Uncertainty Quantification in Operational Modal Analysis and Continuous Monitoring of Special Structures

This thesis addresses three key issues of a real-life vibration-based structural health monitoring system. The first is related to the estimation of the modal parameters of the monitored structures from output-only data together with their confidence intervals. Since the source of vibration of the monitored structures are mostly the unmeasurable ambient excitations, all estimates from the output responses are contaminated with disturbances of statistical nature which are, in turn, disseminated to the identified modal parameters. Hence the need to consider not only the modal parameter estimates, but also their uncertainties in damage assessment. Therefore, apart from discussing the strategies and techniques employed to automatically track the dynamic properties of the monitored structures, the techniques used to estimate the confidence bounds are also addressed and two approaches are proposed to estimate these uncertainties in the present work. The second key issue involves the automation of the modal parameter estimation. In fact, a successful assessment of the health condition based on modal properties is only feasible if these parameters are automatically extracted from the vibration raw data acquired over the course of a continuous monitoring. Given the huge amount of datasets acquired over time, such task is required to be performed by automated applications which are capable of tracking, amongst other useful information, the modal parameters from these data. Once they are initially configured, it is expected that such applications are capable of extracting this information with no further intervention. Finally, the third key issue concerns the detection of damage under varying environmental conditions. In real-life applications structures are subjected to changes in such conditions (e.g., temperature, humidity, wind, traffic, etc.). Therefore, if the modal parameter estimates are intended to be used as damage indicators, the variations induced by these conditions must be taken into account, otherwise they may mask the changes caused by structural damage. If these variations are not accounted, false-positive or negative damage diagnosis may occur and, therefore, vibration-based health monitoring becomes inefficient. In these conditions, environmental models can be applied to such properties, so that they can be used to diagnose damage. In order to discuss the application of such models from a practical point of view, a thorough analysis of data from a continuous monitoring of a football stadium suspension roof is presented. The result of this analysis indicates that a slight structural change has occurred in the roof structure.