Research On Remaining Useful Life Prediction Method Of Turbine Engine Based On Deep Learning

As the key power equipment of aircrafts,vehicles or ships,the operation state and reliability of turbine engine plays a decisive role in the safe and stable operation of the equipment.However,the complex and changeable working conditions,high temperature and high pressure working environment and other factors are likely to lead to the problem of engine performance degradation or even failure.Predicting remaining useful life(RUL)in the early stage of engine degradation and arranging reasonable maintenance plans can effectively avoid losses caused by unplanned downtime for maintenance and improve equipment safety.Effective prediction methods are critical for timely RUL prediction.In order to meet the demand of accurate prediction of turbine engine RUL,this paper carried out research on multi-dimensional perception data screening and preprocessing methods and remaining useful life prediction model construction based on deep learning.And a remaining useful life fusion prediction method of Stacked Denoised Auto-encoder(SDAE)and multi-layer Long Short-Term Memory(LSTM)network based on particle swarm optimization was formed.The main research contents include:(1)Research on screening and preprocessing methods of monitoring data.Aiming at the multi-dimensional perception data of turbine engine,the linear combination of the correlation and the monotonicity between variables and degradation state is taken as the evaluation criteria to reduce the data dimension and screen out the variables that can characterize the engine degradation trend.The data standardization processing is carried out,and the training samples of remain useful life prediction model are constructed through sliding time window.(2)Research on RUL prediction method based on multi-layer LSTM network.Aiming at the turbine engine data with long time series,a RUL prediction model based on multi-layer LSTM network is built to fully exploit the degradation characteristics and the time correlation between the front and back in the sensor data.The model network structure is determined and optimized by experimental methods.The prediction performance of the model is verified by using the NASA turbine engine data set.The score function,RMSE,MAE and other evaluation indexes and confidence interval are used to quantitatively evaluate the prediction results.And through the comparison experiments with other prediction methods,it is verified that the constructed model has higher accuracy and stability in RUL prediction.(3)Research on RUL prediction method based on particle swarm optimization optimized multi-layer LSTM network.Aiming at the optimization problem of multi-layer LSTM network model,a hyperparameter optimization method of prediction model based on particle swarm algorithm is proposed,which optimizes batch size and the number of neurons in each layer of multi-layer LSTM network.Meanwhile,random weight parameters and asynchronous changing learning factors are used to improve the particle swarm optimization algorithm to improve the convergence speed and the convergence ability of the algorithm.Through multiple iterations,the prediction accuracy of the multi-layer LSTM model is improved.(4)Research on RUL fusion prediction method based on SDAE and multi-layer LSTM.Aiming at the monitoring data of multiple sensor types,the nonlinear relationship between degeneration state variables is automatically learned through SDAE,and the feature data after smooth processing is input into the multi-layer LSTM model optimized by particle swarm optimization for RUL prediction,giving full play to the advantages of the two models and overcoming the limitations of the single model.The research results show that the RUL fusion prediction method based on the two models can fully extract the deep features of the data,reduce the depth of the model and improve the training speed.Taking the multi-dimensional monitoring data of turbine engine as the input,this paper carries out in-depth research on the RUL prediction method based on deep learning and constructs a RUL fusion prediction model based on SDAE and multi-layer LSTM.The particle swarm optimization is used to optimize the model,which improves the model performance and reduces the engine RUL prediction error.It has great engineering application value for engine reliability maintenance and condition based maintenance in the fields of aerospace,vehicle or ship.