Predicting Residual Life Of Rolling Bearing Using IMMFE And Improved Recurrent Neural Network

In rotating machinery operation process,due to the acquisition of signal noise,the nonlinear and non-stationary signal,caused by the residual service life prediction of rolling bearings recognition rate is low,rolling bearing is usually in the middle of the high speed rotating machinery or heavy load,which has a great effect,if the bearing become malfunction,which can lead to heavy load mechanical systems become paralyzed,lead to loss of people’s life safety and property damage.Therefore,the feature extraction of rolling bearing and the recognition of rolling bearing life prediction are very necessary for predicting the remaining life of rolling bearings.In this paper,predicting the remaining life of rolling bearing is studied based on IMMFE(Improved Mean Multiscale Fuzzy Entropy)and Bi LSTM-GRULR(Bidirectional Long Short Term Memory-Gated Recurrent Unit-Lasso Regression).The specific contents are as follows:(1)The principle of CEEMDAN(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise)algorithm is explained,a Complete Empirical Mode Decomposition algorithm is analyzed and verified respectively through the rolling bearing simulation signals and the real fault signals of rolling bearings.The results of Ensemble Empirical Mode Decomposition EEMD(Ensemble Empirical Mode Decomposition)and CEEMD(Complete Ensemble Empirical Mode Decomposition)are compared respectively to select the most appropriate bearing signal denoising method.(2)For the IMF(Intrinsic Mode Function)decomposed by CEEMDAN,the entropy method featuring improved mean multi-scale fuzzy entropy is innovatively proposed.First of all,the value of the improved mean multi-scale fuzzy entropy of each IMF component is calculated,and then the features of the improved mean multi-scale fuzzy entropy and the traditional time-frequency domain are screened by cosine similarity,so as to select the optimal features similar to the rolling bearing degradation curve.(3)Bearing state identification is needed to predict rolling bearings.In order to predict the remaining using life of the rolling bearing accurately,the degradation state of the rolling bearing is divided by the RMS feature of the rolling bearing life signal.The degradation state of the rolling bearing is divided into the stationary period and the degradation period.A Bi STM-GRULR(Bidirectional Long Short Term Memory-recurrent unit-Lasso Regression)degrade state classifier was proposed,which divided the life period of rolling bearings into stationary period and degradation period,and established the remaining life model of rolling bearings in different states.(4)The multi-stage prediction model for the remaining life of rolling bearings based on Bi LSTM-GRU-LR network is proposed.In order to verify the effectiveness of the proposed algorithm,the remaining using life of the tested bearings under the same working condition is predicted in this paper.In view of the unsolved problems about the rolling bearing life prediction at present,the next step of the future work is planned in the conclusion and summary,hoping to improve it step by step in the future.