Local validation of structural health monitoring strain measurements

Abstract

A Structural Health Monitoring (SHM) local strain sensor validation scenario is considered in this paper. The statistical control of the sensor output data is a well established and commonly used technique. However, the assumption that the same probability distribution will continue to represent the observed variable as the process goes on without faults cannot be met for non-stationary processes. Such is the case for many SHM strain measurement time series, which have a non-stationary nature. To enable the local validation of those time series, new methods for the empirical segmentation of the strain time series into two periods (nighttime and daytime), and for the temporal normalization of the daytime time series are proposed. The segmentation of the normalized non-stationary time series—through the application of a heuristic segmentation algorithm—and a new method for statistical control of the non-stationary strain sensor trend output data, are proposed and validated. The proposed local validation method is validated on real suspension bridge SHM strain data. The results show that the statistical control of the daytime period allowed the detection of faults that otherwise would not be detected in the nighttime period. The proposed methods allowed the local validation of the non-stationary strain time series.