Research On Remaining Useful Life Of Turbofan Engine Based On Deep Learning

Turbofan engine is the core of the aircraft,the health status of turbofan engine analysis for the evaluation of aircraft,safe use and maintenance strategy is crucial,and the remaining useful life(RUL)is the key to measure the health of turbofan engine indicator,to accurately predict the remaining useful life,both to avoid safety accident,failed to timely maintenance It can also reduce the expensive cost of excessive maintenance.Due to the characteristics of many measuring points,complex working conditions and large amount of data,the prediction of the remaining useful life of turbofan engine is faced with great challenges such as feature extraction and low prediction accuracy.How to extract effective features and improve the prediction effect of the remaining useful life of turbofan engine is a hot and difficult topic in the field of life prediction today.Based on the above background,based on turbofan engine as the research object,this thesis combined with deep learning method,in view of the original data set noise interference,feature extraction of multi-dimensional sensors is complicated due to the variety of sensors and large amount of data,unable to effectively describe data or degradation trends,traditional data driven model to predict the effect not beautiful,remaining useful life prediction was studied for the turbofan engine,The main contents are as follows:1.A method for predicting the remaining useful life of turbofan engines based on the fusion of spatio-temporal features was proposed.The method was combined with 1D convolutional layer neural network(1D-FCLNN)and long short-term memory network(LSTM)to predict the remaining useful life of turbofan engines.The data set was input into the whole model,and 1D full convolutional layer neural network(1D-FCLNN)extracted the spatial features of the data set.Meanwhile,LSTM was used to extract the temporal features of the data set,and the second-class features were used as the input of the next 1D convolutional neural network 1D-CNN after fusion.Finally,the remaining useful life of the turbofan engine is obtained.In the case of using the same data set,compared with the currently popular remaining useful life prediction model,comparison results show that the proposed fusion the spatio-temporal features of turbofan engine remaining useful life prediction model in terms of life prediction accuracy than other models have bigger promotion effect,the final evaluation index also suggests that the validity and superiority of the model.2.Taking the turbofan engine data set as the research object,the features of the remaining useful life prediction model proposed above based on the fusion of spatio-temporal features are automatically extracted by one-dimensional full-convolutional layer neural network(1D-FCLNN)and Long and short-term memory network(LSTM)respectively.In order to obtain more accurate features and further improve the prediction effect,an improved stacked sparse autoencoder structure(im SSAE)combined with attentional mechanism based echo state network(Atten-ESN)was proposed to predict the remaining useful life.The original noise is removed and the data is reconstructed using the3 sigma criterion.The improved stacked sparse auto-encoder is used to extract the features of each engine cycle and perform feature dimension reduction.In addition,Batch Normalization layer and Dropout layer are used in the encoder to solve the problems of gradient disappearance and over-fitting.Then extracted features built into the HI value engine,engine deterioration trend,finally introduced in ESN network attention mechanism,the adaptive processing of different types of input and get the remaining useful life prediction,the data set of the results showed that the turbofan engine compared with the currently popular remaining useful life prediction model,this model has higher prediction accuracy in RUL prediction and its HI index and evaluation index results are better than other prediction models.