Model selection, updating, and averaging for probabilistic fatigue damage prognosis

► A hierarchical Bayesian framework for uncertainty management is developed. ► A generic two-step procedure for reversible jump MCMC simulation is included. ► The computational efficiency is significantly improved comparing with traditional methods. This paper presents a method for fatigue damage propagation model selection, updating, and averaging using reversible jump Markov chain Monte Carlo simulations. Uncertainties from model choice, model parameter, and measurement are explicitly included using probabilistic modeling. Response measurement data are used to perform Bayesian updating to reduce the uncertainty of fatigue damage prognostics. All the variables of interest, including the Bayes factors for model selection, the posterior distributions of model parameters, and the averaged results of system responses are obtained by one reversible jump Markov chain Monte Carlo simulation. The overall procedure is demonstrated by a numerical example and a practical fatigue problem involving two fatigue crack growth models. Experimental data are used to validate the performance of the method.