Probabilistic Damage Severity Estimation of Stay Cables in a Cable-stayed Bridge Using Ambient Vibration Data

Abstract

In the management of infrastructures, ambient-vibration-based structural health monitoring (SHM) has become an important technique to evaluate the structural integrity due to its real-time and cost-efficient benefits. However, the inevitable uncertainty limits the application of this technique for practical use. This study addresses prevailing challenges by introducing a statistical method that integrates Multivariate Analysis (MVA) and Bayesian Inference (BI). It aims to estimate the damage severity of stay cables of a cable-stayed bridge using the monitored data while considering the monitoring uncertainties, such as environmental effects and measurement errors. Firstly, the Bayesian Fast FFT is applied to process the ambient vibration data, which can identify the most probable value (MPV) of modal parameters and the associated covariance matrix. Then, the tension of cables can be identified as an intuitive damage sensitive feature and also the evidence in BI. Based on the long-term measured data and the damage simulation results of the updated FE model, a bivariate probability density function (PDF) of the damage severity 𝜃 and monitored tension 𝑇 can be constructed using crude Monte-Carlo Simulation (MCS) and Kernel Density Estimate (KDE). Conditioning is then applied to derive various posterior distributions of damage severity given different monitored cable tensions, which gives a complete description of the uncertainty around the damage parameter estimate and the credible interval. The method is applied to the monitoring of a real cable-stayed bridge, where the first year of monitoring data is utilized for model construction, and the pseudo monitoring data is used for testing.