The use of subspace-based methods for damage detection in civili structures: a machine learning approach

Events in recent years have shown how important it is to monitor the structural health of existing civil works. Structural Health Monitoring systems are a useful tool to provide an objective and automatic valuation of the state of health of a structure, in order to detect the emergence of anomalies in its behavior. They are also an auxiliary tool in the decision-making phase for maintenance work or after extraordinary events. The Thesis work explores the topic of damage detection based on the analysis of subspaces of dynamical systems matrices. The aim of the research was to investigate a method for the development of levels in the damage diagnosis scale, ranging from the identification of the anomaly to the localization and subsequent assessment of the entity of the damage occurred. The study is carried out following two approaches: in the first, damage indices present in the literature are considered, and a newly developed one is presented. In the second approach, the identification issue is addressed by introducing the most recent Machine Learning tools: the goals are achieved through the supervised training of a classifier, with the task of localizing and quantifying the damage. In both cases, the methods used are model-driven type, based on simulations of the damage scenarios through Finite Element modeling. The thesis work therefore aimed to evaluate the effectiveness of the same subspace- based indices as objective functions in the optimization process related to the model updating process of the FE model, such that the simulated response in subsequent analyses would be as close as possible to the real one. The findings of the tests, both numerical and experimental, confirm the effectiveness of both proposed methods, highlighting their shortcomings and strengths. The concepts developed were subsequently applied to a case study, represented by the Basilica of Santa Maria di Collemaggio, in L'Aquila. In the first stage, the dynamic behavior of the Basilica was investigated, over the years of monitoring. Subsequently, traditional and Machine Learning algorithms have been implemented for the purpose of anomaly detection: the procedure has been performed considering as damaged a case, one produced after a structural intervention subsequent to the installation of the monitoring system. The studies on the Basilica showed a complex dynamic behavior, strongly influenced by environmental factors: nevertheless, the implemented algorithms proved to be effective for the defined purpose.