Research On Mechanical Equipment Health Index And Remaining Useful Life Prediction Based On Recurrent Neural Network

With the advent of the industrial information age,people have higher requirements for manufacturing efficiency.How to avoid the shutdown and production stoppage caused by equipment failure has become a difficult problem to be overcome.At present,factories and enterprises generally adopt methods based on experience and regular maintenance.This will undoubtedly increase labor costs and operational management expenditures,making the company less competitive in the market.Mechanical equipment degradation trend modeling and Remaining Useful Life(RUL)estimation techniques are the key to effectively solving this problem.At the same time,with the continuous upgrading of intelligent algorithms and the improvement of numerical computing capabilities,as well as the innovation of sensors and storage technologies,data-driven mechanical device degradation trend modeling and estimation of remaining service life are increasingly achievable.Deep learning can often learn higher-dimensional abstract features by simulating the human brain's thinking mechanism,and has achieved excellent results in many fields.Now it is gradually used in industry.The advantage that the cyclic neural network has strong learning ability for time series is just suitable for the learning of multi-sensor time series degradation model.The first section first describes how the data-driven approach addresses device degradation trend modeling and remaining life prediction issues.The data preprocessing methods of industrial datasets and how to model time series by two different methods are introduced.Then the CIndRNN model is proposed by combining the convolutional neural network and the independent recurrent neural network.The high-dimensional features are extracted from the sensor data and the reconstructed multivariate time series is used to learn the equipment degradation information,which is used for the RUL of the aircraft turbine engine.The forecast has achieved good results.Then Long Short-Term Memory(LSTM)is used to propose a new codec method to model the icing process of the fan blade,and use the reconstruction error to establish the health index(HI)of the fan blade.It helps to grasp the timing of deicing and prolong the service life of the fan.Finally,the above two methods are summarized to construct a relatively complete solution for monitoring the health of the equipment in the remaining service life prediction.At the same time,it also points out that the two methods still have shortcomings,which can be further studied in the following.