Particle filtering prognostic estimation of the remaining useful life of nonlinear components

Bayesian estimation techniques are being applied with success in component fault diagnosis and prognosis. Within this framework, this paper proposes a methodology for the estimation of the remaining useful life of components based on particle filtering. The approach employs Monte Carlo simulation of a state dynamic model and a measurement model for estimating the posterior probability density function of the state of a degrading component at future times, in other words for predicting the time evolution of the growing fault or damage state. The approach avoids making the simplifying assumptions of linearity and Gaussian noise typical of Kalman filtering, and provides a robust framework for prognosis by accounting effectively for the uncertainties associated to the estimation. Novel tailored estimators are built for higher accuracy. The proposed approach is applied to a crack fault, with satisfactory results.