A generic Bayesian approach using Laplace approximation for model-based failure prognosis

Abstract

A generic Bayesian framework using Laplace approximation for model-based remaining useful life prognosis is presented in this thesis. The developed generic Bayesian prognosis approach models and updates the remaining useful life distributions by incorporating timely evolving sensory data using a general Bayesian inference mechanism, and employs an efficient Bayesian updating approach using Laplace approximation (LA) method. The developed Bayesian prognosis approach eliminates the dependency of evolutionary updating process on a selection of distribution types for the parameters for a given system degradation model. Furthermore, with the developed LA method, the Bayesian updating process can be carried out efficiently which makes the proposed approach possible for real-time prognosis applications. The proposed Bayesian prognosis methodology is generally applicable for different degradation models without prior distribution constraints as faced by conjugate or semi-conjugate Bayesian inference models. Electric resistor prognosis application is employed in this study to demonstrate the efficacy of the proposed prognosis methodology.