Vertical and Horizontal Combined Algorithm for Missing Data Imputation in Bridge Health Monitoring System.

Structural health monitoring (SHM) provides strong support for bridge safety evaluation by tracking the real-time structural performance. An accurate and efficient imputation method for missing data in the SHM system is of vital importance for bridge management. In this paper, an innovative vertical–horizontal combined (VHC) algorithm is proposed to estimate the missing SHM data by a more comprehensive consideration of different types of information reflected in different time dimensions of the monitoring data. In the vertical time dimension, the variation trends and changing ranges for the missing data are analyzed by the relational analysis from the global view of data. In the horizontal time dimension, the exact value of the missing data is estimated based on the adjacent data inside the missing data set from the local view of data. The proposed VHC algorithm is validated and compared with the current methods in the data imputation for the missing deflection values in the SHM system of an example tie-arch bridge. The results show that the proposed algorithm demonstrates the best imputation performance among all methods for both missing types with the highest imputation accuracy/similarity and satisfying the 95% confidence bound. The application of reward coefficient significantly increases the imputation accuracy for the continuous missing situation and the improvement grows with the missing step number. Loss ratio and sample size also have significant influences on the imputation performance with the proposed method. Data missing in the structural health monitoring (SHM) system affects the continuity of monitoring data and the structural conditions reflected by them. This paper proposed an innovative data imputation algorithm combining information from both vertical and horizontal time dimensions of the monitored data in the bridge SHM system. The proposed vertical–horizontal combined (VHC) algorithm shows better imputation performance when compared with the current methods widely used in engineering practice and shows a higher applicability in a small sample size situation. Thus the proposed algorithm provides an accurate and efficient tool for dealing with the data missing phenomenon in the SHM system and is of great importance to the safety evaluation and further management decision process for bridge structure. Moreover, the methodologies and approaches proposed in this paper are quite general and can be applied to the missing data imputation problems in many research fields with time series data. [ABSTRACT FROM AUTHOR]