calibration approach for our application, given practical limitations, to be a full calibration cycle (with ambient and high calibration cylinders) about once per day and an ambient calibration cylinder sampled hourly. Measurement uncertainty was assessed, including the uncertainty due to instrument noise as a function of calibration frequency, uncertainty in the water vapor correction, and cylinder assignment uncertainty. Instrument noise was the dominant source of uncertainty for C 2 H 6 , while the water vapor correction dominated the CH 4 uncertainty. For Version 2 systems with hourly calibrations and a Nafion dryer with counterflow, the mean total uncertainty, including both systematic errors and noise, of hourly averages was 0.8–3.0 ppb CH 4 and 0.35–0.37 ppb C 2 H 6 . Laboratory intercomparisons showed network compatibility within 1.2 ppb CH 4 and 0.23 ppb C 2 H6, and a collocated deployment with a NOAA Picarro system agreed within 1.8 ppb CH 4 . Instrument noise varied substantially amongst the instruments, with errors reaching up to 11 ppb CH 4 and 2 ppb C 2 H 6 for hourly means, with similar variability indicated in a 50 h cylinder test. With appropriate engineering and calibration, the Aeris MIRA Ultra has the potential to distinguish regional methane emission sources in many field settings.

03 NATURAL GAS↗
DOE OSTI2020

A Step Towards Machine Learning-based Coherent Generator Grouping for Emergency Control Applications in Modern Power Grid

A dynamic neural network (NN) based multi-class classifier is proposed for improving online prediction of coherent generator groups (CGGs), following the occurrences of various contingencies in the power grid. This is motivated by the increasing availability of the measurements from phasor measurement units (PMUs) and the number of grouping schemes is limited. The proposed method consists of three steps. First, by performing offline simulations, a library of system dynamic responses characterized by post-contingency rotor angles and speeds of individual generators is obtained. To generate sufficient data, up to N-2 contingencies and the uncertain parameters associated with the power grid including type and location of disturbance and fault clearing times are modeled. Secondly, the training data-set is produced by generating labels for individual contingencies using a hierarchical clustering method based on rotor angle and speed data. Lastly, the dynamic NN models are trained for online applications such as emergency controls and controlled islanding. The proposed method is tested on the standard 16-generator 68-bus system to demonstrate its performance. Furthermore, the impact of the sample data lengths on the CGG numbers is evaluated. It is interesting to observe that the time domain stability behaviors can be determined by examining the changes in the CGG numbers.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2022

High-fidelity retrieval from instantaneous line-of-sight returns of nacelle-mounted lidar including supervised machine learning

Abstract. Wind turbine applications that leverage nacelle-mounted Doppler lidar are hampered by several sources of uncertainty in the lidar measurement, affecting both bias and random errors. Two problems encountered especially for nacelle-mounted lidar are solid interference due to intersection of the line of sight with solid objects behind, within, or in front of the measurement volume and spectral noise due primarily to limited photon capture. These two uncertainties, especially that due to solid interference, can be reduced with high-fidelity retrieval techniques (i.e., including both quality assurance/quality control and subsequent parameter estimation). Our work compares three such techniques, including conventional thresholding, advanced filtering, and a novel application of supervised machine learning with ensemble neural networks, based on their ability to reduce uncertainty introduced by the two observed nonideal spectral features while keeping data availability high. The approach leverages data from a field experiment involving a continuous-wave (CW) SpinnerLidar from the Technical University of Denmark (DTU) that provided scans of a wide range of flows both unwaked and waked by a field turbine. Independent measurements from an adjacent meteorological tower within the sampling volume permit experimental validation of the instantaneous velocity uncertainty remaining after retrieval that stems from solid interference and strong spectral noise, which is a validation that has not been performed previously. All three methods perform similarly for non-interfered returns, but the advanced filtering and machine learning techniques perform better when solid interference is present, which allows them to produce overall standard deviations of error between 0.2 and 0.3 m s−1, or a 1 %–22 % improvement versus the conventional thresholding technique, over the rotor height for the unwaked cases. Between the two improved techniques, the advanced filtering produces 3.5 % higher overall data availability, while the machine learning offers a faster runtime (i.e., ∼ 1 s to evaluate) that is therefore more commensurate with the requirements of real-time turbine control. The retrieval techniques are described in terms of application to CW lidar, though they are also relevant to pulsed lidar. Previous work by the authors (Brown and Herges, 2020) explored a novel attempt to quantify uncertainty in the output of a high-fidelity lidar retrieval technique using simulated lidar returns; this article provides true uncertainty quantification versus independent measurement and does so for three techniques rather than one.

47 OTHER INSTRUMENTATION↗
DOE OSTI2025

Anomaly Detection in Seismic Data with Deep Learning: Application for Instrument Failure Detection and Forecasting

Seismic data quality assessment (QA) is the first and one of the most important steps before conducting any further data analysis. Traditional methods involve checking various metrics, such as spike detection and power spectral density, by setting strict thresholds or comparing data against synthetic benchmarks. However, these approaches often rely on pre-existing knowledge and assumptions about data anomalies, leading to potential misclassification of unusual cases. Here, in this study, we propose a deep autoencoder model, an unsupervised learning approach that evaluates data quality without making assumptions about normal and anomalous data, which can be used to identify deviations in recorded data that may indicate nascent instrument failure. We test the model with the U.S. International Monitoring System (IMS) seismic stations and demonstrate the capability of detecting anomalies on a monthly scale. This could prompt station operators to examine potential problems early, allowing sufficient time for instrument maintenance to prevent data outages. In addition, we use a new manually selected testing dataset to compare our model performance against two supervised machine learning (ML) approaches and a standard QA package, as baseline models. When applied to the dataset containing known data anomalies, performance of the supervised and unsupervised ML approaches is similar, with an accuracy of 88.1% for our model compared to ∼90% for the supervised ML approach and 78.2% for the standard QA package. Our model outperforms the baseline models when applied to new stations, where new types of data anomalies can be station-specific and not included in the training dataset. Finally, we show model transferability by training the model with data from the Global Seismograph Network only and applying it to the IMS network data. The results suggest that our model is generalizable and can be applied to new stations with good accuracy.

Lin, Jiun-Ting [Lawrence Livermore National Labora↗
DOE OSTI2025

The Sensor Dilemma in Intelligent Transportation Systems

Intelligent Transportation Systems (ITS) are at the forefront in advancing the way we interact and perceive with the transportation network. This revolution is fueled by the significant advancement in sensor perception technologies such as Radar, LiDAR and Video Imaging which are the most popular modalities for ITS. Real-time perception data from these sensors allows intelligent infrastructure side decision making to improve the energy, efficiency and safety at traffic intersections. As traffic departments across the United States are transitioning from traditional loop detectors / emulators and embracing newer technologies, they are often left with a dilemma in choosing a sensor technology for infrastructure-based perception which is reliable, inexpensive, easy to setup and has robust performance in varying weather conditions. However, choosing a sensor which checks all boxes is not straightforward as every sensor type has unique benefits and drawbacks. Radar is excellent at detecting long range vehicles and weather resistance but lacks high resolution. LiDAR is expensive and weather-sensitive, while cameras provide rich visual data at a low cost but are constrained by lighting and visibility. This study examines Radar, LiDAR and camera sensors capabilities to ascertain whether any of these qualifies as the "best" sensor for ITS perception. Through this evaluation, we hope to draw attention to the necessity of National Renewable Energy Laboratory's (NREL) Infrastructure Perception and Control (IPC) framework which presents a multi-sensor track data fusion engine to assimilate multiple data streams in order to provide robust and reliable perception. While no single sensor can meet all the demands of ITS, a hybrid approach combining multiple sensor modalities like Radar, LiDAR and cameras, offers the most robust solution for enhancing the safety and efficiency in intelligent transportation systems.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2024

Catalytic Resonance Theory: Forecasting the Flow of Programmable Catalytic Loops

Chemical transformations on catalyst surfaces occur through series and parallel reaction pathways. These complex networks and their behavior can be most simply evaluated through a three-species surface reaction loop (A* to B* to C* to A*) that is internal to the overall chemical reaction. Application of an oscillating dynamic catalyst to this reactive loop has been shown to exhibit one of three types of behavior: (1) a positive net flux of molecules about the loop in the clockwise direction, (2) a negative net flux of molecules about the loop in the counterclockwise direction, or (3) negligible flux of molecules about the loop at the limit cycle of reaction. Three-species surface loops were simulated with microkinetic modeling to assess the reaction loop behavior resulting from a catalytic surface oscillating between two or more catalyst surface energy states. Selected input parameters for the simulations spanned an 11-dimensional parameter space using 127 688 different parameter combinations. Their converged limit cycle solutions were analyzed for their loop turnover frequencies, the majority of which were found to be approximately zero. Classification and regression machine learning models were trained to predict the sign and magnitude of the loop turnover frequency and successfully performed above accessible baselines. Notably, the classification models exhibited a baseline weighted F1 score of 0.49, whereas trained models achieved weighted F1 scores of 0.94 and 0.96 when trained on the parameters used to define the simulations and derived rate constants, respectively. The trained models successfully predicted catalytic loop behavior, and interpretation of these models revealed all input parameters to be important for the prediction and performance of each model.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

Algorithm-Based Fault Tolerance for Convolutional Neural Networks

Convolutional neural networks (CNNs) are becoming more and more important for solving challenging and critical problems in many fields. CNN inference applications have been deployed in safety-critical systems, which may suffer from soft errors caused by high-energy particles, high temperature, or abnormal voltage. Of critical importance is ensuring the stability of the CNN inference process against soft errors. Traditional fault tolerance methods are not suitable for CNN inference because error-correcting code is unable to protect computational components, instruction duplication techniques incur high overhead, and existing algorithm-based fault tolerance (ABFT) techniques cannot protect all convolution implementations. In this paper, we focus on how to protect the CNN inference process against soft errors as efficiently as possible, with the following three contributions. (1) We propose several systematic ABFT schemes based on checksum techniques and analyze their fault protection ability and runtime thoroughly. Unlike traditional ABFT based on matrix-matrix multiplication, our schemes support any convolution implementations. (2) We design a novel workflow integrating all the proposed schemes to obtain a high detection/correction ability with limited total runtime overhead. (3) We perform our evaluation using ImageNet with well-known CNN models including AlexNet, VGG-19, ResNet-18, and YOLOv2. Here, experimental results demonstrate that our implementation can handle soft errors with very limited runtime overhead (4%~8% in both error-free and error-injected situations).

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Evaluating proxies for the drivers of natural gas productivity using machine-learning models

We report the extensive development of unconventional reservoirs using horizontal drilling and multistage hydraulic fracturing has generated large volumes of reservoir characterization and production data. The analysis of this abundant data using statistical methods and advanced machine-learning (ML) techniques can provide data-driven insights into well performance. Most predictive modeling studies have focused on the impact that different well completion and stimulation strategies have on well production but have not fully exploited the available in situ rock property data to determine its role in reservoir productivity. We have used machine-learning techniques to rank rock mechanical properties, microseismic attributes, and stimulation parameters in the order of their significance for predicting natural gas production from an unconventional reservoir. The data for this study came from a hydraulically fractured well in the Marcellus Shale in Monongalia County, West Virginia. The data classes included measurements aggregated by well completion stage that included (1) gas production, (2) well-log-derived measurements including bulk density, elastic moduli, shear impedance, compressional impedance, brittleness, and gamma measurements, (3) microseismic attributes, (4) long-period long-duration (LPLD) event counts, (5) fracture counts, and (6) stimulation parameters that included the fluid injection volume and average pumping pressure. To identify observable proxies for the drivers of gas production, we evaluated five commonly used ML approaches including multivariate adaptive regression spline, Gaussian mixture model, random forest, gradient boosting, and neural network. We selected five variables including LPLD event count, seismogenic b-value, hydraulic diffusivity, cumulative moment, and fluid volume as the features most likely to impact gas productivity at the stage level in the study area. The data-driven selection of these parameters for their importance in determining gas production can help reservoir engineers design more effective hydraulic-fracture treatments in the Marcellus Shale and other similar unconventional reservoirs. Plain language summary: We use machine-learning methods and data-driven selection of reservoir parameters to rank and better understand their importance in determining gas production, which can help reservoir engineers design more effective hydraulic-fracture treatments in the Marcellus Shale and other similar unconventional reservoirs.

58 GEOSCIENCES↗
DOE OSTI2023

Neural Image Compression: Generalization, Robustness, and Spectral Biases

Recent advances in neural image compression (NIC) have resulted in models which are starting to outperform traditional codecs. While this has led to growing excitement about using these methods in real-world applications, the successful adoption of any machine learning system (including NIC) in the wild requires it to generalize (and be robust) to unseen distribution shifts at deployment time. Unfortunately, current research lacks comprehensive datasets and informative tools to evaluate and understand compression performance in real-world settings. To bridge this crucial gap, first, this paper presents a comprehensive benchmark suite to evaluate the out-of-distribution (OOD) performance of image compression methods. Specifically, we design CLIC-C and Kodak-C by introducing 15 common corruptions to popular CLIC and Kodak benchmarks. Next, we propose spectrally inspired introspection tools to gain a deeper understanding of errors introduced by image compression methods as well as their OOD performance. To this end, we carry out a detailed performance comparison of the classical codec with various variants of NIC (e.g., original, variable rate, pruned), revealing intriguing findings that challenge our current understanding of the strengths and limitations of NIC. Finally, we corroborate our empirical findings with theoretical analysis, providing an in-depth view of the OOD performance of NIC. Our benchmarks, spectral introspection tools, and findings provide a crucial bridge to the real-world adoption of NIC. We hope that our work will propel future efforts in designing more robust and generalizable NIC methods.

neural networks, Variational Autoencoder, robustne↗