Event detection in sensor networks for modern oil fields

Abstract

We report the experience of implementing event detection analytics to monitor and forewarn oil production failures in modern, digitized oil fields. Modern oil fields are equipped with thousands of sensors and gauges to measure various physical and chemical characteristics of oil and gas from underground reservoirs to distribution systems. Data from these massive sensor networks weave a picture depicting the state of oil production and potentially hinting at troubles ahead. Continuous streams of sensor readings can be tapped and fed into analytical algorithms in real time to estimate the likelihood of failure events and generate alerts for possible engineering actions. However, the large amount of main memory required to maintain algorithmic states on cumulative stream data poses challenges to today's webcentric, short-message oriented IT infrastructure. Familiar techniques such as data aggregation, selective sampling and window truncating cannot be applied to some sophisticated algorithms. The paper details our end-to-end solution, points out mismatches with the prevalent transactional web model and suggests new research directions. Copyright 2008 ACM.