Structural identification through continuous monitoring: data cleansing using temperature variations

The aim of structural performance monitoring is to infer the state of a structure from measurements and thereby support decisions related to structural management. Complex structures may be equipped with hundreds of sensors that measure quantities such as temperature, acceleration and strain. However, meaningful interpretation of data collected from continuous monitoring remains a challenge. MPCA (Moving principal component analysis) is a model-free data interpretation method which compares characteristics of a moving window of measurements against those derived from a reference period. This paper explores a data cleansing approach to improve the performance of MPCA. The approach uses a smoothing procedure or a low-pass filter (moving average) to exclude the effects of seasonal temperature variations. Consequently MPCA can use a smaller moving window and therefore detect anomalies more rapidly. Measurements from a numerical model and a prestressed beam are used to illustrate the approach. Results show that removal of seasonal temperature effects can improve the performance of MPCA. However, improvement may not be significant and there remains a trade off when choosing the window size. A small window increases the risk of false-positives while a large window increases the time to detect damage.