Damage identification using Bayesian derivative method

Resumo

Structural Health Monitoring (SHM) is one of the most challenging topics in structural dynamic analyses. This challenging characteristic comes from several uncertainties related to boundary conditions, damping description, material properties, and unknown excitation inputs. In this regard, formulating damage identification strategies based on the Bayesian framework is a feasible alternative to conjugating prior information, computational models, and measured data. Nevertheless, building metrics based on comparisons between experimental and computational dynamic features of structure leads to difficulties in determining the likelihood model. In this context, the present work brings an approach for damage identification that considers a metric for the inverse problem, which contains information about the relative difference of natural frequencies. The lack of knowledge about the likelihood structure is tackled using the Approximate Bayesian Computation (ABC), where draws from the approximate posterior are obtained using the Sequential Monte Carlo Approximate Bayesian Computation (SMC-ABC). The feasibility of this approach is assessed considering data from an experimental set-up composed of an aluminium beam containing lumped masses at some positions. The present strategy estimates the position and magnitude of the lumped masses attached to the supported beam. The central role of these lumped masses is to simulate structural anomalies. The inference process considers uncertainties related to structural damping by calibrating several concurrent structural models whose differences reside in the model damping. Finally, the strategy provides the probability of each concurrent model at the end of the process.