Application of a classification algorithm to the early-stage damage detection of a masonry arch

The early-stage identification of structural damage still represents a relevant challenge in civil engineering. Localized damages if not readily detected can lead to disruption or even collapse, involving hazard to people and economical losses. Although the final goal of the identification is to localize and quantify the damage, a reliable discrimination between normal and abnormal states of the structure in the very early stage of the damage onset is not an easy task. In the field of Structural Health Monitoring (SHM) great attention has been paid to the development of damage detection methods based on continuous and automatic registration of the system response to unknown ambient inputs. The numerical algorithms exploited must be: (1) easy to implement and computationally inexpensive, eventually being embedded in the sensors; (2) as much independent on human decision as possible; (3) robust to the many sources of uncertainties affecting the monitoring; (4) able to detect small damage extents in order to provide an early warning; (5) suitable for the application in the case of few and sparse measurements collected only in the normal condition. The performance of a novel version of Negative Selection Algorithm, recently developed by the authors, is here analyzed with attention to these issues. The algorithm is tested against data collected on a segmental masonry arch built in the laboratory of the University of Minho and subject to progressive lateral displacement of one support.