A Statistical Health-monitoring Framework for Transportation Infrastructure.

摘要

As transportation systems in the U.S. mature, significant portions of its infrastructure are nearing the end of their service lives and need to be replaced. This is especially true for bridges which attract the greatest attention among all types of infrastructure. Structural Health Monitoring (SHM) is a direct and deliberate response to address this issue which involves the detection and characterization of damage in civil structures using advanced sensing technologies as well as reliable analytical methods to process the ensuing data.;This dissertation presents a rigorous and generally-applicable statistical framework to support long-term SHM of transportation infrastructure. The first part of the framework integrates regression analysis with sophisticated time series models to explain common-cause variations in the data that can be attributed to usual operating conditions. The ensuing innovation series are analyzed in the second part of the framework, which consists of using Shewhart and complementary control charts to detect special-cause or unusual events. As multiple measurements are monitored simultaneously, Multivariate Statistical Process Control methods are employed to capture the underlying covariance as well as to ensure rigorous control of Type I error rate. Such methods include the Hotelling's T2 control chart followed by the Mason-Young-Tracy decomposition technique to interpret the results.;This dissertation further exploits Structural Time Series Models as an alternative statistical approach suited to analyze measurement errors in SHM. The models decompose deterioration properties of a facility into meaningful components such as linear trend and seasonality and formulate them in a state-space form where true condition of each component is treated as a latent variable. Kalman filter is implemented to estimate the latent variables and the ensuing innovation series are analyzed to distinguish measurement errors from randomness inherent in the deterioration process.;Models described in this dissertation are validated through empirical studies on the SHM data acquired from a highway bridge in Hurley, Wisconsin where concerns are raised regarding premature structural deterioration caused by loads associated with logging trucks. Special-cause events detected by the framework are reported and plausible link is presented between several notable events with excessive truck loads as well as extreme weather changes.