Research On Key Technologies For Similarity-based Residual Life Prediction Of Electromechanical Device Under Different Degradation Variables

The normal operation of electromechanical equipment is closely related to the daily life of people and the operation and production of enterprises.Accurate residual useful life(RUL)prediction of electromechanical equipment is an important means to guarantee their safe operation.The similarity-based residual life prediction(SbRLP)approach is an emerging data-driven RUL technique.It has many advantages like modeling without degradation signal,strong robustness,and high prediction accuracy and thus it is suitable to be applied in RUL estimation of electromechanical equipment.However,up to present,literatures concerned are not only limited in quantity,but also has many problems,for example,not considering the influence of operation condition and that of maintenance,scarce researches on the SbRLP method with multiple degradation variables(e.g.the MSbRLP method)and not considering the selection of degradation variables.Those issues greatly limit the performance improvement of the SbRLP method and finally affect the accuracy of the SbRLP method of electromechanical equipment.Therefore,in order to obtain more accurate RUL of electromechanical equipment and enrich existing theories about the SbRLP method,it's essential to take the Sb RLP method of electromechanical equipment as the object,selectively analyze key problems which influence its accuracy improvement and finally put forward models,techniques,or methods to solve these problems.For this reason,based on fuzzy mathematics theory and grey theory,this paper has had an intensive study on key technologies for electromechanical equipment's SbRLP under different degradation variables.The main contents consist of the following parts:(1)The performance of the SSbRLP method of electromechanical equipment is improved from two aspects: considering operating conditions and considering maintenance and consequently a novel weight function and a fresh similarity measurement are advanced respectively.During the research of considering operating conditions,the triangular fuzzy number and data standardization are firstly hired to preprocess operating conditions of the operating sample and reference samples;secondly,the variable weight method is employed to calculate weights of reference samples according to operating conditions after pretreatment;finally,the RUL of the operating sample is predicted based on weights of reference samples.During the research of considering maintenance,firstly,similarities between the operating sample and reference samples are calculated on the basis of maintenance effect;then the RUL of the operating sample is predicted united with similarities between the operating sample and reference samples.Meanwhile,in the example analysis of the gyroscope's RUL estimation and the case study of steam feed pump's RUL estimation of enterprise A,the effectiveness and superiority(in terms of statistically more accurate RUL prediction results)of the SSbRLP method considering operating conditions and that considerting maintenance are demonstrated through comparisons with the classical SSbRLP method.(2)The prediction scheme construction for the MSbRLP method of electromechanical equipment is advanced.First of all,limitations of the SSbRLP method are analyzed and hence the MSbRLP method is mooted;secondly,the key problem of the MSbRLP method is analyzed and combined with information fusion level,two kinds of prediction schemes are put forward,namely the scheme implementing the fusion after prediction and that implementing the fusion before prediction;thirdly,two abovementioned prediction schemes are constructed through principal component analysis(PCA)and the weight adjustment coefficient.Eventually,in the example analysis of the gyroscope's RUL estimation and the case study of steam feed pump's RUL estimation of enterprise A,the necessity of researching on the MSbRLP method and the effectiveness and reasonability of two proposed prediction schemes for the MSbRLP method are illustrated.(3)On the basis of the scheme implementing the fusion after prediction,an original selection process for degradation variables and that for reference samples are introduced,and thus the MSbRLP method of electromechanical equipment based on the scheme implementing the fusion after prediction and two selection processes is mooted.Firstly,the correlation coefficient,Spearman coefficient and multiple objective function are utilized to establish the selection function for degradation variables,whose purpose is to obtain suitable degradation variables and then initial weights of those selected degradation variables are determined;secondly,failed historical samples are screened in light of the type and quantity of selected degradation indicators and their standard degree are computed;thirdly,united with the standard degree of each selected reference system,initial RULs of the operating sample is estimated by the SSbRLP method;in the end,integrated with their initial weights,decisive weights of degradation variables are obtained by the weighted average method and then the ultimate RUL of the operating sample is forecasted linked with with its initial RULs and decisive weights of degradation variables.Simultaneously,in example analysis of the gyroscope's RUL estimation,the proposed selection process for degradation variable,that for reference samples and the MSbRLP method based on the scheme implementing the fusion after prediction and two selection processes are illustrated to be feasible and advantageous through comparisons with the classical MSbRLP method.(4)On the basis of the scheme implementing the fusion before prediction,a new fusion method on dimension of degradation variable(i.e.fusing multiple degradation variables into real-time health degree),and a novel fusion method on dimension of operating time(i.e.fusing multiple degradation variables into the grey generation rate sequence)are respectively presented.During the research of the former,firstly,PCA,SVDD,Markov distance and negative conversion function are all hired to calculate the real-time health degrees of the operating sample and reference samples;then,according to the real-time health degree of samples,the RUL of the operating sample can be predicted by the classical SSbRLP method.During the research of the latter,firstly,the grey accumulation generation relational analysis model is employed to compute grey generation rate sequences of the operating sample and reference samples;then based on grey generation rate sequences of samples,the similarities between the operating sample and reference samples are calculated by the grey similarity correlation analysis method;finally,the RUL of the operating sample is predicted through the similarity weighted average based on the similarities between the operating sample and reference samples.Simultaneously,in the example analysis of the gyroscope's RUL prediction,the reasonability and effectiveness of the MSbRLP method based on the fusion on dimension of operating time and that based on the fusion on dimension of degradation variables are validated through comparisions with other methods.