UMD Theses and Dissertations
Permanent URI for this collectionhttp://hdl.handle.net/1903/3
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a given thesis/dissertation in DRUM.
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
3 results
Search Results
Item MACHINERY ANOMALY DETECTION UNDER INDETERMINATE OPERATING CONDITIONS(2018) Tian, Jing; Pecht, Michael; Mechanical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Anomaly detection is a critical task in system health monitoring. Current practice of anomaly detection in machinery systems is still unsatisfactory. One issue is with the use of features. Some features are insensitive to the change of health, and some are redundant with each other. These insensitive and redundant features in the data mislead the detection. Another issue is from the influence of operating conditions, where a change in operating conditions can be mistakenly detected as an anomalous state of the system. Operating conditions are usually changing, and they may not be readily identified. They contribute to false positive detection either from non-predictive features driven by operating conditions, or from influencing predictive features. This dissertation contributes to the reduction of false detection by developing methods to select predictive features and use them to span a space for anomaly detection under indeterminate operating conditions. Available feature selection methods fail to provide consistent results when some features are correlated. A method was developed in this dissertation to explore the correlation structure of features and group correlated features into the same clusters. A representative feature from each cluster is selected to form a non-correlated set of features, where an optimized subset of predictive features is selected. After feature selection, the influence of operating conditions through non-predictive variables are removed. To remove the influence on predictive features, a clustering-based anomaly detection method is developed. Observations are collected when the system is healthy, and these observations are grouped into clusters corresponding to the states of operating conditions with automatic estimation of clustering parameters. Anomalies are detected if the test data are not members of the clusters. Correct partitioning of clusters is an open challenge due to the lack of research on the clustering of the machinery health monitoring data. This dissertation uses unimodality of the data as a criterion for clustering validation, and a unimodality-based clustering method is developed. Methods of this dissertation were evaluated by simulated data, benchmark data, experimental study and field data. These methods provide consistent results and outperform representatives of available methods. Although the focus of this dissertation is on the application of machinery systems, the methods developed in this dissertation can be adapted for other application scenarios for anomaly detection, feature selection, and clustering.Item Anomaly Detection in Time Series: Theoretical and Practical Improvements for Disease Outbreak Detection(2009) Lotze, Thomas Harvey; Shmueli, Galit; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The automatic collection and increasing availability of health data provides a new opportunity for techniques to monitor this information. By monitoring pre-diagnostic data sources, such as over-the-counter cough medicine sales or emergency room chief complaints of cough, there exists the potential to detect disease outbreaks earlier than traditional laboratory disease confirmation results. This research is particularly important for a modern, highly-connected society, where the onset of disease outbreak can be swift and deadly, whether caused by a naturally occurring global pandemic such as swine flu or a targeted act of bioterrorism. In this dissertation, we first describe the problem and current state of research in disease outbreak detection, then provide four main additions to the field. First, we formalize a framework for analyzing health series data and detecting anomalies: using forecasting methods to predict the next day's value, subtracting the forecast to create residuals, and finally using detection algorithms on the residuals. The formalized framework indicates the link between the forecast accuracy of the forecast method and the performance of the detector, and can be used to quantify and analyze the performance of a variety of heuristic methods. Second, we describe improvements for the forecasting of health data series. The application of weather as a predictor, cross-series covariates, and ensemble forecasting each provide improvements to forecasting health data. Third, we describe improvements for detection. This includes the use of multivariate statistics for anomaly detection and additional day-of-week preprocessing to aid detection. Most significantly, we also provide a new method, based on the CuScore, for optimizing detection when the impact of the disease outbreak is known. This method can provide an optimal detector for rapid detection, or for probability of detection within a certain timeframe. Finally, we describe a method for improved comparison of detection methods. We provide tools to evaluate how well a simulated data set captures the characteristics of the authentic series and time-lag heatmaps, a new way of visualizing daily detection rates or displaying the comparison between two methods in a more informative way.Item A Performance Characterization of Kernel-Based Algorithms for Anomaly Detection in Hyperspectral Imagery(2007-04-25) Goldberg, Hirsh Reid; Chellappa, Rama; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This thesis provides a performance comparison of linear and nonlinear subspace-based anomaly detection algorithms. Using a dual-window technique to separate the local background into inner- and outer-window regions, pixel spectra from each region are projected onto subspaces defined by projection vectors that are generated using three common pattern classification techniques; the detection performances of these algorithms are then compared with the Reed-Xiaoli anomaly detector. Nonlinear methods are derived from each of the linear methods using a kernelization process that involves nonlinearly mapping the data into a high-dimensional feature space and replacing all dot products with a kernel function using the kernel-trick. A projection separation statistic determines how anomalous each pixel is. These algorithms are implemented on five hyperspectral images and performance comparisons are made using receiver operating characteristic (ROC) curves. Results indicate that detection performance is data dependent but that the nonlinear methods generally outperform their corresponding linear algorithms.