A set-based uncertainty quantification of evolving fuzzy models for data-driven prognostics

Recent years have seen a great deal of innovation in the field of systems prognostics and health management. However, even with these advancements, some pertinent issues related with uncertainty in remaining useful life predictions are still open for investigation. One such area of interest is on how to account for the distribution of these predictions such that all uncertainty sources are duly captured and represented. Practically, these uncertainty quantification procedures must be computationally feasible for real-life deployment and reflect real-life situations devoid of strong assumptions. This article thus, proposes a data-based prognostics technique that uses a set-based quantification of uncertainty based on the set-membership paradigm, the interval predictor approach. The methodology is applied in the framework of the Evolving Ellipsoidal Fuzzy Information Granule which has recently proven its potency in prognostics applications. As a case study, the method is tested on the prognostics of insulated bipolar transistors utilising an accelerated aging IGBT dataset from the NASA Ames Research Center.
Peer Reviewed
Postprint (author's final draft)