A statistical based damage detection approach for highway bridge structural health monitoring
Upon the request of Iowa DOT, the Bridge Engineering Center (BEC) of Iowa State University developed a fiber optic sensor (FOS) structural health monitoring (SHM) system to monitor the fatigue crack formation for fracture critical bridges in 2006. The system enables bridge owners to remotely monitor bridges for gradual and sudden damage formation. However, the correlation between the data analysis results and damage was not objectively defined; bridge owners need to interpret the results according to their experiences. To improve the existing SHM system, a statistical damage detection method was proposed and analytically evaluated in this work.
The basic idea of this method is that the response of a normal structure is different from that of the damaged structure. To define the difference mathematically, Shewhart control chart analysis was carried out over a strategically defined damage indicator (i.e. residuals calculated from cross prediction models). With different prediction models and different damage indicator calculation procedures, three damage detection methods (one-to-one model direct evaluation method, two-level evaluation method, and many-to-one model method) were studied. As a result, the one-to-one model direct evaluation method was recommended due to the best performance in terms of both damage sensitivity and damage location detection accuracy. For this method, significant efforts were made to select the performance matrix, the load condition, and residual matrix simplification procedure. Results show that using event strain ranges obtained from right-lane, five-axle, heavy trucks along with the combined summation residual matrix simplification approach lead to the best damage detection performance. The optimal group size used in the control chart analysis was also recommended based on the synthetic data verification.
Compared to the previously developed damage detection method, improvements were made by:
1. Developed a strain-based truck detection sub-system, which can detect truck events and calculate relevant parameters including the number of axles, axle spacings, speed, event start and end time, and weight group autonomously in a near-real-time fashion. The sub-system has been integrated into the existing SHM system successfully.
2. The truck detection subsystem allowed for a successful data selection procedure. Using a single truck type during the structural evaluation improved the performance of the system significantly.
3. Using the simplified residual as the damage indicator enabled reliable mathematical control limits selection. The strategically defined damage indicator significantly improved the damage detection power of the method.
4. By using the uniquely one-to-one cross prediction model, the developed damage detection approach can be applied for virtually all types of bridges.
5. The control chart analysis results show the damage occurrence and damage location directly.