Abstract: Aircraft engines are with highly complex structures, a large number of performance parameters monitored and low remaining useful life (RUL) prediction accuracy in their degradation processes. A method based on the similarity of degradation characteristics is proposed for predicting the RUL of aircraft engines. Firstly, the Relief algorithm is used for feature selection, principal component analysis is used for feature extraction, and the Kernel algorithm is used for feature trajectory smoothing. Through this, a few lowdimensional orthogonal degradation characteristics are obtained for RUL prediction. Then, a similaritybased matching algorithm is adopted to find out a group of similar degradation segments in the referenced samples, and the corresponding RUL information of these degradation segments are integrated via a density weighting method to obtain the final RUL prediction of the current sample. Finally, the proposed method is evaluated by using the turbofan engine degradation simulation data announced by national aeronautics and space administration (NASA), and the results can show its superiority compared with several popular RUL prediction methods.

Key words: remaining useful life (RUL) prediction, performance degradation, Relief feature selection, similarity, density weighting