Research On RUL Prediction Method Of Rolling Bearings Based On CEF And Integrated KELM

Rolling bearings are one of the most important components in rotating machinery.The complex working environment and operating conditions make it extremely vulnerable,which in turn affects the operation of the machine.Rolling bearing remaining useful life(RUL)prediction can provide a theoretical basis for preventive maintenance decisions,and develop a good plan for fault maintenance in advance,thereby the operating cycle of the entire mechanical equipment can be lengthened and the accidents will be avoided.The data-driven method is one of the effective methods for rolling bearing RUL prediction,including data acquisition,feature processing and RUL prediction.The paper takes the RUL prediction method of rolling bearing as the main line,and the research is gradually unfolded from two aspects: feature extraction and RUL prediction.For feature extraction,in order to represent the performance degradation trend of the rolling bearing comprehensively and accurately,first,the original vibration signal is smoothed by using the five-point three-spacing method to reduce the interference noise component of the original vibration signal.Then the variational mode decomposition(VMD)is used to decompose the smoothed original vibration signal into several modes,the time domain and frequency domain features of each mode are extracted,and finally the original feature set can be constructed.Aiming at the problem that the amplitude range of different characteristics of different rolling bearings is inconsistent,the similarity measurement algorithm is studied,and the structured feature set is normalized and reduced.At the same time,in order to obtain the features which have monotonicity,stability and similarity,a new cycle enhancement feature(CEF)is proposed and regarded as the performance degradation feature of the bearing.The validity of the proposed CEF feature is verified byexperimental comparison.The RUL prediction model includes two parts: the construction of the kernel extreme learning machine(KELM)prediction model and the integrated KELM prediction model.Aiming at the problem that the prediction error is large for single prediction model,an integrated forest KELM prediction model is proposed based on random forest(RF).The CEFs of multiple rolling bearings are input into multiple KELMs,and the strong predictor model KELM-RF can be constructed by the RF integrated algorithm to predict the future failure time and calculate the RUL.In order to make the obtained RUL more accurate,the second order exponential smoothing method is used to fit the current life,and the life of each point in the future life trends is predicted.The experimental results show that the proposed method has higher prediction accuracy than single limit learning machine and traditional integration method.