Research On Rolling Bearing Fault Diagnosis And Remaining Life Prediction Method Based On Big Data Technology

Rolling bearing is the most widely used part of mechanical equipment,especially turbomachinery,but it is also one of the most vulnerable parts.The traditional prediction method of rolling bearings' residual life usually takes a feature of time domain or frequency domain to evaluate the performance degradation of bearing after signal de-noising,and the model is established to predict the residual life.However,a signal index is difficult to fully characterize the degradation process of bearing performance and the model is too simple,which lead to inaccurate prediction result.Therefore,with the help of large data technology,this paper studied the feature extraction from multi domain to be weighted,constructed a new evaluation index,and established an improved combination prediction model based on support vector machine to pretest the remaining life of bearing.Firstly,research backgrounds and research significance of feature extraction,fault diagnosis and residual life(P-F time)prediction of rolling bearings were investigated.The current research level of signal de-noising technology was summarized in this paper.The development of the big data technology and its application in the field of fault diagnosis was analyzed,and the key questions in the residual life prediction were summed up.The means of extracting feature information for appraising bearings' working condition were discussed.In addition,with the analysis of the P-F curve,the fault evolution law of rolling bearings and its performance degradation process were deeply studied.The meaning of residual life was accurately interpreted so as to carry out subsequent work.Secondly,the method of LMD and FastICA were adopted to eliminate noise,and the purpose was to reduce the inaccuracy of the diagnosis and the life prediction resulted from a large amount of noise in the bearing vibration signal.Through the simulation analysis,the feasibility of the method was verified.After that,the validity of the method for vibration signal de-noising was testified by the actual engineering data.Compared with the signal processing results of the commonly used filtering methods,it was found that the proposed method had a better effect in dealing with the noise reduction of the bearing vibration signals.Then,in order to solve the issue that the performance degradation evaluation index was constructed difficultly,features of time domain and frequency domain were extracted to form a set.The KPCA method was applied to analyze the set and the first kernel principal component containing the key information of the feature set was selected as the evaluation index.Measuring the evaluation capability of this index by the bearing full life test data and found that it not only reflected the law of bearing performance degradation well,but also gave good expression to initial failure.Finally,aiming at the problem that the accuracy of traditional SVM model was low when predict rolling bearing's residual life,the LSSVR model was studied.As the input of the prediction model,principal components achieved from the time domain and frequency domain feature set after kernel principal component analysis and meeting the requirements were chosen.Subsequently,the model was established to predict the residual life of the rolling bearing.Moreover,the genetic algorithm was used to optimize the model parameters.The validity of the prediction method was validated by the bearing full life test data.The results showed that the method could pretesting more precisely than other prediction models.