Study, qualification and application of AR techniques for SHM

Structural Health Monitoring (SHM) is any process to conduct an automatic damage detection strategy, usually in the context of aerospace, civil and mechanical structures. A focus of this thesis is to perform a qualification of the Autoregressive (AR) models in the field of the unsupervising learning and where only the output of a system provided by a single sensor is available. This is the worst scenario in terms of the amount of information available on the system taken into consideration and this is chosen to conduct an analysis of these techniques, investigating their limits of application. The first part of the thesis deals with the variance analysis of the AR parameters used to represent the states of simulated mechanical systems, to quantify the performances of the AR parameters in terms of sensitivity to a change of the represented system. The qualification of the AR parameters in this context is a lack of the literature and some unexpected results will be found. These results justify the use of the Principal Component Analysis (PCA) on the AR parameters describing different scenarios of the system in order to enhance the damage identification. In the second part of the thesis, this procedure, AR modelling coupled with PCA, is evaluated on a test bench structure in relation to Environmental and Operational variations (EOVs) effects. The experimental validation considers a single sensor and actual temperature variations, which is very rare in the literature. Since, this kind of scenario is usually performed representing the effects of temperature like a stiffness reduction of the bench test. Cointegration is applied to the same test bench by using the information provided by a single sensor. The purpose is a comparison between the use of PCA and another technique which is specifically developed to deal with environmental and operational effects. Finally, the use of autoregressive models in combination with PCA is applied to a real operative structure, the Meazza stadium based in Milan. The results show the effectiveness of the structural health monitoring procedure studied in this thesis. Different scenarios of the structure can be clearly identified and the systematic trends on the data during the year prove the ability of the methodology to follow the variation of the system due to environmental and operational conditions.