Anomaly Detection of Piezometer Data Collected from Embankment Dams
There are more than 85,000 dams in the U.S., the majority of which were built decades ago. It is not surprising then that the number of deficient dams, as qualified by different evaluation methods, has increased in recent years. Dams can pose significant risks to people living around them, and since they are exposed to harsh and largely unpredictable environments, it is important to maintain and inspect the conditions of dams effectively. In the United States (US), the current practice of analyzing the structural integrity of embankment dams relies primarily on manual a posteriori analysis of instrument data by engineers, leaving much room for improvement through the application of automated data analysis techniques. Accurately evaluating measurements collected from instruments used to monitor dam behavior is not an easy task, requiring sound engineering judgment and analysis, as well as robust statistical analysis techniques to prevent misinterpretation. In the research presented in this thesis, different types of anomaly detection techniques are investigated in an effort to i) propose which data analytics are appropriate for various anomalous scenarios as well as piezometer locations, and ii) to test if the widely-held assumptions on piezometer data, i.e., linearity between piezometer data and pool levels, as well as normally distributed piezometer data, are necessary in the anomaly detection task. This thesis specifically focuses on anomaly detection techniques that are applicable in analyzing piezometer data (collected from embankment dams) and anomalous scenarios that may lead to internal erosion. In order to validate how well the anomaly detection techniques perform, various anomalous scenarios and piezometer data of a case study dam are simulated using a numerical model. For each technique, various parameter settings are also examined. In the real world, piezometers (and other instruments) in an embankment dam may not always be located at the optimum places, and their behaviors may not continue consistently over years. Thus, it is important to recognize how the data from a set of piezometers have changed over time as a group rather than evaluating only a single significant piezometer at a time. Since anomalies occurring inside dams may initiate with very small deformations without any visual signs, it is difficult to know exactly when and where the anomalies have initiated and would be located. However, by observing deviations among multiple (or grouped) piezometer readings over time, or the piezometers around a specific location, there is potential for obtaining better interpretations. Thus, this thesis also presents the research work on analyzing multiple piezometers together. The research examines if analyzing multiple piezometers together can improve interpretation and detection of piezometric anomalies, and shows the impacts of analyzing multiple piezometers compared to individual piezometers.