Title: A Pattern-Based Method for Defective Sensors Detection in an Instrumented Bridge

Author(s): M.S. Islam, A. Bagchi and A. M. Said

Publication: Symposium Paper

Volume: 298

Issue:

Appears on pages(s): 1-18

Keywords: auto regressive xeogenous, binary search, damage detection, filter, sensor, sequential search, simulated data, pattern matching

DOI: 10.14359/51687088

Date: 6/5/2014

Abstract:
The most advanced method of investigating the performance of a structure is to continuously track the strain, deflection, and acceleration by analysing data collected from a series of wireless sensors installed on the structural member. Before analysing the data, it is important to assure the reliability of the data by verifying that all sensors are working properly. For an instance, in the reinforced concrete structure sensors are attached to the reinforcement bars and might be destroyed while pouring the concrete. Besides, sensors might malfunction due to excessive variation of temperature, load, or any other causes. Data-driven and structural models-based are two damage detection techniques in civil structures. In this study, the data driven method, a direct approach to damage assessment, was followed; this approach does not require structural modeling, such as finite element analysis. In this method, the existence of damage and its location are interpreted by pattern matching of the data series at different time ranges. The objective of this study was to develop new techniques to detect defective sensors based on the pattern matching method that included the Auto Regression Xeogeneous model. As a case study, Portage Creek Bridge was selected, located in British Colombia, Canada. Data sets of strain and temperature gages were downloaded from a database connected to the instrumented pier of the bridge and filtered and normalized continuously. The condition of a set of sensors installed in the pier was determined, using a method developed based on the concept of the sequential and binary search techniques. Using sensitivity analyses of the developed models, defective sensors were detected by pattern matching of simulated and measured or real data.