Remaining Useful Life Prediction Of Rolling Bearing Based On Time Frequency Analysis And Deep Learning

Rolling bearing is one of the most commonly used core basic parts in mechanical equipment.Its health condition will affect the precision and safety of the whole mechanical system.The rolling bearing in operation may be worn,the inner and outer rings of the bearing broken or the ball damaged due to uneven lubrication,corrosion,excessive load and other reasons,thus affecting the normal operation of mechanical equipment and causing industrial production losses.Therefore,the prediction of the remaining useful life(RUL)of rolling bearing can provide guidance for the maintenance of mechanical equipment,and then carry out maintenance in advance,reduce the maintenance cost of mechanical equipment and improve its safety.Based on the direction of data driven RUL prediction technology,based on the theory of deep learning artificial neural network,a new rolling bearing RUL prediction model based on time-frequency analysis and depth learning is proposed in this paper,which is based on the characteristics of large noise interference,adulteration of useless signal and low energy in real industrial environment The work is as follows:(1)Aiming at the problems of incomplete analysis and low prediction accuracy caused by the original process data including noise,low energy and no use signal,the RUL prediction method of most rolling bearings is proposed.The original time-domain vibration data of rolling bearings is analyzed by using time-frequency analysis technology.The single time domain analysis ignores the frequency domain characteristics of data,which leads to incomplete analysis of degradation characteristics,which has certain limitations.Therefore,time-frequency analysis technology is needed to solve this problem.In order to fully exploit the degradation characteristics of the vibration signal of rolling bearing and improve the prediction accuracy of RUL,this paper uses wavelet transform with high time and frequency resolution to process the sample data to obtain the corresponding time-frequency data set.(2)The existing RUL prediction models based on theoretical mechanics and statistics are more and more difficult to predict RUL under the increasingly complex situation of mechanical equipment,and rely on expert experience.With the rapid development of information technology,the data collection of industrial process becomes efficient and convenient,which provides sufficient data resources for data driven RUL prediction technology,which gradually shows great development potential.In view of the above problems,this paper takes rolling bearing as the research object,and carries out the research work of the deep learning RUL prediction model based on the hot branch of data driven.In order to reduce the burden of neural network,the time-frequency data set of rolling bearing is reduced by bilinear interpolation algorithm,and finally used as input of neural network.After that,the paper studies the RUL prediction methods of rolling bearing under different structures,such as BP neural network,traditional convolution neural network,full convolution neural network and global pool neural network.After the output of the network is obtained,weighted average noise reduction algorithm is used to reduce the noise interference and fluctuation phenomenon of the prediction results,so as to further improve the prediction accuracy.The experimental results on the open data set show that the RUL prediction model architecture based on time-frequency analysis and depth learning has high prediction accuracy.