Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Anomaly Detection”

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 865 records · Page 48

Advances in Bayesian Model Based Clustering Using Particle Learning

Recent work by Carvalho, Johannes, Lopes and Polson and Carvalho, Lopes, Polson and Taddy introduced a sequential Monte Carlo (SMC) alternative to traditional iterative Monte Carlo strategies (e.g. MCMC and EM) for Bayesian inference for a large class of dynamic models. The basis of SMC techniques involves representing the underlying inference problem as one of state space estimation, thus giving way to inference via particle filtering. The key insight of Carvalho et al was to construct the sequence of filtering distributions so as to make use of the posterior predictive distribution of the observable, a distribution usually only accessible in certain Bayesian settings. Access to this distribution allows a reversal of the usual propagate and resample steps characteristic of many SMC methods, thereby alleviating to a large extent many problems associated with particle degeneration. Furthermore, Carvalho et al point out that for many conjugate models the posterior distribution of the static variables can be parametrized in terms of [recursively defined] sufficient statistics of the previously observed data. For models where such sufficient statistics exist, particle learning as it is being called, is especially well suited for the analysis of streaming data do to the relative invariance of its algorithmic complexity with the number of data observations. Through a particle learning approach, a statistical model can be fit to data as the data is arriving, allowing at any instant during the observation process direct quantification of uncertainty surrounding underlying model parameters. Here we describe the use of a particle learning approach for fitting a standard Bayesian semiparametric mixture model as described in Carvalho, Lopes, Polson and Taddy. In Section 2 we briefly review the previously presented particle learning algorithm for the case of a Dirichlet process mixture of multivariate normals. In Section 3 we describe several novel extensions to the original implementation of Carvalho et al that allow us to retain the computational advantages of particle learning while improving the suitability of the methodology to the analysis of streaming data and simultaneously facilitating the real time discovery of latent cluster structures. Section 4 demonstrates our methodological enhancements in the context of several simulated and classical data sets, showcasing the use of particle learning methods for online anomaly detection, label generation, drift detection, and semi-supervised classification, none of which would be achievable through a standard MCMC approach. Section 5 concludes with a discussion of future directions for research.

99 GENERAL AND MISCELLANEOUS↗

A Novel Passive Millimeter Imager for Broad-Area Search (Final Report)

This report describes research and development efforts toward a novel passive millimeter-wave (mm-wave) electromagnetic imaging device for broad-area search. It addresses the technical challenge of detecting anomalies that occupy a small fraction of a pixel. The purpose of the imager is to pinpoint suspicious locations for cuing subsequent higher-resolution imaging. The technical basis for the approach is to exploit thermal and polarization anomalies that distinguish man-made features from natural features.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Numerical Methods for Large-Scale Data Factorization (Final Report (Milestone Date 9/30/11) for Subcontract No. B594099)

The following work has been performed by PI Hans De Sterck and graduate student Manda Winlaw for the required tasks 1-5 (as listed in the Statement of Work). Graduate student Manda Winlaw has visited LLNL January 31-March 11, 2011 and May 23-August 19, 2010, working with Van Henson and Mike O'Hara on non-negative matrix factorizations (NMF). She has investigated the dense subgraph clustering algorithm from 'Finding Dense Subgraphs for Sparse Undirected, Directed, and Bipartite Graphs' by Chen and Saad, testing this method on several term-document matrices and adapting it to cluster based on the rank of the subgraphs instead of the density. Manda Winlaw was awarded a first prize in the annual LLNL summer student poster competition for a poster on her NMF research. PI Hans De Sterck has developed a new adaptive algebraic multigrid algorithm for computing a few dominant or minimal singular triplets of sparse rectangular matrices. This work builds on adaptive algebraic multigrid methods that were further developed by the PI and collaborators (including Sanders and Henson) for Markov chains. The method also combines and extends existing multigrid algorithms for the symmetric eigenproblem. The PI has visited LLNL February 22-25, 2011, and has given a CASC seminar 'Algebraic Multigrid for the Singular Value Problem' on this work on February 23, 2011. During his visit, he has discussed this work and related topics with Van Henson, Geoffrey Sanders, Panayot Vassilevski, and others. He has tested the algorithm on PDE matrices and on a term-document matrix, with promising initial results. Manda Winlaw has also started to work, with O'Hara, on estimating probability distributions over undirected graph edges. The goal is to estimate probabilistic models from sets of undirected graph edges for the purpose of prediction, anomaly detection and support to supervised learning. Graduate student Manda Winlaw is writing a paper on the results obtained with O'Hara which will be submitted some time later in 2011 to a data mining conference. PI Hans De Sterck has developed a new optimization algorithm for canonical tensor approximation, formulating an extension of the nonlinear GMRES method to optimization problems. Numerical results for tensors with up to 8 modes show that this new method is efficient for sparse and dense tensors. He has written a paper on this which has been submitted to the SIAM Journal on Scientific Computing. PI Hans De Sterck has further developed his new optimization algorithm for canonical tensor approximation, formulating an extension in terms of steepest-descent preconditioning, which makes the approach generally applicable for nonlinear optimization. He has written a paper on this extension which has been submitted to Numerical Linear Algebra with Applications.

97 MATHEMATICS AND COMPUTING↗

Multilevel Method for Modeling Large-Scale Networks

Understanding the behavior of real complex networks is of great theoretical and practical significance. It includes developing accurate artificial models whose topological properties are similar to the real networks, generating the artificial networks at different scales under special conditions, investigating a network dynamics, reconstructing missing data, predicting network response, detecting anomalies and other tasks. Network generation, reconstruction, and prediction of its future topology are central issues of this field. In this project, we address the questions related to the understanding of the network modeling, investigating its structure and properties, and generating artificial networks. Most of the modern network generation methods are based either on various random graph models (reinforced by a set of properties such as power law distribution of node degrees, graph diameter, and number of triangles) or on the principle of replicating an existing model with elements of randomization such as R-MAT generator and Kronecker product modeling. Hierarchical models operate at different levels of network hierarchy but with the same finest elements of the network. However, in many cases the methods that include randomization and replication elements on the finest relationships between network nodes and modeling that addresses the problem of preserving a set of simplified properties do not fit accurately enough the real networks. Among the unsatisfactory features are numerically inadequate results, non-stability of algorithms on real (artificial) data, that have been tested on artificial (real) data, and incorrect behavior at different scales. One reason is that randomization and replication of existing structures can create conflicts between fine and coarse scales of the real network geometry. Moreover, the randomization and satisfying of some attribute at the same time can abolish those topological attributes that have been undefined or hidden from researchers. We propose to develop multilevel methods to model complex networks. The key point of the proposed strategy is that it will help to preserve part of the unknown structural attributes by guaranteeing the similar behavior of the real and artificial model on different scales.

99 GENERAL AND MISCELLANEOUS↗

Summary of the Graduate Mentoring Assistance Program for PNNL PNNL.12.006 (April 2012 – May 31, 2013)

Summary of Students and Projects: Sarah Bender: The possible improvement of the Multi-Isotope Process (MIP) Monitor by the addition of Compton suppression is investigated for this study. The Compton continuum from the dominant 661.7 keV Cs-137 fission product peak obscures lower energy lines which could be used for spectral and multivariate analysis. A novel Compton suppression system which uses fast, room-temperature materials will be developed using simulation. As opposed to the standard high purity germanium (HPGe) detector suppressed using a thallium-doped sodium iodide (NaI(Tl)) guard detector array, cerium-doped lanthanum bromide (LaBr3) will be used as the primary spectroscopic detector. For Sarah’s Master’s degree, she constructed and validated the unsuppressed detector response of the Pennsylvania State University (PSU)-Compton suppression system. For Sarah’s PhD, she is completing that simulation by including the suppressed operation which will be validated with spent fuel samples. She is also going to use the validation of suppressed operation to predict the detector response of a LaBr3 based Compton suppression system. Richard Clark: Richard Clark completed his study this FY of a new technique known as Intrinsic Dosimetry. Intrinsic Dosimetry is the method of measuring total absorbed dose received by the walls of a container holding radioactive material. By considering the total absorbed dose received by a container in tandem with the physical characteristics of the radioactive material housed within that container, this method has the potential to provide enhanced pathway information regarding the history of the container and its radioactive contents. The latest in a series of experiments designed to validate and demonstrate this newly developed tool are reported. Kenneth Dayman: The MIP Monitor is an online, non-destructive approach to process monitoring in a nuclear fuel reprocessing facility using gamma spectra taken of process streams and multivariate data analysis to detect anomalies in operation. Last year, Mr. Dayman investigated methods to classify spent nuclear fuel according to both reactor type and burn up based on the constituent nuclide activities, which may be used to more accurately calibrate other models in an analysis scheme in order to improve the accuracy and precision of predictions. He is currently writing up his results for publication and has transitioned to another project with Dr. Derrick Haas at PNNL.

Schwantes, Jon M.↗

Sampling Approaches for Multi-Domain Internet Performance Measurement Infrastructures

The next-generation of high-performance networks being developed in DOE communities are critical for supporting current and emerging data-intensive science applications. The goal of this project is to investigate multi-domain network status sampling techniques and tools to measure/analyze performance, and thereby provide “network awareness” to end-users and network operators in DOE communities. We leverage the infrastructure and datasets available through perfSONAR, which is a multi-domain measurement framework that has been widely deployed in high-performance computing and networking communities; the DOE community is a core developer and the largest adopter of perfSONAR. Our investigations include development of semantic scheduling algorithms, measurement federation policies, and tools to sample multi-domain and multi-layer network status within perfSONAR deployments. We validate our algorithms and policies with end-to-end measurement analysis tools for various monitoring objectives such as network weather forecasting, anomaly detection, and fault-diagnosis. In addition, we develop a multi-domain architecture for an enterprise-specific perfSONAR deployment that can implement monitoring-objective based sampling and that adheres to any domain-specific measurement policies.

97 MATHEMATICS AND COMPUTING↗

Energy information systems (EIS): Technology costs, benefit, and best practice uses

Energy information systems are the web-based software, data acquisition hardware, and communication systems used to store, analyze, and display building energy data. They often include analysis methods such as baselining, benchmarking, load profiling, and energy anomaly detection. This report documents a large-scale assessment of energy information system (EIS) uses, costs, and energy benefits, based on a series of focused case study investigations that are synthesized into generalizable findings. The overall objective is to provide organizational decision makers with the information they need to make informed choices as to whether or not to invest in an EIS--a promising technology that can enable up to 20 percent site energy savings, quick payback, and persistent low-energy performance when implemented as part of best-practice energy management programs.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

Modeling an Application's Theoretical Minimum and Average Transactional Response Times

The theoretical minimum transactional response time of an application serves as a basis for the expected response time. The lower threshold for the minimum response time represents the minimum amount of time that the application should take to complete a transaction. Knowing the lower threshold is beneficial in detecting anomalies that are results of unsuccessful transactions. On the converse, when an application's response time falls above an upper threshold, there is likely an anomaly in the application that is causing unusual performance issues in the transaction. This report explains how the non-stationary Generalized Extreme Value distribution is used to estimate the lower threshold of an application's daily minimum transactional response time. It also explains how the seasonal Autoregressive Integrated Moving Average time series model is used to estimate the upper threshold for an application's average transactional response time.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

Cyber Graph Queries for Geographically Distributed Data Centers

We present new algorithms for a distributed model for graph computations motivated by limited information sharing we first discussed. Two or more independent entities have collected large social graphs. They wish to compute the result of running graph algorithms on the entire set of relationships. Because the information is sensitive or economically valuable, they do not wish to simply combine the information in a single location. We consider two models for computing the solution to graph algorithms in this setting: 1) limited-sharing: the two entities can share only a polylogarithmic size subgraph; 2) low-trust: the entities must not reveal any information beyond the query answer, assuming they are all honest but curious. We believe this model captures realistic constraints on cooperating autonomous data centers. We have algorithms in both setting for s - t connectivity in both models. We also give an algorithm in the low-communication model for finding a planted clique. This is an anomaly-detection problem, finding a subgraph that is larger and denser than expected. For both the low- communication algorithms, we exploit structural properties of social networks to prove performance bounds better than what is possible for general graphs. For s - t connectivity, we use known properties. For planted clique, we propose a new property: bounded number of triangles per node. This property is based upon evidence from the social science literature. We found that classic examples of social networks do not have the bounded-triangles property. This is because many social networks contain elements that are non-human, such as accounts for a business, or other automated accounts. We describe some initial attempts to distinguish human nodes from automated nodes in social networks based only on topological properties.

97 MATHEMATICS AND COMPUTING↗

A Clustering Graph Generator

In very simple terms a network can be de ned as a collection of points joined together by lines. Thus, networks can be used to represent connections between entities in a wide variety of elds including engi- neering, science, medicine, and sociology. Many large real-world networks share a surprising number of properties, leading to a strong interest in model development research and techniques for building synthetic networks have been developed, that capture these similarities and replicate real-world graphs. Modeling these real-world networks serves two purposes. First, building models that mimic the patterns and prop- erties of real networks helps to understand the implications of these patterns and helps determine which patterns are important. If we develop a generative process to synthesize real networks we can also examine which growth processes are plausible and which are not. Secondly, high-quality, large-scale network data is often not available, because of economic, legal, technological, or other obstacles [7]. Thus, there are many instances where the systems of interest cannot be represented by a single exemplar network. As one example, consider the eld of cybersecurity, where systems require testing across diverse threat scenarios and validation across diverse network structures. In these cases, where there is no single exemplar network, the systems must instead be modeled as a collection of networks in which the variation among them may be just as important as their common features. By developing processes to build synthetic models, so-called graph generators, we can build synthetic networks that capture both the essential features of a system and realistic variability. Then we can use such synthetic graphs to perform tasks such as simulations, analysis, and decision making. We can also use synthetic graphs to performance test graph analysis algorithms, including clustering algorithms and anomaly detection algorithms.

97 MATHEMATICS AND COMPUTING↗

Curvature Based Analysis to Identify and Categorize Trajectory Segments

Since the attacks carried out against the United States on September 11, 2001, which involved the commandeering of commercial aircraft, interest has increased in performing trajectory analysis of vehicle types not constrained by roadways or railways, i.e., aircraft and watercraft. Anomalous trajectories need to be automatically identified along with other trajectories of interest to flag them for further investigation. There is also interest in analyzing trajectories without a focus on anomaly detection. Various approaches to analyzing these trajectories have been undertaken with useful results to date. In this research, we seek to augment trajectory analysis by carrying out analysis of the trajectory curvature along with other parameters, including distance and total deflection (change in direction). At each point triplet in the ordered sequence of points, these parameters are computed. Adjacent point triplets with similar values are grouped together to form a higher level of semantic categorization. These categorizations are then analyzed to form a yet higher level of categorization which has more specific semantic meaning. This top level of categorization is then summarized for all trajectories under study, allowing for fast identification of trajectories with various semantic characteristics.

42 ENGINEERING↗

Distributed Magnetic Measurements Using Fiber Optic Pulsed Polarimetry (Final Report)

Measurement of the magnetic fields surrounding magnetic fusion energy (MFE) devices, such as tokamaks, is critical for characterizing and monitoring the plasma inside these devices, and it is especially important for detecting plasma instabilities, which can lead to the plasma damaging the inside of these devices. This program successfully demonstrated a new type of magnetic field diagnostic that is capable for the first time of measuring simultaneously the spatial and temporal magnetic field distributions of MFE devices in real-time and with high resolution. This is a capability not feasible with existing magnetic field diagnostics and, moreover, the technique we developed provides other advantages, such as less susceptibility to noise, the ability to operate during both pulsed and steady-state plasma conditions, the ability to detect relatively weak poloidal fields, and potentially lower costs. This new capability of being able in real-time to detect anomalies in the magnetic field distribution corresponding to the onset of plasma instabilities offers the ability to automatically shut down the MFE device before the plasma can damage the machine. Hence, this machine protection capability can be vital for future fusion power stations. Our technique uses optical fibers where changes in the polarization of the laser light traveling through the fiber provide a direct measurement of the magnetic field at any point along the fiber. The small size of the fibers allows them to be easily installed either outside or inside the MFE vacuum vessel, and in any direction one wishes (e.g., poloidal or toroidal). We demonstrated the basic capabilities of our new diagnostic during a proof-of-principle experiment performed on the General Atomics DIII-D tokamak, and published the results in: W. D. Kimura, D. S. Dhillon, S. Munaretto, E. J. Strait, D. Du, S. Loranger, J.-S. Boisvert, and R. Kashyap, “Measurements of DIII-D poloidal field by fiber-optic pulsed polarimetry,” Rev. Sci, Inst. 89, 10J102 (2018).

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Parallel Tensor Decompositions for Massive Heterogeneous, Incomplete Data (Final Report: LDRD Project 199986).

We extended the fundamental capabilities of tensor decomposition to a broader range of problems - handling new data types and larger problems. This has implications for data analysis across a range of applications in sensor monitoring, cybersecurity, treaty verification, signal processing, etc. Identifies latent structure within data, enabling anomaly detection, process monitoring, scientific discovery.

97 MATHEMATICS AND COMPUTING↗

Machine Learning for Synchrophasor Analysis

The report presents results from the development of a cloud-based, Big Data analysis framework for power systems. The computational pipeline uses the Apache Spark framework running in an OpenStack cloud infrastructure. A real-world phasor measurement unit (PMU) dataset has been used to carry out the analysis. Several Machine Learning (ML) methods have been developed and implemented for event and anomaly detection and classification. Actual examples of power system events detection and analysis using synchrophasor data are presented. It has been shown that applications of the cloud-based computing environment and the Apache Spark framework enable a significant increase in the computational efficiency of large-scale PMU data analysis.

20 FOSSIL-FUELED POWER PLANTS↗

The Fingerprints of Stratospheric Aerosol Injection in E3SM

The June 15, 1991 Mt. Pinatubo eruption is simulated in E3SM by injecting 10 Tg of SO 2 gas in the stratosphere, turning off prescribed volcanic aerosols, and enabling E3SM to treat stratospheric volcanic aerosols prognostically. This experimental prognostic treatment of volcanic aerosols in the stratosphere results in some realistic behaviors (SO 2 evolves into H 2 SO 4 which heats the lower stratosphere), and some expected biases (H 2 SO 4 aerosols sediment out of the stratosphere too quickly). Climate fingerprinting techniques are used to establish a Mt. Pinatubo fingerprint based on the vertical profile of temperature from the E3SMv1 DECK ensemble. By projecting reanalysis data and preindustrial simulations onto the fingerprint, the Mt. Pinatubo stratospheric heating anomaly is detected. Projecting the experimental prognostic aerosol simulation onto the fingerprint also results in a detectable heating anomaly, but, as expected, the duration is too short relative to reanalysis data.

54 ENVIRONMENTAL SCIENCES↗

Machine Learning Modeling Pipeline for Extracting Nuclear Proliferation Events of Interest from Open Data Sources (U)

In FY2020, the Savannah River National Laboratory (SRNL) and the Sanghani Center for Artificial Intelligence and Data Analytics at Virginia Polytechnic Institute and State University entered a collaboration funded by Department of Energy’s (DOE) Office of Defense Nuclear Nonproliferation Research and Development. The project’s mission was to take the first steps toward developing a demonstration prototype system that uses multiple machine learning and data analytics methods on largescale open data sources to identify new, developing, and/or undeclared nuclear programs. Given the SRNL team’s on-site perspective of events culminating in the DOE’s decision to pursue the Savannah River Plutonium Processing Facility (SRPPF), the team targeted the identification of events and indicators in retrospective datasets that pointed to the activity of “fissile core fabrication at the Savannah River Site” prior to the official announcement in May of 2018. A preliminary modeling pipeline was developed in FY20 that showed the datasets contained adequate signal for continuation of efforts. In FY21, a modular demonstration prototype modeling pipeline has continued in development for two text-based data sources: a broad internet archive (Webhose Ltd.) and a decahose Twitter database (i.e., a global sampling of one in every ten Tweets). The techniques that have been developed rely on graph theory and anomaly detection to identify contextual shifts in key words and phrases at various points in time such that indicators of events of interest could be identified and subsequently, events could be extracted from the corpuses. The foundational concept behind the approaches is that contextual shifts in key words and phrases can act as indicators of events of interest. Both datasets have proven successful in extracting events of interest related to pit production at the Savannah River Site prior to the official announcement. In addition, the pipelines have generated a wide range of events broadly summarized as: the awarding of DOE contracts at major sites, DOE investments in various programs, accidents at DOE national laboratories, speculations about the fate of pit production in the DOE complex, domestic and international shipments and receipts of nuclear materials at DOE sites, termination of non-proliferation agreements with Russia, termination of MOX, new weapons development approvals/testing, nuclear posture reviews, major DOE cleanup/production milestones, political opinions, and nuclear watch groups’ opinions, among many others.

97 MATHEMATICS AND COMPUTING↗

Data Science and Machine Learning for Genome Security

This report describes research conducted to use data science and machine learning methods to distinguish targeted genome editing versus natural mutation and sequencer machine noise. Genome editing capabilities have been around for more than 20 years, and the efficiencies of these techniques has improved dramatically in the last 5+ years, notably with the rise of CRISPR-Cas technology. Whether or not a specific genome has been the target of an edit is concern for U.S. national security. The research detailed in this report provides first steps to address this concern. A large amount of data is necessary in our research, thus we invested considerable time collecting and processing it. We use an ensemble of decision tree and deep neural network machine learning methods as well as anomaly detection to detect genome edits given either whole exome or genome DNA reads. The edit detection results we obtained with our algorithms tested against samples held out during training of our methods are significantly better than random guessing, achieving high F1 and recall scores as well as with precision overall.

59 BASIC BIOLOGICAL SCIENCES↗

Multi-modal Energy-optimal Trip Scheduling in Real-time (METS-R) for Transportation Hubs (Final Report)

This report summarizes the work performed under the award number EE0008524. The project develops the Multi-modal Energy-optimal Trip Scheduling in Real-time (METS-R) platform as the next-generation transportation solution based on autonomous electric vehicles (AEV) serving passenger trips from and to urban transportation hubs, to substantially reduce transportation energy consumption. Extensive data collection and analyses were first conducted to understand the demand patterns and energy consumption of hub-based on-road trips. Then, a data-driven framework that consists of an analytical module and a simulation module was proposed. For the analytical module, five planning + operation tools were developed to support the planning and energy-efficient operations of urban AEV services: the charging station planning that robotically allocates charging supplies based on the stationary charging demand distribution; the transit planning and demand adaptive scheduling model that efficiently generates\ candidate transit routes from hubs to other places and dynamically adjusts the transit time table to fit the current demand; the online energy-efficient routing that learns the energy-optimal paths from observations of link-level energy consumption in real-time; the hub-based ridesharing that matches trip requests together with account for the uncertainty of future trip demand and vehicle supply; and finally, the integrated demand prediction and anomaly detection pipeline that leverages the flight/train time table and support other planning/operation tools. To demonstrate the performance of these tools, a scalable high-performance agent-based simulator was built. We divided the urban space into multiple service zones where each zone was considered as an agent for passenger generation and vehicle charging. Two types of AEV agents were coded to model two types of mobility services: AEV taxi and AEV transit. For the AEV taxi, the team implemented the functions of pickup/drop-off passengers, energy-efficient routing, ridesharing, fleet rebalancing, and recharging. For the AEV bus, the team implemented the functions of demand-adaptive route scheduling, passenger boarding, and recharging. A high-performance computing framework was introduced to receive various profiling information (such as link energy updates, vehicle speed) from the simulator instances and communicate the operational commands back to the instances. The numerical experiments show that each of the proposed operational algorithms can reduce energy consumption and improve system efficiency. Furthermore, there exists the need to collectively consider multiple planning + operational strategies as multiple strategies can influence each other in terms of performance impacts. Recommendations for future work related to AEV planning and simulation are discussed.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗