Research On Similarity-based Remaining Useful Life Prediction Technology

Remaining useful life(RUL)prediction is one of the key technologies in the field of prognostics and health management(PHM).This thesis proposes a similarity-based RUL prediction method,which explores the impacts of multiple degradation behaviors.The work and contributions are summarized as follows:(1)The past degradation behaviors of equipment may reflect the operation status of the equipment to a certain extent.This thesis proposes an RUL prediction method based on the similarity of multiple degradation behaviors.It transforms the test equipment's history into multiple degradation behaviors,finds similar degradation behaviors for each of them as references,and then uses the corresponding reference RULs to predict the test equipment RUL.The experimental results show the proposed method performs better than the existed algorithms for it can make full use of the past degradation data and avoid candidate insufficiency.(2)Equipment degradation behavior may contain abundant information.Therefore,this thesis introduces several adjustment and optimization methods based on degradation behavior characteristics.Considering the length differences among similar degradation behaviors,this thesis proposes a degradation behavior extraction method based on data morphological characteristics,and develops a multi-scale scaling based search strategy to match similar degradation behaviors with uneven lengths.A degradation rate based RUL adjustment strategy and a degradation bias based RUL aggregation method are further proposed to make the reference RUL more suitable for RUL prediction.Experiments show the proposed methods can effectively improve the RUL prediction accuracy.