Residual life prediction method of bearing based on improved JRD and error correction
At present,it is difficult to identify the initial degradation of the gearbox performance of wind turbines,and the existing degradation indicators are prone to dramatic fluctuations and poor monotonicity,which can not accurately predict the remaining useful life(RUL)of key components of the gearbox such as bearings.To solve this problem,a dual-exponential model bearing RUL prediction method with improved Jensen-Rényi divergence(JRD)and error correction is proposed.Firstly,the multi-domain characteristic index was extracted from the vibration signal sample,the posterior probability distribution vector of the sample was obtained by using Gaussian mixture model(GMM)and exponential weight JRD,and then the confidence value(CV)was obtained by normalization.Then,phase space reconstruction was performed on the CV values of the bearing from the initial healthy state up to the current inspection moment,in order to extract the dynamics of the CV sequence and use it as a training set for the relevance vector machine(RVM)to obtain the correlation vectors that underpin the entire degradation trajectory.Finally,the resulting correlation vector was fitted using a biexponential model,the trend was extrapolated to the failure threshold,and the RUL was calculated;at the same time,the differential autoregressive integrated moving average model(ARIMA)was introduced to predict the fitting error generated by the fitting correlation vector to correct the prediction results.The experimental results verify that the monotony index of the improved degradation index is increased by 14.3％,and after the error correction,the RUL prediction results of bearings under different working conditions and at different times are significantly improved.The research results show that this method can provide reference value for predictive maintenance planning of important components of wind turbine gearbox.
rolling bearingremaining useful life(RUL)predictionGaussian mixture model(GMM)Jensen-Rényi divergence(JRD)error correctiondual-exponential modelconfidence value(CV)differential autoregressive integrated moving average model(ARIMA)
万方数据
维普
