Similarity-based Prognostics for Remaining Useful Life Prediction of Engineered Systems

Over the past two decades, prognostics that aims at forecasting remaining useful life (RUL) has evolved as an enabling technique for the predictive maintenance of engineered systems. Thus, one novel similarity-based prognostic approach for accurate RUL prediction is presented in this work. In the offline stage, historical degradation trajectories are properly abstracted into the common degrading characteristics (i.e., one mean trend and a few varying modes) for the whole system population by functional principal component analysis, so that the capacity of degradation trajectory library is greatly reduced. During the online stage, reference degradation trajectories are automatically generated by the whole observed degradation trajectory of the testing system, thus the problem of selecting an appropriate degradation trajectory horizon/window for similarity comparison in traditional similarity-based methods is gently solved. Case study on cooling fan run-to-failure experiments demonstrates that the proposed method can achieve reliable RUL prediction and the errors decrease with the accumulation of the degradation observations.