Research On Remaining Useful Life Prediction Of Rolling Elements Bearings With GRU Network

Rolling bearing is widely used in rotating machinery,whose running state is related to the overall performance of the equipment.Therefore,the health management and predictive maintenance of rolling bearings is significance to prolong the life of the equipment and improve the operational reliability of the whole mechanical equipment.Among them,the remaining useful life(RUL)prediction is one of the most important technologies to implement the health management and predictive maintenance of rotating machinery,which has received widespread attention.The gated recurrent unit(GRU)network is used to study the RUL prediction of rolling bearings.The research contents are as follows:Due to the different degradation states of each bearing,this paper establishes a health indicator(HI)estimation model for vibration signals,and then combines particle filter algorithm to predict RUL.In this model,firstly,aiming at the problem of fluctuation and inconspicuous degradation trend in basic features,the basic feature extraction is combined with the complete ensemble empirical mode decomposition with adaptive noise method is proposed to extract trend features.Furthermore,monotonicity and correlation were used to select features.Then,aiming at the problem that the bearing failure threshold is difficult to determine,the selected best feature set is input into GRU network to obtain HI and the particle filter algorithm is used to obtain the RUL of the tested bearings.In view of the fact that the degradation characteristics of rolling bearings are not obvious in the healthy stage,this paper further studies the method of monitoring its running state in the health stage,alarming and triggering RUL prediction device when the rolling bearings begin to degrade.On the basis,the algorithm takes into account the different degradation modes of bearings,and proposes a method to build different prediction model for different degradation modes.In the fast degradation mode,the degradation information of the training bearings is less.Firstly,the time-domain features are selected,and the selected degradation features are enhanced with the generated adversarial network.Next,the features are transformed by relative transformation and anomaly correction entropy transformation and then input into the GRU network model to predict RUL.In the slow degradation model,RUL is directly predicted without degradation feature enhancement.Finally,simulation experiments are performed on the IEE-PHM-2012 challenge data set and the XJTU-SY data set,and compared with several common machine learning algorithms.The results show that the proposed method can not only effectively improve the prediction accuracy,but also accurately track the degradation process of bearings.This provides an effective maintenance strategy for equipment maintenance,so as to extend the life of equipment.