Fault Detection in Embedded Networked Sensing

• Activity Number: 317
• Type: Contributed
• Date/Time: Tuesday, August 5, 2008 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Sheela Nair*+ and Mark H. Hansen
• Companies: University of California, Los Angeles and University of California, Los Angeles
• Keywords: sensor networks ; data quality ; fault detection ; spatial temporal modeling

Abstract

Recent advances in sensor technology, computing, and low-power communication have facilitated the development of embedded networked sensing (ENS). A major issue limiting the use of ENS is quality of sensor data, which can be prohibitively faulty. Fault detection systems for ENS face unique statistical and computing challenges. Data collected by ENS are often highly complex and non-stationary. Algorithms that are both computationally feasible and scalable must be designed so that data quality can be assessed in near-real time. We discuss the use of signatures for fault detection, borrowing on similar ideas used in fraud detection. The proposed algorithm adaptively estimates and tracks sensor signatures as data are collected. We also outline current work in spatial-temporal modeling of ENS data. Environmental ENS deployments will be used to illustrate the problems and methodology.