Predicting fatigue damage in composites: A Bayesian framework

Modeling the progression of damage in composites materials is a challenge mainly due to the uncertainty in the multi-scale physics of the damage process and the large variability in behavior that is observed, even for tests of nominally identical specimens. As a result, there is much uncertainty related to the choice of the class of models among a set of possible candidates for predicting damage behavior. In this paper, a Bayesian prediction approach is presented to give a general way to incorporate modeling uncertainties for inference about the damage process. The overall procedure is demonstrated by an example with test data consisting of the evolution of damage in glass–fiber composite coupons subject to tension–tension fatigue loads. Results are presented for the posterior information about the model parameters together with the uncertainty associated with the model choice from a set of plausible fatigue models. This approach confers an efficient way to make inference for damage evolution using an optimum set of model parameters and, in general, to treat cumulative damage processes in composites in a robust sense