Data-driven Aero-engine Remaining Useful Life Prediction Method

Aero-engine is prone to failure and resulting in disastrous accidents because of its complex structure and harsh working environment.Thus,it is of great significance to analyze and model the performance of the engine through scientific and effective methods,and predict the remaining life of the engine,as well as preventing and reducing the occurrence of catastrophic accident.This paper investigates the aero-engine’s remaining life prediction methods from a data-driven perspective,in which the performance degradation of engine is studied from two aspects: monotonous performance degradation and non-linear performance degradation.The main work and innovation of this paper are summarized as follows:(1)Aiming at the problem of traditional remaining life prediction methods which highly require prior knowledge and limits their applications,a trajectory similarity-based and a support vector regression-based prediction methods are respectively used to predict the remaining useful life of engine.Meanwhile,in order to solve the parameter setting problem of support vector regression,the differential evolution algorithm is brought to optimize the objective function to reduce the influence of subjective factors and improve the prediction accuracy of the algorithm.(2)Due to the limitation of its own structure and model parameters,the single prediction algorithm cannot achieve high prediction accuracy with respect to the degrees of degradation and it is difficult to guarantee its accuracy.In order to improve the robustness of the algorithm,an ensemble prediction model is proposed in this paper.The model not only combines the advantages of the trajectory similarity-based prediction methods and the support vector regression improved with the differential evolution algorithm,but also adopts the Ada-boost algorithm in the ensemble learning to fuse the two prediction algorithms organically.The experiments demonstrate that the ensemble prediction model yields a higher prediction accuracy compared with other high-performing algorithms.(3)The traditional methods require the priori degradation distribution function of aero-engine when they predict the remaining useful life of the engine with non-linear trend of degradation,and have poor applicability.To solve this problem,the deep learning algorithm is applied in this paper to predict the remaining useful life and excavates the potential rules from engine’s sensor data.As aero-engine data is a type of time series data,this paper utilizes the long and short memory recirculation neural network to overcome the gradient vanishing problem in traditional recurrent neural network.Meanwhile,this network is also improved to solve the problem of low prediction accuracy in the regression prediction so that it can accurately predict the remaining useful life of the aero-engine.Experiments on C-MAPSS dataset show that the model has good performance on predicting aero-engine’s remaining useful life.