Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Preprocessing Systems”

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 379 records · Page 21
All sourcesNASA NTRSDOE OSTI
NASA NTRS1996

First Two Years of Observations NASA ACTS Propagation Experiment Central Oklahoma Site

Continuous observations from December 1, 1993 through November 30, 1995 were made at the ACTS Propagation Terminal on the roof of the Sarkeys Energy Center at the University of Oklahoma in Norman, Oklahoma. Beacon and radiometer observations were combined to calibrate the beacon system for the estimation of total attenuation (attenuation relative to free space) and attenuation relative to clear sky (gaseous absorption component removed). Empirical cumulative distributions (edf's) were compiled for each month of observation and for each year. The annual edf's are displayed in the figures, the monthly and annual edf's are listed in the tables. The tables are organized by blocks and pages within a block. The blocks correspond to the headings in the edf files generated by the ACTS Preprocessing (actspp) software and contained in the fourth disk in the set of ACTS Propagation Experiment CD-ROMs generated by the University of Texas.

Crane, Robert K.↗
NASA NTRS1974

Development of a Korotkov sound processor for automatic identification of auscultatory events. I - Specification of preprocessing bandpass filters

Frequency bands that best discriminate the Korotkov sounds at systole and at diastole from the sounds immediately preceding these events are defined. Korotkov sound data were recorded from five normotensive subjects during orthostatic stress (lower body negative pressure) and bicycle ergometry. A spectral analysis of the seven Korotkov sounds centered about the systolic and diastolic auscultatory events revealed that a maximum increase in amplitude at the systolic transition occurred in the 18-26-Hz band, while a maximum decrease in amplitude at the diastolic transition occurred in the 40-60-Hz band. These findings were remarkably consistent across subjects and test conditions. These passbands are included in the design specifications for an automatic blood pressure measuring system used in conjuction with medical experiments during NASA's Skylab program.

Golden, D. P., Jr.↗
DOE OSTI2021

Phasor-Measurement-Unit-Based Data Analytics Using Digital Twin and PhasorAnalytics Software

A major objective of this project was to apply GE’s commercial machine learning and data analytics toolsets to large-scale, real-world, anonymized Phasor Measurement Unit (PMU) datasets in order to extract signatures, correlated and/or causal factors, and precursor patterns associated with significant power system phenomena. The project had a particular emphasis on extraction of insights relevant to asset health monitoring, real-time load modeling and cybersecurity monitoring. Additionally, the team was directed to undertake a comprehensive data quality analysis for the provided datasets and encouraged to estimate the ‘machine-learning readiness’ of the datasets by documenting any major obstacles to the application of commercial machine learning algorithms. To accomplish the aforementioned objectives, the project team’s work centered around the identification of key event signatures and application of the identified event signatures for event detection and event classification. The industry-validated, semi-supervised machine learning strategy employed for event signature identification involved several major tasks, including data-preprocessing, generation of an overabundance of features, normal data identification, normality modeling, and event signature identification through a methodical, quantitative ranking of features in order of relevance to each studied event type. Throughout the project, data quality issues and mitigation techniques were investigated. In this report, insights are provided regarding the readiness of the provided synchrophasor datasets for application of machine learning and data analytics. The methodologies employed for this technical strategy are summarized in this report. With regards to data preprocessing and feature generation, the provided Training and Test Datasets were ingested into GE’s big data environment. Subsequently, the team applied bad data cleansing and data imputation scripts, event detection scripts, and application programming interfaces (APIs) to the datasets for convenient data access. The project team completed development and validation of dozens of physics-based, statistics-based and transformation-based feature functions used for the extraction of over 60 synchrophasor features. Using a new parallel feature generation technology developed on this project, over 60 features have been rapidly generated for the full two years’ worth of Training and Test Dataset data associated with both the Eastern and Western interconnects. Even accommodating for temporal down-sampling inherent to the feature extraction procedure, this parallel feature generation activity resulted in a massive feature set with a storage requirement approximately equal to that of the raw training dataset itself. With regards to normal data identification and normality modeling, a normality model was built using the feature data extracted from the Training Dataset and iteratively refined subsequent to incremental adjustments and expansions of the Training Dataset feature data. With respect to event characterization and signature identification, an event signature identification pipeline was developed and used in conjunction with the normality model to identify over 15 event signatures for key event categories within the Training Dataset. The identified event signatures were used to characterize hundreds of key events in terms of relative severity, duration, and location of the event. An investigation was undertaken to identify correlated and causal factors involved in transformer events. A separate investigation into temporal trends in ring-down analysis results was undertaken to determine possible associations between system dynamics and various other factors such as loading, season or year. To validate the identified event signatures, additional work was undertaken to develop signature-based anomaly detection and classification tools suitable for convenient application to the synchrophasor datasets. The anomaly detection and classification tools, suitable for online application, were then applied to the entirety of the Eastern Interconnect Training and Test Datasets. Performance of the event detection and classification tools was evaluated upon receipt of the Test Dataset event logs (i.e., the labels for events contained in the Test Dataset), and promising results were obtained despite several challenges (documented herein) associated with application of supervised or semi-supervised machine learning methods to large-scale, anonymized datasets. Finally, the detection and classification tools were used to detect, classify, and characterize thousands of new events not included in the original event logs provided by the DOE within both the Training and Test Datasets.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2021

Brady Geodatabase for Geothermal Exploration Artificial Intelligence

These files contain the geodatabases related to Brady's Geothermal Field. It includes all input and output files for the Geothermal Exploration Artificial Intelligence. Input and output files are sorted into three categories: raw data, pre-processed data, and analysis (post-processed data). In each of these categories there are six additional types of raster catalogs which are titled Radar, SWIR, Thermal, Geophysics, Geology, and Wells. These inputs and outputs were used with the Geothermal Exploration Artificial Intelligence to identify indicators of blind geothermal systems at the Brady Hot Springs Geothermal Site. The included zip file is a geodatabase to be used with ArcGIS and the tar file is an inclusive database that encompasses the inputs and outputs for the Brady Hot Springs Geothermal Site.

15 GEOTHERMAL ENERGY↗
NASA NTRS2011

Visual SLAM Using Variance Grid Maps

An algorithm denoted Gamma-SLAM performs further processing, in real time, of preprocessed digitized images acquired by a stereoscopic pair of electronic cameras aboard an off-road robotic ground vehicle to build accurate maps of the terrain and determine the location of the vehicle with respect to the maps. Part of the name of the algorithm reflects the fact that the process of building the maps and determining the location with respect to them is denoted simultaneous localization and mapping (SLAM). Most prior real-time SLAM algorithms have been limited in applicability to (1) systems equipped with scanning laser range finders as the primary sensors in (2) indoor environments (or relatively simply structured outdoor environments). The few prior vision-based SLAM algorithms have been feature-based and not suitable for real-time applications and, hence, not suitable for autonomous navigation on irregularly structured terrain. The Gamma-SLAM algorithm incorporates two key innovations: Visual odometry (in contradistinction to wheel odometry) is used to estimate the motion of the vehicle. An elevation variance map (in contradistinction to an occupancy or an elevation map) is used to represent the terrain. The Gamma-SLAM algorithm makes use of a Rao-Blackwellized particle filter (RBPF) from Bayesian estimation theory for maintaining a distribution over poses and maps. The core idea of the RBPF approach is that the SLAM problem can be factored into two parts: (1) finding the distribution over robot trajectories, and (2) finding the map conditioned on any given trajectory. The factorization involves the use of a particle filter in which each particle encodes both a possible trajectory and a map conditioned on that trajectory. The base estimate of the trajectory is derived from visual odometry, and the map conditioned on that trajectory is a Cartesian grid of elevation variances. In comparison with traditional occupancy or elevation grid maps, the grid elevation variance maps are much better for representing the structure of vegetated or rocky terrain.

Howard, Andrew B.↗
NASA NTRS1990

Parallel asynchronous hardware implementation of image processing algorithms

Research is being carried out on hardware for a new approach to focal plane processing. The hardware involves silicon injection mode devices. These devices provide a natural basis for parallel asynchronous focal plane image preprocessing. The simplicity and novel properties of the devices would permit an independent analog processing channel to be dedicated to every pixel. A laminar architecture built from arrays of the devices would form a two-dimensional (2-D) array processor with a 2-D array of inputs located directly behind a focal plane detector array. A 2-D image data stream would propagate in neuron-like asynchronous pulse-coded form through the laminar processor. No multiplexing, digitization, or serial processing would occur in the preprocessing state. High performance is expected, based on pulse coding of input currents down to one picoampere with noise referred to input of about 10 femtoamperes. Linear pulse coding has been observed for input currents ranging up to seven orders of magnitude. Low power requirements suggest utility in space and in conjunction with very large arrays. Very low dark current and multispectral capability are possible because of hardware compatibility with the cryogenic environment of high performance detector arrays. The aforementioned hardware development effort is aimed at systems which would integrate image acquisition and image processing.

Coon, Darryl D.↗
NASA NTRS2019

Analysis Ready Data in Analytics Optimized Data Stores for Analysis of Big Earth Data in the Cloud

Cloud computing offers the possibility of making the analysis of Big Data approachable for a wider community due to affordable access to computing power, an ecosystem of usable tools for parallel processing, and migration of many large datasets to archives in the cloud, allowing data-proximal computing. Generally, data analysis acceleration in the cloud comes from running multiple nodes in a split-combine-apply strategy. Data systems such as the Earth Observing System Data and Information System are in a position to "pre-split" the data by storing them in a data store that is optimized for data parallel computing, i.e., an Analytics-Optimized Data Store (AODS). A variety of approaches to AODS are possible, from highly scalable databases to scalable filesystems to data formats optimized for cloud access (e.g., zarr and cloud-optimized datasets), with the optimal choice dependent on both the types of analysis and the geospatial structure of the data. A key question is how much preprocessing of the data to do, both before splitting and as the first part of the apply step. Again, the geospatial structure of the data and the analysis type influence the decision, with the added complexity of the user type. Trans-disciplinary users who are not well-versed in the nuances of quality-filtering and georeferencing of remote sensing orbit/swath/scene data tend to ask for more highly processed data, relying on the data provider to make sensible decisions on preprocessing parameters. (This accounts for the popularity of "Level 3" gridded data, despite the lower spatial resolution it provides.) In this case, data can be preprocessed before the split, resulting in higher performance in the rest of the "apply" step, which can be transformative for use cases such as interactive data exploration at scale. Discipline researchers who are experienced with remote sensing data often prefer more flexibility in customizing the preprocessing data into Analysis Ready Data, resulting in more need for on-the-fly preprocessing.

Lynnes, Christopher↗
DOE OSTI2022

Statistical characterization of experimental magnetized liner inertial fusion stagnation images using deep-learning-based fuel–background segmentation

Significant variety is observed in spherical crystal x-ray imager (SCXI) data for the stagnated fuel–liner system created in Magnetized Liner Inertial Fusion (MagLIF) experiments conducted at the Sandia National Laboratories Z-facility. As a result, image analysis tasks involving, e.g., region-of-interest selection (i.e. segmentation), background subtraction and image registration have generally required tedious manual treatment leading to increased risk of irreproducibility, lack of uncertainty quantification and smaller-scale studies using only a fraction of available data. We present a convolutional neural network (CNN)-based pipeline to automate much of the image processing workflow. This tool enabled batch preprocessing of an ensemble of N scans = 139 SCXI images across N exp = 67 different experiments for subsequent study. The pipeline begins by segmenting images into the stagnated fuel and background using a CNN trained on synthetic images generated from a geometric model of a physical three-dimensional plasma. The resulting segmentation allows for a rules-based registration. Our approach flexibly handles rarely occurring artifacts through minimal user input and avoids the need for extensive hand labelling and augmentation of our experimental dataset that would be needed to train an end-to-end pipeline. Here we also fit background pixels using low-degree polynomials, and perform a statistical assessment of the background and noise properties over the entire image database. Our results provide a guide for choices made in statistical inference models using stagnation image data and can be applied in the generation of synthetic datasets with realistic choices of noise statistics and background models used for machine learning tasks in MagLIF data analysis. We anticipate that the method may be readily extended to automate other MagLIF stagnation imaging applications.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
NASA NTRS2005

An Automated Flying-Insect-Detection System

An automated flying-insect-detection system (AFIDS) was developed as a proof-of-concept instrument for real-time detection and identification of flying insects. This type of system has use in public health and homeland security decision support, agriculture and military pest management, and/or entomological research. Insects are first lured into the AFIDS integrated sphere by insect attractants. Once inside the sphere, the insect's wing beats cause alterations in light intensity that is detected by a photoelectric sensor. Following detection, the insects are encouraged (with the use of a small fan) to move out of the sphere and into a designated insect trap where they are held for taxonomic identification or serological testing. The acquired electronic wing beat signatures are preprocessed (Fourier transformed) in real-time to display a periodic signal. These signals are sent to the end user where they are graphically displayed. All AFIDS data are pre-processed in the field with the use of a laptop computer equipped with LABVIEW. The AFIDS software can be programmed to run continuously or at specific time intervals when insects are prevalent. A special DC-restored transimpedance amplifier reduces the contributions of low-frequency background light signals, and affords approximately two orders of magnitude greater AC gain than conventional amplifiers. This greatly increases the signal-to-noise ratio and enables the detection of small changes in light intensity. The AFIDS light source consists of high-intensity Al GaInP light-emitting diodes (LEDs). The AFIDS circuitry minimizes brightness fluctuations in the LEDs and when integrated with an integrating sphere, creates a diffuse uniform light field. The insect wing beats isotropically scatter the diffuse light in the sphere and create wing beat signatures that are detected by the sensor. This configuration minimizes variations in signal associated with insect flight orientation.

Vann, Timi↗
NASA NTRS1976

Digital preprocessing and classification of multispectral earth observation data

The development of airborne and satellite multispectral image scanning sensors has generated wide-spread interest in application of these sensors to earth resource mapping. These point scanning sensors permit scenes to be imaged in a large number of electromagnetic energy bands between .3 and 15 micrometers. The energy sensed in each band can be used as a feature in a computer based multi-dimensional pattern recognition process to aid in interpreting the nature of elements in the scene. Images from each band can also be interpreted visually. Visual interpretation of five or ten multispectral images simultaneously becomes impractical especially as area studied increases; hence, great emphasis has been placed on machine (computer) techniques for aiding in the interpretation process. This paper describes a computer software system concept called LARSYS for analysis of multivariate image data and presents some examples of its application.

Anuta, P. E.↗
NASA NTRS1984

Timing performance of phased-locked loops in optical pulse position modulation communication systems

An optical digital communication system requires that an accurate clock signal be available at the receiver for proper synchronization with the transmitted signal. Phase synchronization is especially critical in M-ary pulse position modulation (PPM) systems where the optimum decision scheme is an energy detector which compares the energy in each of M time slots to decide which of M possible words was sent. Timing errors cause energy spillover into adjacent time slots (a form of intersymbol interference) so that only a portion of the signal energy may be attributed to the correct time slot. This effect decreases the effective signal, increases the effective noise, and increases the probability of error. A timing subsystem for a satellite-to-satellite optical PPM communication link is simulated. The receiver employs direct photodetection, preprocessing of the detected signal, and a phase-locked loop for timing synchronization. The variance of the relative phase error is examined under varying signal strength conditions as an indication of loop performance, and simulation results are compared to theoretical calculations.

Lafaw, D. A.↗
NASA NTRS2008

Overview of the LaNCETS Flight Experiment and CFD Analysis. Supplemental Movies

This presentation focuses on nearfield airborne pressure signatures from the Lift and Nozzle Change Effect on Tail Shocks (LaNCETS) flight test experiment. The primary motivation for nearfield probing in the supersonic regime is to measure the shock structure of aircraft in an ongoing effort to overcome the overland sonic boom barrier for commercial supersonic transportation. LaNCETS provides the opportunity to investigate lift distribution and engine plume effects. During Phase 1 flight testing an F-15B was used to probe the F-15 LaNCETS aircraft in order to validate CFD and pre-flight prediction tools. A total of 29 probings were taken at 40,000 ft. altitude at Machs 1.2, 1.4 and 1.6. LaNCETS Phase 1 flight data are presented as a detailed pressure signature superimposed over a picture of the LaNCETS aircraft. The attenuation of the Inlet-Canard shocks with distance as well as its forward propagation and the coalescence of the noseboom shock are illustrated. A detailed CFD study on a simplified LaNCETS aircraft jet nozzle was performed providing the ability to more accurately capture the shocks from the propulsion system and emphasizing how under- and over-expanding the nozzle affects the existence of shock trains inside the jet plume. With Phase 1 being a success preparations are being made to move forward to Phase 2. Phase 2 will fly similar flight conditions, but this time changing the aircraft's lift distribution by biasing the canard positions, and changing the plume shape by under- and over-expanding the nozzle. Nearfield probing will again be completed in the same manner as in Phase 1. An additional presentation focuses on LaNCETS CFD solution methodology. Discussions highlight grid preprocessing, grid shearing and stretching, flow solving and contour plots. Efforts are underway to better capture the flow features via grid modification and flow solution methodology, which will help to achieve better agreement with flight data. An included CD-ROM provides animations of the nearfield probing procedure and of real data from one of the probings integrated with GPS positional and velocity data. An additional in-flight video from the rear seat of the probing aircraft is also provided.

Cliatt, Larry J., II↗
NASA NTRS2013

Mars Science Laboratory Entry, Descent, and Landing Trajectory and Atmosphere Reconstruction

On August 5th 2012, The Mars Science Laboratory entry vehicle successfully entered Mars atmosphere and landed the Curiosity rover on its surface. A Kalman filter approach has been implemented to reconstruct the entry, descent, and landing trajectory based on all available data. The data sources considered in the Kalman filtering approach include the inertial measurement unit accelerations and angular rates, the terrain descent sensor, the measured landing site, orbit determination solutions for the initial conditions, and a new set of instrumentation for planetary entry reconstruction consisting of forebody pressure sensors, known as the Mars Entry Atmospheric Data System. These pressure measurements are unique for planetary entry, descent, and landing reconstruction as they enable a reconstruction of the freestream atmospheric conditions without any prior assumptions being made on the vehicle aerodynamics. Moreover, the processing of these pressure measurements in the Kalman filter approach enables the identification of atmospheric winds, which has not been accomplished in past planetary entry reconstructions. This separation of atmosphere and aerodynamics allows for aerodynamic model reconciliation and uncertainty quantification, which directly impacts future missions. This paper describes the mathematical formulation of the Kalman filtering approach, a summary of data sources and preprocessing activities, and results of the reconstruction.

Karlgaard, Christopher D.↗
DOE OSTI2020

Jefferson Laboratory C100 Superconducting Radio-Frequency Cavity Fault Data, 2020

The dataset was created to train machine learning models for the task of identifying the (1) cavity and (2) fault type from C100-type cryomodules at the Thomas Jefferson National Accelerator Facility (Jefferson Lab), thereby replacing the time-consuming efforts of a subject matter expert. Superconducting radio-frequency (SRF) cavity trips represent a significant source of accelerator downtime. Real-time – rather than post-mortem – identification of the offending cavity and classification of the fault type would give control room operators valuable feedback for corrective action planning. The anticipated benefit is increased beam-on-target time for users and provides performance metrics that can be used to improve future cavity designs. A series of 17 RF signals are recorded for each of the 8 cavities in a C100 cryomodule every time a cavity trips. These time-series signals are written to file using a specially designed data acquisition system. The dataset represents fault events recorded during Continuous Electron Beam Accelerator Facility (CEBAF) beam operations between January 18, 2019 and March 9, 2020. The following filtering steps were applied to collected data; (1) only 4 of the 17 signals per cavity are retained (GMES, GASK, CRFP, DETA2) (2) only events with data from each of the eight cavities in the cryomodule are kept, (3) only events that were sampled at 5 kHz were kept, (4) events from cryomodule 0L04 were neglected, (5) events occurring between February 4, 3PM and February 5, 12PM were neglected. As a result of preprocessing, the dataset is comprised of 2,375 unique events. The full dataset is comprised of three files: features.csv, cavity_labels.csv, fault_labels.csv. Each instance in faults.csv includes a timestamp (“date_time”), a label for the cryomodule which experienced the trip (“zone_label”), and 192 features (“feature_1”, “feature_2”... “feature_192”). The features correspond to 6 autoregressive features for each of 4 signals per cavity for each of the 8 cavities (6 × 4 signals/cavity × 8 cavities/cryomodule = 192). To deal with the large variation of signal amplitudes, time-series standardization via the z-score (standard score) function was applied prior to computing the features. For each instance, there is an associated label for the (1) cavity which faulted first (cavity_labels.csv) and (2) the type of fault that caused the trip (fault_labels.csv). The cavity identification can take values of [0, 1, 2, 3, 4, 5, 6, 7, 8] and the fault type can take values of [‘Microphonics’, ‘Quench_100ms’, ‘Controls_Fault’, ‘E_Quench’, ‘Quench_3ms’, ‘Single_Cav_Turn_Off’ , ‘Heat_Riser_Choke’, ‘Multi_Cav_Turn_Off’].

43 PARTICLE ACCELERATORS↗
DOE OSTI2022

Robust Event Classification Using Imperfect Real-world PMU Data

Here, this paper studies robust event classification using imperfect real-world phasor measurement unit (PMU) data. By analyzing the real-world PMU data, we find it is challenging to directly use this dataset for event classifiers due to the low data quality observed in PMU measurements and event logs. To address these challenges, we develop a novel machine learning framework for training robust event classifiers, which consists of three main steps: data preprocessing, fine-grained event data extraction, and feature engineering. Specifically, the data preprocessing step addresses the data quality issues of PMU measurements (e.g., bad data and missing data); in the fine-grained event data extraction step, a model-free event detection method is developed to accurately localize the events from the inaccurate event timestamps in the event logs; and the feature engineering step constructs the event features based on the patterns of different event types, in order to improve the performance and the interpretability of the event classifiers. Based on the proposed framework, we develop a workflow for event classification using the real-world PMU data streaming into the system in real time. Using the proposed framework, robust event classifiers can be efficiently trained based on many off-the-shelf lightweight machine learning models. Numerical experiments using the real-world dataset from the Western Interconnection of the U.S power transmission grid show that the event classifiers trained under the proposed framework can achieve high classification accuracy while being robust against low-quality data.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗
NASA NTRS1984

The MST Radar Technique

The coherent radar technique is reviewed with special emphasis to mesosphere-stratosphere-troposphere (MST) radars operating in the VHF band. Some basic introduction to Doppler radar measurements and the radar equation is followed by an outline of the characteristics of atmospheric turbulence, viewed from the scattering and reflection processes of radar signals. Radar signal acquisition and preprocessing, namely coherent detection, digital sampling, pre-integration and coding, is briefly discussed. The data analysis is represented in terms of the correlation and spectrum analysis, yielding the essential parameters: power, signal-to-noise ratio, average and fluctuating velocity and persistency. The techniques to measure wind velocities, viz. the different modes of the Doppler method as well as the space antenna method are surveyed and the feasibilities of the MST radar interferometer technique are elucidated. A general view on the criteria to design phased array antennas is given. An outline of the hardware of a typical MST radar system is presented.

Roettger, J.↗
DOE OSTI2024

Real-time confinement regime detection in fusion plasmas with convolutional neural networks and high-bandwidth edge fluctuation measurements

Abstract A real-time detection of the plasma confinement regime can enable new advanced plasma control capabilities for both the access to and sustainment of enhanced confinement regimes in fusion devices. For example, a real-time indication of the confinement regime can facilitate transition to the high-performing wide-pedestal (WP) quiescent H-mode, or avoid unwanted transitions to lower confinement regimes that may induce plasma termination. To demonstrate real-time confinement regime detection, we use the 2D beam emission spectroscopy (BES) diagnostic system to capture localized density fluctuations of long wavelength turbulent modes in the edge region at a 1 MHz sampling rate. BES data from 330 discharges in either L-mode, H-mode, quiescent H (QH)-mode, or WP QH-mode were collected from the DIII-D tokamak and curated to develop a high-quality database to train a deep-learning classification model for real-time confinement detection. We utilize the 6×8 spatial configuration with a time window of 1024 µ s and recast the input to obtain spectral-like features via fast Fourier transform preprocessing. We employ a shallow 3D convolutional neural network for the multivariate time-series classification task and utilize a softmax in the final dense layer to retrieve a probability distribution over the different confinement regimes. Our model classifies the global confinement state on 44 unseen test discharges with an average F 1 score of 0.94, using only ∼1 ms snippets of BES data at a time. This activity demonstrates the feasibility for real-time data analysis of fluctuation diagnostics in future devices such as ITER, where the need for reliable and advanced plasma control is urgent.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2020

Adapting In Situ Accelerators for Sparsity With Granular Matrix Reordering

Neural network (NN) inference is an essential part of modern systems and is found at the heart of numerous applications ranging from image recognition to natural language processing. In situ NN accelerators can efficiently perform NN inference using resistive crossbars, which makes them a promising solution to the data movement challenges faced by conventional architectures. Although such accelerators demonstrate significant potential for dense NNs, they often do not benefit from sparse NNs, which contain relatively few non-zero weights. Processing sparse NNs on in situ accelerators results in wasted energy to charge the entire crossbar where most elements are zeros. To address this limitation, this paper proposes Granular Matrix Reordering (GMR): a preprocessing technique that enables an energy-efficient computation of sparse NNs on in situ accelerators. GMR reorders the rows and columns of sparse weight matrices to maximize the crossbars' utilization and minimize the total number of crossbars needed to be charged. The reordering process does not rely on sparsity patterns and incurs no accuracy loss. Finally, GMR achieves an average of 28% and up to 34% reduction in energy consumption over seven pruned NNs across four different pruning methods and network architectures.

97 MATHEMATICS AND COMPUTING↗