ABSTRACT
Structural health monitoring (SHM) methods are essential for the identification of system properties and the accurate prediction of the performance and resilience of structural systems subjected to time-dependent actions. It is common to estimate the response of existing structures by numerical models that depend on unknown mechanical and/or geometrical parameters and to characterize probabilistically these parameters from input/output measurements by solving inverse problems in a stochastic setting. The quality of the resulting models depends on the magnitude of the measurement errors and the methods for solving the posed inverse problems. We propose a robust framework for damage identification that is based on the Bayesian representation of the uncertain parameters of the finite element model for dynamic analysis. The potential of the proposed method is assessed by virtual response samples that are polluted by measurement noise. The method is crucial to quantify the time evolution of structural performance and reliability and to develop suitable damage detection procedures for early warning. A cable-stayed footbridge is used as a case study to demonstrate the implementation and the capability of our method. The proposed method advances the field of data-driven SHM in three directions. First, the efficiency of the classical Bayesian updating procedure is improved by using polynomial chaos. Second, the updating algorithm is viewed as a model prediction error, e.g., the differences between measured and calculated natural frequencies and modal shapes. Third, the measurement errors are explicitly handled in the Bayesian updating procedures for solving the damage identification and localization problems.
