Bayesian Finite Element Model Updating and Assessment of Cable-Stayed Bridges Using Wireless Sensor Data

Area of Science:

• Structural Engineering
• Computational Mechanics
• Bayesian Inference

Background:

• Finite element (FE) model updating is crucial for assessing the structural health of bridges.
• Long-term monitoring data from wireless sensor networks (WSNs) offer valuable insights but require robust analysis methods.
• Accurate quantification of uncertainty in FE models is essential for reliable structural predictions.

Purpose of the Study:

• To develop and validate a Bayesian inference framework for FE model updating of a long-span cable-stayed bridge.
• To improve the accuracy of modal property prediction and uncertainty quantification using long-term WSN data.
• To assess and compare different approaches for handling prediction-error precisions within the Bayesian framework.

Main Methods:

• A robust Bayesian inference method is proposed, incorporating marginalization of prediction-error precisions.
• The Transitional Markov Chain Monte Carlo (TMCMC) algorithm is employed for sampling posterior distributions.
• Bayes' Theorem is applied at the model class level for Bayesian model class assessment, implementing a Bayesian Ockham's razor.

Main Results:

• The proposed method, marginalizing prediction-error precisions, yields a more accurate FE model with robust uncertainty quantification and modal property prediction.
• The TMCMC sampler effectively characterized posterior distributions of structural parameters and assessed the plausibility of different model classes.
• The updated FE model demonstrated high accuracy in predicting the modal properties of the cable-stayed bridge using real-world monitoring data.

Conclusions:

• The developed Bayesian inference framework provides a robust approach for FE model updating of long-span bridges using WSN data.
• Marginalizing prediction-error precisions enhances the reliability and accuracy of structural model updating.
• The Bayesian model class assessment procedure effectively balances model fit and complexity, leading to accurate predictions of structural behavior.