Bayesian model averaging based reliability analysis method for monotonic degradation dataset based on inverse Gaussian process and Gamma process

A Bayesian model averaging based reliability analysis method for monotonic degradation modeling and inference is proposed in this paper. Considering the model uncertainty, the Bayesian model averaging method is applied to combine the candidate monotonic processes, specifically the Gamma process and inverse Gaussian process. To evaluate the population reliability, the unit-to-unit variations and heterogeneities within product population are highlighted, so the random effects of both the model parameters and model probabilities are taken in to account. The fully Bayesian inference is applied to estimate distribution hyper-parameters, in which the priors are obtained by moment estimation combined with maximum-likelihood estimation. The proposed Bayesian model averaging based reliability analysis method is verified using previously published GaAs laser degradation dataset. The results indicate that the proposed Bayesian model averaging based method provides flexibility when evaluating the population reliability.