Research On Remaining Useful Life Prediction Method Of Rolling Bearing Based On Neural Network

As a key component of rotating machinery,the service performance of rolling bearings has a great impact on the reliability and safety of mechanical operation.Rolling bearings have been in service under complex conditions for a long time,and once failure occurs,it may cause significant economic losses or endanger life and health.At present,scheduled periodic maintenance was still the main form of bearing maintenance,but this method has under-maintenance and over-maintenance.Therefore,it is very necessary to carry out maintenance based on the prediction of the remaining useful life of bearings,which effectively reduce maintenance costs,reduce accident rates,and ensure the safe and reliable operation of equipment.Taking rolling bearings as the research object,this thesis carried out a research on the remaining useful life prediction based on data-driven neural network,and realizes the remaining useful life prediction of rolling bearings by establishing a neural network model,The specific research contents of this thesis are as follows:Taking rolling bearings as the research object,the remaininguseful life prediction method based on neural network was carried out.By establishing and training neural network model,the accurate prediction of the remaining useful life of rolling bearings was realized.Specific research contents were as follows:(1)The accelerated life experiment of rolling bearings based on BPS experimental platform was built,load and speed experiment conditions were set,bearing failure conditions were designed,and accelerated life test of rolling bearings was carried out.Acceleration life tests were carried out on rolling bearings,acceleration signals were collected,and the changing trend of acceleration signals during the life evolution period of dynamic bearings was studied.(2)Based on feature analysis and convolutional neural network,the method of early fault recognition for rolling bearings was studied.The characteristic quantities in time and frequency domain that were more sensitive to bearing faults were studied.Principal component analysis was used to fuse the selected characteristic information and determine the approximate time when rolling bearing faults occur.A rolling bearing early fault recognition model based on convolutional neural network was established,and the difference between normal signals and fault signals was highlighted by the advantage of weight sharing of convolutional neural network.The effects of different network super parameters and sliding window step on early fault identification were compared,and the comparative experiments were carried out on IEEE-PHM-2012 data set and BPS experimental data set.The experimental results showed that the proposed fault recognition algorithm can accurately extract the early fault characteristics of test bearings,which lays a foundation for the future bearing life prediction.(3)Prediction method of Residual life of rolling bearings based on SE Module attention mechanism and Conv LSTM neural network.The ensemble empirical mode decomposition algorithm was used to decompose the original vibration signal,and the decomposed signal was reconstructed based on kurtosis criterion.The proposed SEConv LSTM neural network was used to construct the health index of the reconstructed signals of rolling bearings,and the particle filter was used to predict the remaining useful life of rolling bearings.Through the comparison of IEEE-PHM-2012 data set and BPS experimental data set,the results showed that the proposed method can effectively identify the remaining service life of experimental bearings.