Enhancing Bridge Structural Health Monitoring: Integrating Physics-Based Knowledge and Machine Learning for Temperature Compensation

Abstract

This study explores the application of a grey-box modeling approach for bridge SHM, integrating physics-based principles with machine learning (ML) techniques to enhance predictive accuracy. Using data collected from a tied-arch bridge, including acceleration, and temperature measurements, the study first evaluates different ML models, such as Linear Regression and Gradient Boosting Regressor, to eliminate the temperature effect from the natural frequency of a bridge cable. While purely data- driven models yield reasonable predictions, their accuracy improves when enhanced with prior physics knowledge. Building upon this, a grey-box model is applied to further refine the temperaturecompensation process, assessing its effectiveness in the SHM domain. The findings demonstrate that a grey-box model offers a more robust and reliable solution by leveraging fundamental physical principles alongside data-driven learning. This approach proves particularly beneficial in real-world SHM scenarios, where environmental and operational variability complicate damage detection and monitoring.