Vibration-based Bayesian updating of hull girder finite element models

Abstract

Recent advances in sensing technology and data processing have enabled using structural response data collected in situ to identify typically unmeasurable quantities of interest governing the response characteristics of complex structural systems. These updated quantities can then be used to identify incipient structural damage or to improve the predictive accuracy of engineering models simulating structural response, thereby enhancing structural operation and management capabilities. The present study applies the principles of Bayesian model updating to identify equivalent stiffness properties of a high-fidelity Finite Element model of a real-world coast guard vessel, using vibrational measurements collected from the actual vessel. Initially, the parameter space of the high-fidelity Finite Element model is reduced heuristically to derive equivalent stiffness-related parameters. Subsequently, these equivalent parameters are updated so that the predicted modal frequencies match those obtained through Operational Modal Analysis performed using acceleration signals measured on the vessel. Parameter updating is performed via Bayesian inference. The influence of different modal information on inference and identifiability is explored and the potential of using the Bayesian posterior predictive distribution for damage identification is discussed.