文摘
In this dissertation,long-term measurement data that is being collected from the Jeremiah Morrow Bridge will be used to quantify annual variation in data and establish boundaries for detecting abnormal behaviors including anomalies from univariate trends or multivariate correlational trajectories. Long-term measurement data from the US Grant Bridge will also be used for calibrating an autoregressive integrated moving average model and distinguishing maintenance events. First,the monitoring system that has been used for the two bridges under evaluation will be overviewed. Second,sensory data will be analyzed as a univariate time series and transformed to a simple regression model using temperature data as exogenous inputs. Third,correlation between temperature and sensory data will be analyzed and abnormal changes or outliers within the bivariate time series will be identified. We will try to identify how temperature trends change over time and use the dynamic trends to probabilistically classify temperature-caused events. Fourth,load responses of a bridge will be used to define load signatures; whenever a lane load exists on a bridge e.g.,halted or slowed traffic),the sensory network responds in a certain way that can be quantified by correlation of measured values. Using the identified signature,we should be able to distinguish lane loads from thermal responses. Finally,combining univariate time series outlier detection,variable correlational coefficients Principle components),extreme thermal response signature,and the load response signature,an integrated monitoring system will be proposed and the results will be compared with previously implemented systems by UCII for these structures.