Automated outlier detection framework for identifying damage states in multi-girder steel bridges using long-term wireless monitoring data

Abstract

Advances in wireless sensor technology have enabled low cost and extremely scalable sensing platforms prompting high density sensor installations. High density long-term monitoring generates a wealth of sensor data demanding an efficient means of data storage and data processing for information extraction that is pertinent to the decision making of bridge owners. This paper reports on decision making inferences drawn from automated data processing of long-term highway bridge data. The Telegraph Road Bridge (TRB) demonstration testbed for sensor technology innovation and data processing tool development has been instrumented with a long-term wireless structural monitoring system that has been in operation since September 2011. The monitoring system has been designed to specifically address stated concerns by the Michigan Department of Transportation regarding pin and hanger steel girder bridges. The sensing strategy consists of strain, acceleration and temperature sensors deployed in a manner to track specific damage modalities common to multigirder steel concrete composite bridges using link plate assemblies. To efficiently store and process long-term sensor data, the TRB monitoring system operates around the SenStore database system. SenStore combines sensor data with bridge information ( e.g ., material properties, geometry, boundary conditions) and exposes an application programming interface to enable automated data extraction by processing tools. Large long-term data sets are modeled for environmental and operational influence by regression methods. Response processes are defined by statistical parameters extracted from long-term data and used to automate decision support in an outlier detection, or statistical process control, framework.