Bio-inspired algorithms for structural health monitoring of civil engineering systems

Nowadays, developed countries are challenged by the management of a wide estate inventory of complex existing structures and infrastructures, which are either close or beyond the end of their service life. Maintenance and prevention have become a significant item of expenditure, while cost-effective strategies are required but still under development. Structural Health Monitoring (SHM) aims at the prompt identification of damage in order to allow an automated health condition assessment of structural systems. The development of such a field of investigation shall provide suitable and reliable methods for detecting the damage outbreak at the earliest possible stage, thus for facing it in a quick, focused and economic way. To this end, damage detection can be formulated as a one-class classification problem and effectively addressed through bio-inspired numerical tools, as the Negative Selection Algorithm (NSA). This thesis aims at developing and testing a damage detection methodology based on an innovative version of NSA with deterministic generation. The methodology is composed of several numerical features to tackle the relevant shortcomings that emerged during the literature review and the pilot tests. The individual features and the global methodology are validated on numerical instances and field-testing case studies, considering multiple and incre sing damage scenarios and varying environmental and operational conditions. All the conclusions drawn in the present work are based on experimental analyses of the algorithms, performed based on a proper statistical design. Additional attention is paid to provide a fair comparison with alternative existing techniques. The proposed methodology results suitable for early-stage damage detection. It can be adapted to different types of structures and damage-sensitive features. It might be suitable for sensor embedment, by performing the detection on the acquisition of a single sensor. It is independent of the type of monitoring tools or excitation. It is robust against sources of uncertainties as the noise in the signals, the error induced by feature extraction and the fluctuation in the monitored features due to varying environmental conditions. Its performance is, instead, largely affected by the algorithm parameter setting. Therefore, different suitable setting designs are presented together with recommended values or ranges. In conclusion, the damage detection strategy based on NSA, that is validated in the context of the present thesis, is deemed effective and the promising results foster more research and further applications.