Predictive monitoring of heritage structures machine learning and time-series models for displacement forecasting

This thesis looks into how data-driven forecasting methods might be able to be used to study how heritage masonry structures change shape when the weather changes. The research examines the correlation between temperature fluctuations and structural displacement utilizing a two-stage forecasting framework. First, we use ARIMA and Ridge Regression models to guess what the temperature will be like each day. The predicted temperature values are then used as input variables to figure out structural displacement indicators that a monitoring system measures. We used the methodology on a monitored historic tower in Bologna, Italy, and tested the model's stability and forecasting ability with different training scenarios. The findings indicate that Ridge Regression yielded more dependable temperature forecasts compared to the ARIMA model. The regression model effectively captured the overall trend of the displacement series, especially in the horizontal direction, when utilized for displacement forecasting, despite persistent discrepancies between predicted and observed values. The results show that temperature is a major factor that affects how masonry towers move during different seasons. The results show, however, that temperature alone is not enough to fully explain how the structure reacts. The study emphasizes the advantages and constraints of data-driven forecasting methodologies for heritage structural monitoring.