A comparative study of particle filters for prognostics implementation in industry

Abstract

Prognostics is an engineering discipline aiming at predicting
the Remaining Useful Life (RUL) of an industrial system or
item. Accuracy and confident prediction of the RUL are very
meaningful and important for anticipating failure, controlling
system operational efficiency as well as optimizing maintenance
operations. Given the important role of the prognostics
or RUL prediction, a number of prognostics approaches has
been proposed and successfully applied in various industry.
Among these approaches, particle filters (PF) are more and
more studied and employed thank to their powerful performance
and their flexibility in predicting the RUL of systems
non-linear and non-Gaussian. However, the prediction performance
strongly depends on the application contexts and
the type of particle filter utilized. The choice of particle filters
is therefore a critical step in real industrial applications. The
paper focuses on a comparison of the three different PF techniques
(Sampling importance resampling, Auxiliary particle
filter, and Regularized particle filter) to support the critical
step. The performance of the three PF techniques is compared
by considering different degradation models, noises level. In
addition, the computing time is also analyzed through different
numerical examples.