Damage sensitive features. From classic parameters to new indicators

Structural Health Monitoring (SHM) is the discipline that concerns about the health condition of engineering structures, mechanical systems, aerospace models, at every moment during their utility life. The primary object of SHM is to spot damage, if present, in the observed system and give a consequent diagnosis. Damage can be considered as a variation in the properties of the system that permanently affects the performance of the structure. This variation is meaningless unless contextualized as a comparison between two different states: damaged and healthy. It is precisely this deviation from normal conditions that approaches like vibration-based algorithms are looking for. Vibration-based SHM aims to implement a strategy to correctly detect damage through the assessment of changes in the identified vibration response of civil structures. The structural response is represented, employing a compact representation of its primary traits, called damage sensitive features (DSFs). It can be stated, therefore, that the effectiveness of vibration-based methodology in identifying damage depends on the robustness of the chosen DSFs. They need to be sensitive enough in spotting anomalies in the structural behavior, but at the same time, they need to be insensitive as much as possible towards temporary or seasonally variation of the structural properties that fall into the common behavior of structural systems. In this dissertation, two typologies of DSFs are investigated: the first type, well known in the SHM research community, is derived from the response of the system using user's dependent extraction algorithms, while the other is directly computed from the response of the system using digital signal processes alone. In both approaches, the effects of external conditions, like the seasonal variation of air temperature, are accounted for. Within the first kind of DSFs, an automated procedure is proposed to reduce the interdependency of the algorithm from the user's abilities, leading to a robust identification procedure, more suitable for long-term monitoring purposes. The second health indicator here proposed offers a very low-burden computation cost, with almost non-existing dependency from the user and its extraction process makes these features less sensible to external variation like temperature. The two DSFs and the associated extraction processes are investigated and validated both numerically and experimentally.