Ambient vibration-based cable tension monitoring and uncertainty analysis in cable-stayed bridge from a fully probabilistic perspective

This study assesses a fully probabilistic method of ambient vibration-based cable tension monitoring by conducting long-term monitoring of a cable-stayed bridge. A data-driven and physics-informed Bayesian cable tension estimation method with Bayesian model selection is proposed, which offers a way to consider the influence of uncertainty in cable tension estimation under ambient vibration. The specificity of ambient vibration in the cables at bridge deck and cable-bridge interaction is assessed along with operational modal analysis (OMA) and subsequent cable tension estimation. A comprehensive study is conducted of identification uncertainty (IU) and environmental and operational variation (EOV) effects on long-term monitoring using a predictive probabilistic model formulating global variability. Temperature compensation is implemented in a probabilistic manner using a Gaussian Mixture Model (GMM) and a conditional negative log-likelihood function (NLLF), which provides a flexible and generalized method for describing complex data with multimodality and arbitrary error distribution. Results verified that conditional NLLF, as a damage-sensitive feature (DSF), performed better than NLLF because it compensates both the temperature effect and the influence of temperature distribution with a more stationary time history under a healthy state and a higher alarm ratio under a damaged state.