Computational damage detection and diagnosis in composite structures based on metamodeling aided Bayesian inference scheme

This proposal develops a comprehensive Bayesian diagnostics scheme for the localization and quantification of damage in composite structures based on metamodeling strategy and ultrasonic guided-waves. A surrogate model is constructed in lieu of expensive explicit FE analysis, with numerous training data calculated from the wave propagation model based on the wave and finite element (WFE) scheme. The Bayesian model selection approach is implemented to rank and select the most plausible time-frequency models for processing measurement data. An enhanced Markov chain Monte Carlo (MCMC) algorithm is employed for the sampling process to obtain the important region of uncertain parameter space. The proposed two-stage damage detection scheme allows (1) the classification of the defect modes based on Bayesian inference, and (2) the localization and quantification of different candidate damage modes identified based on specific parameters.