Damage Diagnosis and Prognosis in a Composite Structure by Surrogate Modelling and Particle Filtering

Abstract

Bayesian model updating has proven to be successful for inverse problem solutions, such as for the evaluation of the posterior probability distribution of the parameters of interest, conditional on the observations of some damagedependent features. Specifically, the sequential Monte-Carlo procedure generates possible damage evolution trajectories based on available process models and then assigns each trajectory a weight according to its likelihood. However, for realistic structures, computationally expensive numerical simulations might be required for the evaluation of the trajectory likelihoods. In this work, the computational problem is addressed by leveraging on surrogate modeling, while particle filtering provides the sequential Bayesian framework for performing simultaneous diagnosis and prognosis of a composite double cantilever beam specimen subject to fatigue delamination growth, based on observations of the strain field pattern acquired at some specific locations. The algorithm is successfully tested with respect to a case study of a propagating delamination.