Local Frequency-Based Estimators for Anomaly Detection in Oil and Gas Applications

• Activity Number: 311
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Lakshminarayan K. Choudur*+ and Evan Kriminger and AlexanderSingh Alvarado
• Companies: Hewlett Packard Laboratories and University of Florida at Gainesville and University of Florida at Gainesville
• Keywords: Fourier analysis ; Dynamical systems ; turbulence ; Oscillations ; Non-linear time series ; frequency methods

Abstract

Modern industrial applications such as the smart grid and oil and gas are continuously monitored. The massive amounts of data collected is then processed, and analyzed to generate actions to ensure smooth operations to positively impact the bottom line. In the oil and gas industry, modern oil rigs are outfitted with thousands of sensors to measure the flow rates, as well as the physical and chemical characteristics that affect production from underground off-shore and on-shore reservoirs. Analytical methods packaged into a surveillance system and applied to the massive network of sensors track the state of the system and issue warning alerts about impending failures. In this setting, real time algorithms are needed to detect a diversity of event types, such as anomalies, trends or forewarn failure events to generate alerts for proactive engineering actions. In this paper, online algorith