| The research on the trend of the rolling bearing degradation and the prediction of the residual life is one of the basic research contents for ensuring the safe and reliable operation of the mechanical equipment.This paper uses data from the bearing fatigue test data of NASA's pre-diagnostic database as the basic data.Using multi-frequency fuzzy entropy and extreme learning machine method,the vibration signal feature extraction of rolling bearings,the construction of bearing performance degradation evaluation index,the degradation trend prediction and the remaining life prediction problems are studied.The main work is as follows:1.Based on the principle of being sensitive to the occurrence of early failures of rolling bearings and performance degradation,eight time-domain features and ten frequency-domain features were screened out of many time-domain and frequency-domain feature quantities.In the time-frequency domain and entropy domain,combining empirical mode decomposition and fuzzy entropy,a feature extraction method of "multi-frequency-scale fuzzy entropy"(IMFFE)is proposed,and more sensitive9-frequency fuzzy entropy feature quantities are obtained.2.Based on the principal component analysis(PCA)method to build bearings performance degradation assessment indicators.According to different fields,PCA dimension reduction fusion is performed on the sensitive features after screening.Fusion has three indicators: time-domain fusion index,frequency-domain fusion index,and IMFFE fusion index.Analyzing the change rule of the three indexes with the sample number in the whole life cycle,and finding that the index after fusion is more sensitive to the bearing performance degradation than the single feature.In contrast,the IMFFE index is more sensitive.3.A bearing operating condition assessment method was proposed.This method uses the same model and approximate working condition of the old bearing with the whole life cycle monitoring data as the "reference bearing".Based on the historical monitoring data of the "tested bearing" being operated,the operating status is evaluated.1)According to the slope change of the polynomial curve of the IMFFE index curve during the whole life cycle,the demarcation points of the three phases of the "reference bearing" in the stationary operation period,the degradation period and the expiration period are determined,and three "state lines" with different slopes are used to mark this three phases.Perform polynomial curve fitting and normalization on the so far IFMFE index curve of "tested bearing",and calculate the current point to the "reference bearing" three state line distance.The state of this bearing is the same as the state of the nearest "state line" in the distance.The test results show that the evaluation method is more intuitive and simple,and the accuracy of the evaluation is relatively high.4.On the basis of state assessment,based on the optimized extreme learning machine(ELM)adaptive prediction algorithm,the degenerate trend and residual life prediction of rolling bearings are performed.The algorithm uses the "traversal selection method" to select two external parameters of ELM.At the same time,two internal parameters of ELM are selected by particle swarm optimization(PSO).The goal of selecting 4 algorithm parameters adaptively at the same time is achieved,and the prediction model based on the optimization parameters is established.According to rolling bearing degradation trend prediction problem,this paper proposes a dynamic multi-step iterative degradation trend prediction method,so as to realize the high-precision multi-step degradation trend forecast.There are many factors that affect the life of rolling bearings,and the single index cannot meet the forecasting requirements.Using the three evaluation indicators established in this paper as input to the multivariable limit learning machine(MELM),the remaining life of rolling bearings is predicted,which improves the accuracy of the prediction. |
PDF Full Text Request