Prediction Of Remaining Useful Life For Aero Engine Based On Feature Optimization

The aero engine health management and remaining useful life prediction is one of the key technologies in the aerospace field,and it is an important technical mean to improve engine safety,reliability and economy.As the core technology of health management,remaining useful life prediction aims to provide technical support for operation planning and maintenance decision by predicting the remaining life of the whole machine or components.In recent years,the research on aero-engine remaining useful life prediction has found that a single feature data will directly affect the prediction accuracy of remaining useful life.And too much feature data has the problems of information redundancy and insufficient consideration of time series information.Therefore,a method of aero engine remaining useful life prediction based on feature optimization is proposed.The research focuses on monitoring data prediction,multi-objective optimization of features and remaining useful life prediction.Firstly,a single sequence monitoring data prediction model is established based on long-term memory network to predict the data sets of engine test samples under different operating conditions.Secondly,based on feature fusion and similarity matching method,the root mean square error of remaining useful life value of predicted data set is calculated.The evaluation index of remaining useful life prediction is defined,and the feature optimization is carried out based on Multi-Objective Particle Swarm Optimization(MOPSO)algorithm and Non-dominated Sorting Genetic Algorithm II(NSGA-II)respectively.Finally,the effect of engine remaining useful life prediction based on feature optimization is verified.A new improved calibration strategy(degradation limit ratio α)is proposed,and the influence of this strategy on engine remaining useful life prediction is analyzed.The results show that:1)The fitting ability of the single sequence monitoring data prediction model based on long-term memory network reaches a good level,and this model can better predict the engine characteristic monitoring data in a certain future period.2)In a certain quantity rang of features,the effect of remaining useful life prediction shows a trend of continuous improvement with the increasing number of features.However,when the number of features continues to increase,the effect of remaining useful life prediction begins to decline.It shows that redundant features will lead to the deterioration of engine remaining useful life prediction,and proves the necessity of feature optimization for engine remaining useful life prediction.3)The sample sets of different working conditions are applied to feature multi-objective optimization and remaining useful life prediction respectively.The results show that the proposed method is correct and robust.As the degradation limit ratio α decreases from 0.65 to0.35,the prediction effect of residual life under different number of features is increased by more than 40%.It indicates that the proposed improved calibration strategy can effectively improve the engine RUL prediction effect.