Real-Time Fault Detection and Isolation and Offline Data Analysis with Distributed Processing and State Space Compression

• Activity Number: 348
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Spencer Graves*+
• Companies: Structure Inspection and Monitoring, Inc.
• Keywords: Data compression ; structural health monitoring ; Bayesian sequential updating ; Distributed filtering ; critical infrastructure ; environmental protection

Abstract

The quantity of data that could be collected in any real-time monitoring application is limited by the costs of data collection, transmission and storage. Smart sensors allow computations to be done anywhere, and analog signals can theoretically be digitized to any number of bits with any sampling frequency. The information obtained for a given budget can be maximized by considering what features of the data are most informative, and where those features can be most cheaply encoded. The high order bits in a digital representation almost never change, and the lowest order bits may be noise. Moreover, only change is informative. To determine the number of bits to retain, we consider "observation = important + unimportant + noise". We model this decomposition with Bayesian state space structures, reporting a new state only when predictions from the last report are not adequate. To support off-line analysis for model improvement, we also report outliers and samples of other observations. This creates a need to develop new methods for analyzing data in this compressed format.