Identification of changes in the dynamics of a reinforced concrete building through a machine learning approach for data normalization

In recent years there has been a growing number of Structural Health Monitoring (SHM) applications based on environmental vibration measurements and Operational Modal Analysis techniques. It is nowadays widely recognised in the scientific literature the need of normalizing data (e.g. modal properties) of vibration-based monitoring systems from variations related to the effects of environmental conditions (e.g. temperature, wind speed, intensity of human activities). In this paper, the results of the dynamic monitoring of a 10-storey reinforced concrete building in central Italy are presented. The building, which is monitored since 2017, hosts the Faculty of Engineering of the Università Politecnica delle Marche and can be considered of strategic importance. During 2021, retrofit works on the structural joints separating the building from adjacent bodies were carried out. This led to a slight modification of its dynamic behaviour; in detail, variations of the modal properties due to the interventions on joints hide beyond those due to environmental conditions and cannot be clearly detected with traditional multivariate statistics techniques for data cleansing. On the contrary, a data cleansing procedure based on the application of an artificial neural network revealed effective to detect the variation of the building dynamic behaviour.