A data-driven framework for remaining useful life estimation

Abstract

Remaining useful life (RUL) estimation is one of the most common tasks in the field of prognostics and structural health management. The aim of this research is to estimate the remaining useful life of an unspecified complex system using some data-driven approaches. The approaches are suitable for problems in which a data library of complete runs of a system is available. Given a non-complete run of the system, the RUL can be predicted using these approaches. Three main RUL prediction algorithms, which cover centralized data processing, decentralize data processing, and in-between, are introduced and evaluated using the data of PHM’08 Challenge Problem. The methods involve the use of some conventional data processing techniques including wavelets denoise and similarity search. Experiment results show that all of the approaches are effective in performing RUL prediction.