Health Monitoring of Structures Using Statistical Pattern Recognition Techniques

The primary objective of structural health monitoring (SHM) is to determine whether a structure is performing as expected or if there is any anomaly in its behavior compared with the normal condition. It is also useful in detecting the existence, location, and severity of damage. Vibration-based damage detection methods are very frequently used in SHM. However, because of complicated features of real-life structures, there are uncertainties involved in the key input parameters (e.g., measured frequencies and mode shape data), which affect the performance of these methods. If vibration-based methods are incorporated with semianalytical methods, such as statistical pattern recognition techniques, better accuracy can result in structural health assessment. This paper explores the statistical pattern recognition techniques for damage detection and/or degradation in structures. A case study, the Portage Creek Bridge in Victoria, British Columbia, Canada, has been used. The following two approaches of the statistical pattern recognition techniques have been used: statistical pattern comparison and statistical model development. After filtering and normalizing the data obtained from the SHM system installed in the bridge, damage sensitive features have been extracted by autoregressive modeling of the time series data. Both idle and excited states of the bridge are considered in this case. From the statistical analysis of the strain and acceleration data, although the bridge is in a good condition, there is a small but steady deterioration in its performance. The study also demonstrates the feasibility of the statistical pattern recognition techniques in assessing the structural condition of a practical structure.