Search NASA⌕ Search

SEARCH · Search NASA

Results for “Automatic Labeling”

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 163 records · Page 9

Lighting System Control Data to Improve Design and Operation: Tunable Lighting System Data from NICU Patient Rooms

The advancement of LED and controls technology, computing capacity, and software provides new opportunities for researchers and designers to work together to further optimize spaces for occupant benefit. Here, lighting system control data from five neonatal intensive care unit patient rooms was collected over a 25-week monitoring period and analyzed to better understand occupant response to a tunable lighting system with automatic transitions throughout the day. Lighting systems are very rarely refined after installation based on actual use. Objective data detailing how the lighting system is used by the actual occupants highlights the opportunities for optimization after installation and provides insight for improving the next design. As use of the data becomes more commonplace, it can be leveraged for design recommendations. The collection of the data required no additional cost beyond the time for examining the data. The analysis revealed several clear opportunities for improvement, including adjustments to the default control setting at night, re-labeling of the control stations, and adjustments to the nighttime fade rate. The patient room occupants were active users of the different zones, dimming options, and manual overrides made available by the lighting system.

60 APPLIED LIFE SCIENCES↗

AGR 5/6/7 Data Qualification Report for ATR Cycles 162B through 168A

This report provides the qualification status of experimental data for the Advanced Gas Reactor (AGR) 5/6/7 fuel irradiation. AGR-5/6/7 was conducted in the Advanced Test Reactor (ATR) at Idaho National Laboratory (INL) in support of development and qualification of tri-structural isotropic (TRISO) low-enriched fuel for use in high temperature gas-cooled reactors. The objectives of the AGR-5/6/7 experiments are to: (i) irradiate reference-design fuel particles to support fuel qualification, (ii) establish operating margins for the fuel beyond normal operating conditions, and (iii) provide irradiated-fuel performance data and irradiated-fuel samples for post-irradiation examination (PIE) and safety testing. The test train contains five separate capsules that were independently controlled and monitored. Each capsule contains multiple 12.51-mm-long compacts filled with low enriched uranium carbide/oxide (UCO) TRISO fuel particles. The primary objective of the AGR-5/6 test (Capsules 1, 2, 4, and 5) is to verify successful performance of the reference-design fuel under normal operating conditions. The AGR-7 test (Capsule 3) was designed to explore fuel performance at higher temperatures to demonstrate the capability of the fuel to withstand conditions beyond normal operating conditions in support of plant design and licensing. AGR 5/6/7 will also provide irradiated-fuel performance data on fission-gas release from failed particles during irradiation. The AGR-5/6/7 capsules were irradiated in the ATR northeast flux trap location. The experiment began on February 16, 2018 and ended on July 22, 2020, spanning nine ATR cycles over two and a half years. Thus, the AGR-5/6/7 fuel compacts were irradiated for a total of 360.9 effective full power days. The AGR 5/6/7 experiment was able to remain in the reactor core during all three Powered Axial Locator Mechanism (PALM) cycles (163A, 165A, and 167A) without overheating its fuel compacts. This report includes irradiation monitoring data from nine ATR Cycles: 162B, 163A, 164A, 164B, 165A, 166A, 166B, 167A, and 168A, as stored in the Nuclear Data Management and Analysis System (NDMAS). During irradiation, data records consisted of instantaneous measurements recorded every minute and provided by text files automatically every 2 hours. The AGR 5/6/7 data streams addressed in this report include thermocouple (TC) temperatures, sweep gas data (flow rates [capsule inlet, outlet, and downstream at detector], pressure, and moisture content), and Fission Product Monitoring System (FPMS) data (release rates and release to birth rate ratios [R/Bs]) for each of the five capsules. A total of 94,989,908 TC temperature and sweep gas data records were received and processed by NDMAS for AGR 5/6/7 irradiation. Of these records, 41,593,387 (or 43.7% of the total) met data collection and accuracy requirements and are labeled as Qualified. A total of 57,746,693 TC temperature readings were captured from 54 installed TCs. Among them, 10,034,676 TC temperature records (only 17.4%) were Qualified and 47,701,371 TC temperatures (or 82.6%) are Failed due to 48 TC failures (63.5%) and due to missing values (19.1%). To assess performance of the operational TCs, analysis of daily correlations between TCs found no evidence of virtual junction failure for any TCs. Analyses on control charts of TC temperature differences revealed trending in TC readings for TC2, 4, 5, and 13 in Capsule 3, but there is no conclusive indication of TC drift failure that caused those trends. Therefore, TC control charts are not used to disqualify TC data, but only for users’ consideration. For sweep gas flow rates, a total of 31,519,747 gas flow records (84.4%) are Qualified for use for AGR-5/6/7 experiment; 5,723,468 gas flow records (15.4%) are Failed due mostly to missing values; and 74,641 high sweep gas flow rates (0.2 %) are Trend. A large number of Failed missing TC temperature and gas flow values were caused by an error in the data output script that outputted a ‘NULL’ value when values were unchanged. This problem was fixed during the outage of Cycle 166B, which led to a substantially decreased number of missing values during the last three cycles. Nonetheless, a large amount of non-missing data remained because of the high data acquisition frequency (1-minute) and still provided sufficient data to effectively monitor the experiment as designed. For FPMS data, NDMAS received and processed fission product release and R/B data for nine ATR cycles, when ATR core reached full power during AGR 5/6/7 irradiation. These data consist of 110,388 release rate records and 110,388 R/B records for the twelve radionuclides (Kr 85m, Kr 87, Kr 88, Kr 89, Kr 90, Xe 131m, Xe 133, Xe 135, Xe 135m, Xe 137, Xe 138, and Xe 139) for each of the five capsules. Equivalent numbers of uncertainty records associated the release rates and R/B values were provided. To date, qualification status of the FPMS data stored in the NDMAS dat

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Automatic segmentation of head and neck CT images for radiotherapy treatment planning using multiple atlases, statistical appearance models, and geodesic active contours

Purpose: Accurate delineation of organs at risk (OARs) is a precondition for intensity modulated radiation therapy. However, manual delineation of OARs is time consuming and prone to high interobserver variability. Because of image artifacts and low image contrast between different structures, however, the number of available approaches for autosegmentation of structures in the head-neck area is still rather low. In this project, a new approach for automated segmentation of head-neck CT images that combine the robustness of multiatlas-based segmentation with the flexibility of geodesic active contours and the prior knowledge provided by statistical appearance models is presented. Methods: The presented approach is using an atlas-based segmentation approach in combination with label fusion in order to initialize a segmentation pipeline that is based on using statistical appearance models and geodesic active contours. An anatomically correct approximation of the segmentation result provided by atlas-based segmentation acts as a starting point for an iterative refinement of this approximation. The final segmentation result is based on using model to image registration and geodesic active contours, which are mutually influencing each other. Results: 18 CT images in combination with manually segmented labels of parotid glands and brainstem were used in a leave-one-out cross validation scheme in order to evaluate the presented approach. For this purpose, 50 different statistical appearance models have been created and used for segmentation. Dice coefficient (DC), mean absolute distance and max. Hausdorff distance between the autosegmentation results and expert segmentations were calculated. An average Dice coefficient of DC = 0.81 (right parotid gland), DC = 0.84 (left parotid gland), and DC = 0.86 (brainstem) could be achieved. Conclusions: The presented framework provides accurate segmentation results for three important structures in the head neck area. Compared to a segmentation approach based on using multiple atlases in combination with label fusion, the proposed hybrid approach provided more accurate results within a clinically acceptable amount of time.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Identification of Cross-Linked Peptides after Click-Based Enrichment Using Sequential Collision-Induced Dissociation and Electron Transfer Dissociation Tandem Mass Spectrometry

Chemical cross-linking combined with mass spectrometric analysis is emerging as a powerful technique for protein-protein interaction and protein structure elucidation studies.1 Cross-linkers covalently link two interacting proteins, often with chemistries specific to certain amino acid side chains. After enzymatic digestion of the proteins, the resulting cross-linked peptides can be subjected to analysis by LC-MS(/MS) to identify cross-linked species.2,3 For studying protein interactions using chemical cross-linking towards global discovery-based applications, the critical needs are the development of cross-linkers that are highly specific, amenable to LC-MS/MS, and resulting spectra are interpretable by bioinformatics tools to automatically assign cross-linked peptides with high confidence.4-10 As recently mentioned by Aebersold and co-workers, due to the low relative abundances of cross-linking products compared to their unmodified counterparts, enrichment of cross-linked species is also highly desirable to improve the likelihood of unambiguous identification of cross-linked peptides.6 Most of the currently available enrichable cross-linkers are bulky and are not amenable to studying protein-protein interactions in vivo. To discover protein-protein interactions with high confidence, there is a need for chemical cross-linkers that can effectively label protein complexes, utilize mass spectrometry based bottom-up proteomics analysis pipelines and also contains enrichment functionality.

59 BASIC BIOLOGICAL SCIENCES↗

Semi-supervised Machine Learning Enables the Robust Detection of Multireference Character at Low Cost

Multireference (MR) diagnostics are common tools for identifying strongly correlated electronic structure that makes single-reference (SR) methods (e.g., density functional theory or DFT) insufficient for accurate property prediction. However, MR diagnostics typically require computationally demanding correlated wave function theory (WFT) calculations, and diagnostics often disagree or fail to predict MR effects on properties. To overcome these challenges, we introduce a semi-supervised machine learning (ML) approach with virtual adversarial training (VAT) of an MR classifier using 15 WFT and DFT MR diagnostics as inputs. In semi-supervised learning, only the most extreme SR or MR points are labeled, and the remaining point labels are learned. The resulting VAT model outperforms the alternatives, as quantified by the distinct property distributions of SR- and MR-classified molecules. Additionally, to reduce the cost of generating inputs to the VAT model, we leverage the VAT model’s robustness to noisy inputs by replacing WFT MR diagnostics with regression predictions in an MR decision engine workflow that preserves excellent performance. We demonstrate the transferability of our approach to larger molecules and those with distinct chemical composition from the training set. This MR decision engine demonstrates promise as a low-cost, high-accuracy approach to the automatic detection of strong correlation for predictive high-throughput screening.

36 MATERIALS SCIENCE↗

Exploring Continuous Seismic Data at an Industry Facility Using Unsupervised Machine Learning

Seismic data recorded at industrial sites contain valuable information on anthropogenic activities. With advances in machine learning and computing power, new opportunities have emerged to explore the seismic wavefield in these complex environments. We applied two unsupervised machine learning algorithms to analyze continuous seismic data collected from an industrial facility in Texas, United States. The Uniform Manifold Approximation and Projection for Dimension Reduction algorithm was used to reduce the dimensionality of the data and generate 2D embeddings. Then, the Hierarchical Density-Based Spatial Clustering of Applications with Noise method was employed to automatically group these embeddings into distinct signal clusters. Our analysis of over 1400 hr (around 59 days) of continuous seismic data revealed five and seven signal clusters at two separate stations. At both stations, we identified clusters associated with background noise and vehicle traffic, with the latter’s temporal patterns aligning closely with the facility’s work schedule. Furthermore, the algorithms detected signal clusters from unknown sources and underline the ability of unsupervised machine learning for uncovering previously unrecognized patterns. Our analysis demonstrates the effectiveness of unsupervised approaches in examining continuous seismic data without requiring prior knowledge or pre-existing labels.

58 GEOSCIENCES↗

Machine Learning-Based Identification of the Interface Regions for Coupling Local and Nonlocal Models

Local-nonlocal coupling approaches provide a means to combine the computational efficiency of local models and the accuracy of nonlocal models. However, the coupling process can be challenging, requiring expertise to identify the interface between local and nonlocal regions. Here, this study introduces a machine learning-based approach to automatically detect the regions in which the local and nonlocal models should be used. The method uses loading functions evaluated at grid points to decide the model selection at those points. Training of the networks is based on datasets provided by classes of loading functions for which reference coupling configurations are computed using accurate coupled solutions, where accuracy is measured in terms of the relative error between the solution to the coupling approach and the solution to the nonlocal model. We study two approaches that vary in data structure. The first, the full-domain input data approach, uses the entire load vector and outputs a complete label vector, performing a global classification. The second, a window-based approach, processes loads into windows and addresses the problem as a node-wise classification where each window's central point is classified individually. The classification problems are solved via deep learning algorithms based on convolutional neural networks. The performance of these approaches is studied on one-dimensional numerical examples using F1-scores and accuracy metrics. Notably, the windowing approach achieves an accuracy of 0.96 and an F1-score of 0.97, highlighting its potential to automate coupling processes effectively and enhance computational efficiency in material science applications.

97 MATHEMATICS AND COMPUTING↗

Graph Learning in Knowledge Bases

The amount of text data has been growing exponentially in recent years, giving rise to automatic information extraction methods that store text annotations in a database. The current state-of-theart structured prediction methods, however, are likely to contain errors and it’s important to be able to manage the overall uncertainty of the database. On the other hand, the advent of crowdsourcing has enabled humans to aid machine algorithms at scale. As part of this project we introduced pi-CASTLE , a system that optimizes and integrates human and machine computing as applied to a complex structured prediction problem involving conditional random fields (CRFs). We proposed strategies grounded in information theory to select a token subset, formulate questions for the crowd to label, and integrate these labelings back into the database using a method of constrained inference. On both a text segmentation task over academic citations and a named entity recognition task over tweets we showed an order of magnitude improvement in accuracy gain over baseline methods.

97 MATHEMATICS AND COMPUTING↗

Innovating the next generation of commercial smart building software

Nearly 30% of commercial building energy use is wasted due to equipment faults and HVAC controls problems. The result is increased emissions, compromised comfort and productivity, and less reliable coordination of building power needs with a clean grid. The energy impact alone represents $17 billion in potential savings. Today’s smart building software provides a robust solution to address these operational deficiencies. Energy management and information systems (EMIS) are saving up to 9% on average, with two-year paybacks. They are being incorporated into energy management processes, commissioning services, and utility programs. As effective as they are, two barriers prevent even deeper benefits; limited personnel to fix problems once they are identified, and the expense and time to manually implement changes in control systems. In partnership with the research community, the EMIS industry is developing new capabilities to overcome these barriers. Moving beyond siloed products for either fault detection and diagnostics, or optimal control, these new capabilities empower users to not only automatically identify faults, but also to push corrective action, and control improvements to their buildings. In this paper, several areas for enhancements are documented: ‘one-time’ correction of faults such as setpoints, schedules, and economizer lockouts; short-term active testing for automated proportional integral derivative (PID) loop tuning and functional testing; and continuous supervisory control for demand flexibility and year-round efficiency. Results are presented from a pair of partner implementations out of a dozen providers integrating these enhancements into their products, including field tests from across the country, and insights into operator acceptance and integration into operations and maintenance practices.

Casillas, Armando↗

3DBFSVBF (3D BatFinder Smart Video BioFilter and Multi-class BatFinder Smart Video BioFilter) [SWR-22-88]

Bats are notoriously difficult to study, therefore, identifying specific behavioral trends and the precise environmental conditions at the time of collision requires a monitoring solution that can reliably collect relevant data. To date, thermal infrared video surveillance has been extensively applied to study bats and has proven to be a powerful yet cumbersome tool. Current analytical approaches are time consuming because data processing data has not been fully automated. In the past, steps have been taken to record avian and bat activity in conjunction with complicated image processing techniques that separate species from other moving objects within the field of view (i.e. clouds and portions of the wind turbine). Once the videos are collected, the post-processing does not allow real time monitoring and identification, leading to a delay in both studying the behavior of these species and determining the effectiveness of any impact reduction strategy being studied. Moreover, object identification capability is lacking, thus limiting the usefulness of video data. To resolve these issues, we are using open source 3D computer vision and machine learning techniques allowing for automatic detection of objects in real-time with the ability to correlate these objects with environmental variables and recording the flight paths of each object. The machine learning has been trained on 3D data and allows for automated real-time data collection, identification and tracking, thereby eliminating the need for long and tedious post-analysis processing of the videos. This machine learning model is an added feature to the previous BatFinder Smart Video BioFilter and increases the accuracy of that systems classification by increasing the accuracy of identifying bats (90% accuracy) and insects (69% accuracy) to a 97% accuracy. There are two object classifier machine learning models, Binary and multi-classification. Binary object classifier labeled BatFinder_Smart_Video_BioFilter.h5 distinguishes between biological objects and non-biological objects. The main goal of this object classifier is to ignore the turbine blades while detecting biological object flying withing the rotor swept area of the turbine. Non-biological objects have a probability of 0 and biological objects have a probability of 1. Multi-classifier labeled Multiclass_BatFinder_Smart_Video_BioFilter.h5 distinguishes between bats, birds, insects and non-biological.

Yarbrough, John↗

Use of Machine Learning on PMU Data for Transmission System Fault Analysis

Synchrophasor technology has been used for monitoring, control, and protection of bulk power system for over 10 years. Deployment of phasor measurement units (PMUs) in the USA power system has surpassed 3000 units installed in the transmission substations as stand-alone intelligent electronic devices (IEDs) or as a software add-on to other devices such as digital protective relays (DPRs) or digital fault recorders (DFRs). By now, thousands of terabytes of PMU data may have been captured and stored by various transmission system operators (TSOs) and independent system operators (ISOs). This creates an opportunity to deploy advanced machine learning (ML) techniques to detect and classify faults recorded by PMUs automatically to be used by the system operators for rapid, critical decision-making when manual analysis of the past or unfolding events is not feasible. In this paper we offer a brief background on how the automated fault analysis may be done using DPR and/or DFR data, and compare some of the legacy approaches to the new ML approaches in the context of the system-wide PMU recordings. We then offer insights from developing practical ML solutions that have been applied on field recordings captured by close to 450 PMUs from all three US interconnections (Western, Eastern and ERCOT) over two years (2016-2017). We identify and illustrate ML challenges we addressed: inaccurate data, data with scarce and temporally imprecise fault labels, data recorded by PMUs sparsely located at substations resulting in the fault records taken afar from the ends of the faulted lines, data containing only positive sequence values, and data taken at different voltage levels. We then illustrate the ML model results for fault analysis under different application scenarios. The novelty of this study is not only in the design, implementation, and performance analysis of the ML algorithms, but also in the use of advanced fault modelling and simulation approaches to improve the training results when developing supervised ML models for fault detection and classification. Extensive simulations of faults were conducted on a 14-bus power system to create a training dataset with over 1400 accurately labelled faults. This dataset was applied to enhance the accuracy of fault detection and classification of machine learning-based models trained with small number of labelled faults in large datasets recorded in the grid interconnections ranging from 5,000 to 70,000 buses.

Synchrophasors, Machine Learning, Fault Analysis, ↗

Gradient-Based Novelty Detection Boosted by Self-Supervised Binary Classification

Novelty detection aims to automatically identify out-of-distribution (OOD) data, without any prior knowledge of them. It is a critical step in data monitoring, behavior analysis and other applications, helping enable continual learning in the field. Conventional methods of OOD detection perform multi-variate analysis on an ensemble of data or features, and usually resort to the supervision with OOD data to improve the accuracy. In reality, such supervision is impractical as one cannot anticipate the anomalous data. In this paper, we propose a novel, self-supervised approach that does not rely on any pre-defined OOD data: (1) The new method evaluates the Mahalanobis distance of the gradients between the in-distribution and OOD data. (2) It is assisted by a self-supervised binary classifier to guide the label selection to generate the gradients, and maximize the Mahalanobis distance. In the evaluation with multiple datasets, such as CIFAR-10, CIFAR-100, SVHN and TinyImageNet, the proposed approach consistently outperforms state-of-the-art supervised and unsupervised methods in the area under the receiver operating characteristic (AUROC) and area under the precision-recall curve (AUPR) metrics. We further demonstrate that this detector is able to accurately learn one OOD class in continual learning.

Sun, Jingbo↗

Separable physics-informed DeepONet: Breaking the curse of dimensionality in physics-informed machine learning

The deep operator network (DeepONet) has shown remarkable potential in solving partial differential equations (PDEs) by mapping between infinite-dimensional function spaces using labeled datasets. However, in scenarios lacking labeled data, the physics-informed DeepONet (PI-DeepONet) approach, which utilizes the residual loss of the governing PDE to optimize the network parameters, faces significant computational challenges, particularly due to the curse of dimensionality. This limitation has hindered its application to high-dimensional problems, making even standard 3D spatial with 1D temporal problems computationally prohibitive. Additionally, the computational requirement increases exponentially with the discretization density of the domain. Here, to address these challenges and enhance scalability for high-dimensional PDEs, we introduce the Separable physics-informed DeepONet (Sep-PI-DeepONet). This framework employs a factorization technique, utilizing sub-networks for individual one-dimensional coordinates, thereby reducing the number of forward passes and the size of the Jacobian matrix required for gradient computations. By incorporating forward-mode automatic differentiation (AD), we further optimize computational efficiency, achieving linear scaling of computational cost with discretization density and dimensionality, making our approach highly suitable for high-dimensional PDEs. We demonstrate the effectiveness of Sep-PI-DeepONet through three benchmark PDE models: the viscous Burgers’ equation, Biot’s consolidation theory, and a parameterized heat equation. Our framework maintains accuracy comparable to the conventional PI-DeepONet while reducing training time by two orders of magnitude. Notably, for the heat equation solved as a 4D problem, the conventional PI-DeepONet was computationally infeasible (estimated 289.35 h), while the Sep-PI-DeepONet completed training in just 2.5 h. These results underscore the potential of Sep-PI-DeepONet in efficiently solving complex, high-dimensional PDEs, marking a significant advancement in physics-informed machine learning.

Neural operator↗

Alert Classification for the ALeRCE Broker System: The Light Curve Classifier

We present the first version of the Automatic Learning for the Rapid Classification of Events (ALeRCE) broker light curve classifier. ALeRCE is currently processing the Zwicky Transient Facility (ZTF) alert stream, in preparation for the Vera C. Rubin Observatory. The ALeRCE light curve classifier uses variability features computed from the ZTF alert stream and colors obtained from AllWISE and ZTF photometry. We apply a balanced random forest algorithm with a two-level scheme where the top level classifies each source as periodic, stochastic, or transient, and the bottom level further resolves each of these hierarchical classes among 15 total classes. This classifier corresponds to the first attempt to classify multiple classes of stochastic variables (including core- and host-dominated active galactic nuclei, blazars, young stellar objects, and cataclysmic variables) in addition to different classes of periodic and transient sources, using real data. We created a labeled set using various public catalogs (such as the Catalina Surveys and Gaia DR2 variable stars catalogs, and the Million Quasars catalog), and we classify all objects with ≥6 g-band or ≥6 r-band detections in ZTF (868,371 sources as of 2020 June 9), providing updated classifications for sources with new alerts every day. For the top level we obtain macro-averaged precision and recall scores of 0.96 and 0.99, respectively, and for the bottom level we obtain macro-averaged precision and recall scores of 0.57 and 0.76, respectively. Updated classifications from the light curve classifier can be found at the ALeRCE Explorer website (http://alerce.online).

47 OTHER INSTRUMENTATION↗

NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations

In the last 50 years there has been a tremendous progress in solving numerically the Navier-Stokes equations using finite differences, finite elements, spectral, and even meshless methods. Yet, in many real cases, we still cannot incorporate seamlessly (multi-fidelity) data into existing algorithms, and for industrial-complexity applications the mesh generation is time consuming and still an art. Moreover, solving ill-posed problems (e.g., lacking boundary conditions) or inverse problems is often prohibitively expensive and requires different formulations and new computer codes. Here, we employ physics-informed neural networks (PINNs), encoding the governing equations directly into the deep neural network via automatic differentiation, to overcome some of the aforementioned limitations for simulating incompressible laminar and turbulent flows. We develop the Navier-Stokes flow nets (NSFnets) by considering two different mathematical formulations of the Navier-Stokes equations: the velocity-pressure (VP) formulation and the vorticity-velocity (VV) formulation. Since this is a new approach, we first select some standard benchmark problems to assess the accuracy, convergence rate, computational cost and flexibility of NSFnets; analytical solutions and direct numerical simulation (DNS) databases provide proper initial and boundary conditions for the NSFnet simulations. The spatial and temporal coordinates are the inputs of the NSFnets, while the instantaneous velocity and pressure fields are the outputs for the VP-NSFnet, and the instantaneous velocity and vorticity fields are the outputs for the VV-NSFnet. This is unsupervised learning and, hence, no labeled data are required beyond boundary and initial conditions and the fluid properties. The residuals of the VP or VV governing equations, together with the initial and boundary conditions, are embedded into the loss function of the NSFnets. No data is provided for the pressure to the VP-NSFnet, which is a hidden state and is obtained via the incompressibility constraint without extra computational cost. Unlike the traditional numerical methods, NSFnets inherit the properties of neural networks (NNs), hence the total error is composed of the approximation, the optimization, and the generalization errors. Here, we empirically attempt to quantify these errors by varying the sampling (“residual”) points, the iterative solvers, and the size of the NN architecture. For the laminar flow solutions, we show that both the VP and the VV formulations are comparable in accuracy but their best performance corresponds to different NN architectures. The initial convergence rate is fast but the error eventually saturates to a plateau due to the dominance of the optimization error. For the turbulent channel flow, we show that NSFnets can sustain turbulence at , but due to expensive training we only consider part of the channel domain and enforce velocity boundary conditions on the subdomain boundaries provided by the DNS data base. We also perform a systematic study on the weights used in the loss function for balancing the data and physics components, and investigate a new way of computing the weights dynamically to accelerate training and enhance accuracy. In the last part, we demonstrate how NSFnets should be used in practice, namely for ill-posed problems with incomplete or noisy boundary conditions as well as for inverse problems. We obtain reasonably accurate solutions for such cases as well without the need to change the NSFnets and at the same computational cost as in the forward well-posed problems. As a result, we also present a simple example of transfer learning that will aid in accelerating the training of NSFnets for different parameter settings.

97 MATHEMATICS AND COMPUTING↗

Low-complexity atlas-based prostate segmentation by combining global, regional, and local metrics

Purpose: To improve the efficiency of atlas-based segmentation without compromising accuracy, and to demonstrate the validity of the proposed method on MRI-based prostate segmentation application. Methods: Accurate and efficient automatic structure segmentation is an important task in medical image processing. Atlas-based methods, as the state-of-the-art, provide good segmentation at the cost of a large number of computationally intensive nonrigid registrations, for anatomical sites/structures that are subject to deformation. In this study, the authors propose to utilize a combination of global, regional, and local metrics to improve the accuracy yet significantly reduce the number of required nonrigid registrations. The authors first perform an affine registration to minimize the global mean squared error (gMSE) to coarsely align each atlas image to the target. Subsequently, atarget-specific regional MSE (rMSE), demonstrated to be a good surrogate for dice similarity coefficient (DSC), is used to select a relevant subset from the training atlas. Only within this subset are nonrigid registrations performed between the training images and the target image, to minimize a weighted combination of gMSE and rMSE. Finally, structure labels are propagated from the selected training samples to the target via the estimated deformation fields, and label fusion is performed based on a weighted combination of rMSE and local MSE (lMSE) discrepancy, with proper total-variation-based spatial regularization. Results: The proposed method was applied to a public database of 30 prostate MR images with expert-segmented structures. The authors’ method, utilizing only eight nonrigid registrations, achieved a performance with a median/mean DSC of over 0.87/0.86, outperforming the state-of-the-art full-fledged atlas-based segmentation approach of which the median/mean DSC was 0.84/0.82 when applying to their data set. Conclusions: The proposed method requires a fixed number of nonrigid registrations, independent of atlas size, providing desirable scalability especially important for a large or growing atlas. When applied to prostate segmentation, the method achieved better performance to the state-of-the-art atlas-based approaches, with significant improvement in computation efficiency. The proposed rationale of utilizing jointly global, regional, and local metrics, based on the information characteristic and surrogate behavior for registration and fusion subtasks, can be extended naturally to similarity metrics beyond MSE, such as correlation or mutual information types.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Automated segmentation of dental CBCT image with prior-guided sequential random forests

Purpose: Cone-beam computed tomography (CBCT) is an increasingly utilized imaging modality for the diagnosis and treatment planning of the patients with craniomaxillofacial (CMF) deformities. Accurate segmentation of CBCT image is an essential step to generate 3D models for the diagnosis and treatment planning of the patients with CMF deformities. However, due to the image artifacts caused by beam hardening, imaging noise, inhomogeneity, truncation, and maximal intercuspation, it is difficult to segment the CBCT. Methods: In this paper, the authors present a new automatic segmentation method to address these problems. Specifically, the authors first employ a majority voting method to estimate the initial segmentation probability maps of both mandible and maxilla based on multiple aligned expert-segmented CBCT images. These probability maps provide an important prior guidance for CBCT segmentation. The authors then extract both the appearance features from CBCTs and the context features from the initial probability maps to train the first-layer of random forest classifier that can select discriminative features for segmentation. Based on the first-layer of trained classifier, the probability maps are updated, which will be employed to further train the next layer of random forest classifier. By iteratively training the subsequent random forest classifier using both the original CBCT features and the updated segmentation probability maps, a sequence of classifiers can be derived for accurate segmentation of CBCT images. Results: Segmentation results on CBCTs of 30 subjects were both quantitatively and qualitatively validated based on manually labeled ground truth. The average Dice ratios of mandible and maxilla by the authors’ method were 0.94 and 0.91, respectively, which are significantly better than the state-of-the-art method based on sparse representation (p-value < 0.001). Conclusions: The authors have developed and validated a novel fully automated method for CBCT segmentation.

61 RADIATION PROTECTION AND DOSIMETRY↗

Automated System-wide Event Detection and Classification Using Machine Learning on Synchrophasor Data

As the number of phasor measurement units (PMUs) deployed in a power system increases, and their data volume streamed to the control canter intensifies, operators are facing challenges related to the analysis of such data, which need to be observed and responded to as the measurements are displayed in the Control Room. Humans are generally unable to process such large amount of data efficiently and rapidly. There is an apparent need for automated ways to analyze the data, extract actionable information about occurrence of specific events, and characterize the events quickly and cost effectively. This paper discusses the use of machine learning (ML) to facilitate such tasks by providing automated, highly computationally efficient, and cost-effective ways of extracting actionable information from synchrophasor big data in real-time. We developed Big Data Smart (BDSmart) ML-based prototype tool for the Control Room use that automatically analyses data properties from synchrophasor system measurements taken across the three grid Interconnections in the USA (Western, Eastern and ERCOT). The data collected from several hundreds of PMUs located across the Interconnections over a period of two years have been made available for our extensive study. As a result, we were able to identify a number of big data properties that influence how ML methodology is applied to select, develop, train and test the data models that can eventually be used for the tool implementation. The resulting set of candidate algorithms spans unsupervised, supervised, semi-supervised and transfer-learning approaches. Many ML techniques, such as decision trees, multinomial logistic regression, feed-forward neural networks, K-nearest neighbor, multiclass support vector machine, and single and multi-channel convolutional neural networks, are implemented, and their performance is examined. We offer the results from testing the data models. The novelty of our study is in the approaches for bad data detection and mitigation, selection of a simplified feature for event detection, and data label improvements. As a result, we came up with a list of recommendations for the utilities on how to improve the PMU recording practices to cater to the future ML applications aimed at automating the analysis of synchrophasor data.

Synchrophasors, Machine Learning, System-wide Even↗