Structural Damage Detection Through Vibrational Feature Analysis with Missing Data

Abstract

To mitigate the expensive and time-consuming nature of visual structural inspections, vibration-based structural damage detection methods have been proposed that rely on different damage-sensitive features. These features are derived from data collected by sensor networks implemented on the structure. Damage detection through vibration-based feature analysis thus far has relied on simulation or the responses of fixed sensor networks for feature creation. Another type of monitoring scheme, called mobile sensing, has the ability to eliminate the limited spatial information constraint of fixed sensor networks. However, regardless of the monitoring approach of a real-world structure, data sets can incur cases of missing data, either due to such situations like sensor malfunction or loss of communication connectivity, or in the case of mobile sensing due to the nature of the approach itself. In this paper, a fixed sensor network is implemented on a scale laboratory frame structure, and observations are removed from the resulting complete datasets to simulate data missingness. Damage is simulated through interchangeable, variable stiffness elements that make up the frame. Damage detection is conducted by fitting a numerical model to the data and assessing the significance of the change in the model parameters when damage is introduced.