Identifying new prognostic features for remaining useful life prediction

An accurate prediction of the remaining useful life (RUL) from a prognosis system relies on a good selection of prognosis features. The latter should well capture the trend of the fault progression. In situation where the existence of the fault to failure data is rare and the development of degradation based model is difficult, we must be addressed to the identification of new features having the qualities mentioned above. This paper present an new selection method based upon the Particle Swarm Optimization (PSO) algorithm to identify the advanced prognosis feature and the Hidden Semi Markov Model (HSMM) for the prediction of the remaining useful life. This method was validated on a set of experimental data collected from bearing failures.