Knowledge Transfer Between Dynamically Monitored Masonry Bell Towers

Structural health monitoring (SHM) is really important in cultural heritage (CH) structures, where invasive interventions are often not allowed. An important challenge in data-driven approaches is the lack of data, especially corresponding to damage states, which hampers the development of pattern recognition algorithms in the field of Machine Learning (ML) models. For CH structures, obtaining a complete dataset of health states—including both undamaged and damaged conditions—is often impossible, particularly for damaged states, as it would require harming the structure in situ. Domain adaptation (DA), a sub-sector of Transfer Learning (TL) methods, addresses this problem by adapting data from a more accessible and monitored system (source) transferring information by leveraging data domains to systems with more limited data (target) but with similar properties. In this work, DA techniques are used to expand information on the health state of the special class of monitored structures represented by CH masonry bell towers. In more detail, continuous-time dynamic monitoring data are obtained from a system installed on the masonry bell tower (source) of the Church of S. Maria and S. Giovenale (Fossano, Piedmont, Italy). This tower is currently monitored because, following the evolution of vertical cracking phenomena, in 2012 it was subjected to reinforcement interventions, through the installation of 11 steel tie rods on its four external sides. Another damaged masonry bell tower, belonging to the ancient parish Church of S. Antonio Abate (Montà, Piedmont, Italy), is instead assumed as target structure, with limited knowledge gained from a single testing campaign.