Novel RUL prediction of assets based on the integration of auto-regressive models and an RUSBoost classifier

Abstract

This paper presents a novel, data-driven algorithm for the computation of the Remaining Useful Life (RUL) of an asset. The algorithm utilizes the asset's state history to learn a prognostic model from data. The prognostic model comprises an ensemble of Auto-Regressive (AR) models, together with a state-of-the-art classifier. The AR part of the algorithm is used to predict the system's state evolution. The classifier discriminates between healthy and faulty operation, given the asset's current state. The predicted state, as computed by the AR model, is fed to the classifier. The first time when the predicted state is classified as faulty is returned as the RUL of the system. The resulting prognostic algorithm was tested on the CMAPSS dataset as provided from NASA Ames Research Center. Cases of unknown future input trajectory as well as cases with multiple faults have been investigated.