Structural health assessment of bridges by long-term vibration monitoring and automated operational modal analysis

The goal of Vibration based structural health monitoring (VBSHM) applications is reliable and consistent non-destructive condition assessments of civil engineering structures. The main objective of this thesis is to contribute to, and to expand on, the current knowledge of VBSHM, with a focus on both the practical aspects such as instrumentation, data collection, data management and large scale data processing, and on the theoretical aspects such as advanced analysis and interpretation through improved system identification, automated operational modal analysis and long-term tracking of modal estimates which will lead to proper condition assessment. The Confederation Bridge's long-term remote vibration monitoring project in eastern Canada provides an important backdrop for the work described in this thesis. The path to reliable and consistent condition assessments from vibration response measurements is through a thorough understanding of the causes of uncertainties and variability in the analysis results. The well-established Stochastic Subspace Identification (SSI) technique is improved and automated to reduce the uncertainty associated with computation and human error. A new Automated Inline Full Space Identification (AI-FSI) technique which integrates all aspects of automated modal parameter estimations (MPE) and modal tracking is presented in this thesis. With the new tools integrated in the third version of the signal processing platform for analysis of structural health (SPLASH), the processing and analysis of all the historical data collected by the Confederation Bridge monitoring project since 1998 was completed. This represents over 250000 logger files, 40000 hours of recording and 28TB of raw and processed data collected over 20 years. Several possible sources of environmental and operational variability are identified and quantified. Through multiple regression analysis it has been shown that these identified sources of variability can explain between 24% and 53% of the variations observed in the estimated modal frequencies (depending on the mode). This is a significant finding that can further improve damage identification techniques that use modal features. A novel approach using a change point algorithm to detect mean shifts in the residual frequencies attributable to possible damage is successful in identifying small shifts in frequencies (from 0.68% to 0.95% of mean frequency).