Predictive Modelling of bridge bearing displacements with Physics‐Enhanced Machine Learning (PEML) environmental effects filtering

In bridge Structural Health Monitoring (SHM), identifying anomalies is challenging due to environmental and operational variability (EOV), such as temperature changes, traffic loads, and else. This study develops a predictive model to isolate normal structural responses, enabling the detection of damage-induced anomalies. Using displacement and temperature sensors, the model evaluates longitudinal dis- placements at the bridge bearings. Temperature is the primary independent varia- ble, combined with time, to capture daily and seasonal cycles characterised by non- linear behaviour. Regression-based Machine Learning algorithms, such as Gaussian Process Regression (GPR), are employed to predict the expected displacements. A Physics-Enhanced Machine Learning (PEML) approach, or grey-box model, integrat- ing physical knowledge with data-driven insights is adopted, improving accuracy and interpretability. Tested on real-world data from a highway viaduct, the grey- box model demonstrates superior performance and robustness, even with limited datasets. This confirms the potential of PEML-based approaches for damage assess- ment with data from static monitoring, paving the way for more reliable SHM sys- tems and enhanced bridge safety.