Development of a damage detection technique with a strong immunity to environmental influence implemented on a laboratory truss girder subjected to ambient variations

In recent years, Structural Health Monitoring (SHM) has become a fundamental branch of industrial, aerospace and civil engineering, offering a various array of techniques designed to predict, assess, and manage potential structural damage. Most of the Structural Health Monitoring (SHM) strategies rely on examining the damage-sensitive features of the system under analysis and conducting ongoing monitoring, evaluating the evolution in time of these characteristics. Several methods have been developed in recent years, some of the most promising ones are based on vibration analysis. This PhD thesis focuses on the development of an unsupervised learning approach for detecting structural damage based on vibrations. The benchmark of the work involves a laboratory truss girder designed ad-hoc for the development of damage detection techniques based on acceleration measurements. The approach is based on statistical pattern recognition and is focused on two key steps: the selection of damage-sensitive features and the automation of the outlier detection process. Particular consideration is given to the practical application of this approach, involving sparse sensor layout, handling long-term monitoring data, and suggesting synthetic damage indicators to aid in the maintenance decision-making process. The validation of the algorithm is performed on real data computed from the laboratory truss girder acceleration measurements. The laboratory structure has been monitored for several months under uncontrolled environmental factors such as ambient temperature drift and sunlight exposure evolution.