Remaining Life Prediction Of Equipment Based On Model Fusion

With the development of aviation technology,the aero-engine becomes more and more precise,more and more complex,more and more difficult to predict the remaining useful life(RUL)of the accurate engine.As an important part of fault prediction and health management(PHM),residual life prediction of equipment is of great significance.In this paper,we design a model-based fusion method,using the C-MAPSS data set provided by NASA platform as a benchmark,to further the research of RUL prediction in turbofan.The main contents are as follows:(1)Firstly,the characteristics of C-MAPSS data set are analyzed,a fusion model of turbofan residual life based on Convolutional Neural Network(CNN)and Bi-directional Long Short-Term Memory(BILSTM)was proposed.The C-MAPSS dataset was used to test the performance of the model by root mean square error and Score function.The network parameters of CNN-Bi LSTM are optimized by experiments.(2)Aiming at the limitation of single network,a method of multi-network model was designed to predict the RUL of engine.RUL was predicted by using the idea of fusion of two-channel features,and two different channels are selected for feature extraction.In Channel 1,the original data was trained by CNN to acquire the local features,and then the Bi-LSTM was input to acquire the features extracted by Channel1.On the basis of channel one,the output of BILSTM network was weighted to the selfattention mechanism.The features of the two-channel output are fused,then the fused features are output to the full connection layer,and finally,the RUL of the turbofan was predicted.Finally,the precision of the two-channel feature extraction model is higher than that of the single network model.