Research On Remaining Useful Life Prediction Method Of Aero Engine Driven By Trajectory Similarity-based Prognostic

With the development of artificial intelligence and big data technology,the equipment in the field of aerospace,transportation,energy has become more intelligent and diversified,lots of sensors are deployed to the various parts of the equipment and monitor the running state of the equipment,the massive data can help us effectively depict the current degradation of equipment,which are used to estimate the residual life,this scheme can effectively reduce the cost of maintenance and improve the reliability of equipment.For the prediction of residual life about aeroengine in C-MAPSS,this paper studies an improved method based on the scheme of trajectory similarity-based prognostic.In view of the abnormal value caused by the traditional scheme of trajectory similarity-based prognostic,I proposed an improved TSBP method.Meanwhile,aiming at the problem of insufficient feature,I proposed a method named GRU-MLRGM.Finally,Analyzing the drawbacks of the mentioned deep learning method,I proposed the improved method based on segmented gradient matching strategy and K-means++method.The main contents of this paper are as follows:Firstly,ITSBP method is proposed.Using normal distribution principle and kernel density estimation technology to screen outliers produced by traditional TSBP method.By verifying the Dataset#1 and comparing the result with other TSBP methods,the experiment shows the effectiveness.Secondly,GRU-MLRGM is proposed which maps multidimensional data into low dimensional time series,then extracting deep feature by gated recurrent neural network.In the stage of similarity metric,determining the range of parameters by greed matching strategy.Comparing with existing method of deep learning,the result is more precise.Thirdly,proposing segmented gradient matching strategy which is used to replace greedy matching to reduce the search range of parameters and K-means++ method which is used to cluster to get a certain trend of degradation.Then the deep feature extraction is carried out through the neural network.The experiment shows faster and higher precision.