Welcome to the new version of CaltechAUTHORS. Login is currently restricted to library staff. If you notice any issues, please email coda@library.caltech.edu
Published April 15, 2011 | Published
Book Section - Chapter Open

Robust Diagnostics for Bayesian Compressive Sensing with Applications to Structural Health Monitoring

Abstract

In structural health monitoring (SHM) systems for civil structures, signal compression is often important to reduce the cost of data transfer and storage because of the large volumes of data generated from the monitoring system. Compressive sensing is a novel data compressing method whereby one does not measure the entire signal directly but rather a set of related ("projected") measurements. The length of the required compressive-sensing measurements is typically much smaller than the original signal, therefore increasing the efficiency of data transfer and storage. Recently, a Bayesian formalism has also been employed for optimal compressive sensing, which adopts the ideas in the relevance vector machine (RVM) as a decompression tool, such as the automatic relevance determination prior (ARD). Recently publications illustrate the benefits of using the Bayesian compressive sensing (BCS) method. However, none of these publications have investigated the robustness of the BCS method. We show that the usual RVM optimization algorithm lacks robustness when the number of measurements is a lot less than the length of the signals because it can produce sub-optimal signal representations; as a result, BCS is not robust when high compression efficiency is required. This induces a tradeoff between efficiently compressing data and accurately decompressing it. Based on a study of the robustness of the BCS method, diagnostic tools are proposed to investigate whether the compressed representation of the signal is optimal. With reliable diagnostics, the performance of the BCS method can be monitored effectively. The numerical results show that it is a powerful tool to examine the correctness of reconstruction results without knowing the original signal.

Additional Information

© 2011 Society of Photo-Optical Instrumentation Engineers (SPIE). One of the authors (Yong Huang) acknowledges the support provided by the China Scholarship Council while he was a Visiting Student Researcher at the California Institute of Technology. This research is also supported by grants from National Natural Science Foundation of China (NSFC grant nos. 50538020, 50278029 and 50525823), which supported the first and third authors (Yong Huang and Hui Li).

Attached Files

Published - 79820J_1.pdf

Files

79820J_1.pdf
Files (381.4 kB)
Name Size Download all
md5:c46e87d278851346277a941dbed5ceff
381.4 kB Preview Download

Additional details

Created:
August 19, 2023
Modified:
January 13, 2024