Impact of ambient temperature on structural response: data-driven models for damage detection in SHM systems

The accuracy and reliability of Structural Health Monitoring (SHM) systems are significantly affected by environmental and operational variations, with ambient temperature playing a major role. Temperature fluctuations modify material properties such as stiffness and alter support conditions, impacting structural response data recorded by SHM systems. These variations may lead to false-positive damage detections or obscure actual structural deterioration, underscoring the necessity of incorporating temperature effects in SHM analysis. This dissertation addresses these challenges by examining thermal inertia, which causes a time lag between temperature changes and structural response measurements. To account for this, sequential temperature measurements are integrated into regression models to enhance the accuracy of structural response predictions and damage detection. A range of regression models is developed and tested, including linear regression, autoregressive with exogenous inputs (ARX), autoregressive-moving average with exogenous inputs (ARMAX), Nonlinear AutoRegressive with eXogenous inputs (NLARX), and the deep learning-based WaveNet model. These models are applied across multiple case studies, including finite element simulations, experimental data from an aluminum truss bridge, and real-world monitoring of the Z24 Bridge and Munich test bridge. The models incorporate temperature history from recent hours, and their predictive accuracy and damage detection capabilities are evaluated. Findings reveal that considering temperature history significantly enhances SHM predictions, reducing errors in identifying structural damage. The NLARX model strikes a balance between computational efficiency and predictive performance, making it highly applicable to real-world scenarios. Although WaveNet offers superior accuracy, its high computational demands limit its practicality. In conclusion, integrating temperature history into SHM models greatly improves long-term infrastructure monitoring. The choice of regression model depends on data availability and computational resources, requiring a trade-off between precision and efficiency. These findings contribute to more robust SHM frameworks, enhancing the reliability of damage detection and infrastructure management.