Multilevel Decision Fusion in a Distributed Active Sensor Network for Structural Damage Detection

Publisher Website
Google Scholar
Abstract
Distributed sensor networks are emerging as a critical technical driver in the application of structural health monitoring for large-scale structures as a result of their excellent abilities to enhance the reliability and robustness of monitoring systems. One of the key technical opportunities in the implementation of a distributed sensor network is the application of information fusion. Not only does this enable the integration of data from all sensors for the comprehensive assessment of structural conditions, but it also facilitates the combination of decisions or perceptions from multiple sources or different approaches. In this article, the feasibility of combining a distributed sensor network and several techniques of multilevel decision fusion are demonstrated for damage detection. The level-one decision fusion is first implemented by individual active sensors to create their own perceptions on structural health status. During the fusion process, the active sensors first interrogated local physical sensor nodes in the network, and then combined the perceptions of the local sensors in terms of correlation between features extracted from raw signals and damage scenarios in a knowledge database. Meanwhile, the perceptions of active sensors on structural health status are integrated to represent the decision fusion at level two. Considering the level-two decisions made from different technical approaches, a further combination of all decisions is conducted with implication of information fusion at a higher decision level. As a consequence, the risks of a malfunction of individual sensors or the inappropriateness of individual assessment procedures are significantly reduced, and a robust and error-tolerant structural health monitoring system can be developed. Such an approach is successfully validated by an experimental case study of damage detection in CF/EP composite structures.