This study proposes a novel data driven structural health monitoring (SHM) algorithm for detection and localization of damage in plates and frame-type structures. The algorithm employs Ising graphical model for considering the dependencies of the structural response at the neighboring sensor locations rather than a sensor level data analysis. The proposed statistical model also provides a measure of uncertainty that allows soft decision making instead of binary classification. The efficacy of the algorithm is experimentally validated by testing a three-story two-bay laboratory steel prototype. The results show that the consideration of spatial dependencies significantly improves the detection/localization accuracy.
doi: 10.12783/SHM2015/152