Comparison on outlier detection methods using measured data from a long span tied-arch bridge

Abstract

As the demand for the detections of outliers in the structural health monitoring data-set increases, numerous approaches are presented for it. However, the characteristics of the existing methods dealing with different kinds of measured data are not yet clear enough for practical use. Therefore, this paper conducts a comparative study of several popular rule-based methods based on monitoring data of an arch-tied bridge in China. For measured data, outliers are not known in advance. In this way, this study evaluates and compares the detection performances rely on two indicators: the quantity of the detected outliers and the extreme value of the outliers deviating from the mean of the data. Conclusions on the features and applicable situations of involved methods are given. Additionally, combining the results of different methods proves to be beneficial. Finally, a software incorporating the research results is developed for outlier detection.