A comparative study of particle filters for prognostics implementation in industry

Abstract

Prognostics is an engineering discipline aiming at predictingthe Remaining Useful Life (RUL) of an industrial system oritem. Accuracy and confident prediction of the RUL are verymeaningful and important for anticipating failure, controllingsystem operational efficiency as well as optimizing maintenanceoperations. Given the important role of the prognosticsor RUL prediction, a number of prognostics approaches hasbeen proposed and successfully applied in various industry.Among these approaches, particle filters (PF) are more andmore studied and employed thank to their powerful performanceand their flexibility in predicting the RUL of systemsnon-linear and non-Gaussian. However, the prediction performancestrongly depends on the application contexts andthe type of particle filter utilized. The choice of particle filtersis therefore a critical step in real industrial applications. Thepaper focuses on a comparison of the three different PF techniques(Sampling importance resampling, Auxiliary particlefilter, and Regularized particle filter) to support the criticalstep. The performance of the three PF techniques is comparedby considering different degradation models, noises level. Inaddition, the computing time is also analyzed through differentnumerical examples.