Research On Evaluation Method Of Oil Monitoring Index Based On Multivariate Statistics

Oil monitoring technology is an industrial technology that analyzes the performance indexes of in-use equipment lube through various detection methods,and is the core link of state maintenance.By analyzing various indexes of the same oil sample,the inspector can obtain the status information of the equipment and evaluate the status of the mechanical equipment based on these information.At present,the acquisition of device status information is mostly achieved by means of numerical records,the work is cumbersome and the index evaluation method is single.Through the numerical simulation method,the related lube performance indicators are summarized,and the relevant models are established to evaluate the lube performance indexes,and more hidden equipment status information can be obtained,so that the oil monitoring technology is better applied actual industrial problems.Based on the past years' experimental data of Marine oil Monitoring Diagnosis Center of Dalian Maritime University,this thesis establishes two types of representative data sets—mixed sample data sets and small sample data sets with time series properties,and discusses the evaluation methods of oil monitoring indexes applicable to these two types of data sets.Then,This thesis combines mathematical methods such as multivariate statistics to improve and perfect several existing oil monitoring indexes evaluation models,and makes some specific examples based on Matlab and SPSS.For the mixed sample data set,this thesis has done the following research:(1)Use the trimmed mean method to calculate the recommended standard for the index.Use the distribution fitting method to obtain the approximate obeyed distribution of the index,and calculate the inverse cumulative distribution function to boundary value of the index.(2)Use factor analysis method to analyze the correlation between indexes and extract common factors.Use the surface fitting method to analyze the correlation between indexes.(3)Use cluster analysis to obtain the index value characteristics of suspected abnormal samples in this data set.For small sample data sets,this thesis has done the following research:(1)Use the trimmed mean method to calculate the recommended standard for the index.Use the distribution fitting method to obtain the approximate obeyed distribution of the index,and calculate the inverse cumulative distribution function to boundary value of the index.(2)Use the phase space reconstruction method to analyze the information loss of the index.(3)Use the GM(1,1)model and the difference statistical model to predict the index.Research on the evaluation method applicable to the mixed sample data set,the following conclusions are obtained:(1)Using the trimmed mean method instead of the average method to calculate the recommended standard can reduce the impact of outliers and improve the accuracy of the recommended standard for each index by 0%to 28%.Using the distribution fitting method can compensate for the deviation caused by the distribution assumption and improve the accuracy of dividing the limit value.(2)Viscosity and viscosity index indexes,lead element and boron element indexes,magnesium element and zinc element indexes,calcium element and silicon element indexes have obvious positive linear correlations.After sorting the common factors after factor rotation,it is concluded that the magnesium,zinc,mechanical impurities,silicon and calcium indicators are important indexes in this mixed sample data set.There are three related formulas for the standardized index:Viscosity-Flash point-Viscosity Index formula,Iron-Aluminum-Copper formula,Viscosity-Viscosity index-Magnesium formula.Using the above three formulas to estimate the index value,the probability that the error does not exceed 20%is greater than 0.83.(3)The viscosity,viscosity-index,mechanical impurities,total base number,silicon,calcium,zinc indexes of suspected abnormal samples are relatively low relative to the recommended standard,while the remaining indexes are relatively high.By writing a clustering model of the computational geometry method,it is possible to further achieve the characterization of the index value association at different scales and realize the identification of the index abnormal value.Research on the evaluation method applicable to the small sample data set,the following conclusions are obtained:(1)Using the trimmed mean method instead of the average method to calculate the recommended standard can reduce the impact of outliers and improve the accuracy of the recommended standard for each index by 0%to 50%.Using the distribution fitting method can also improve the accuracy of the boundary value division.(2)The K entropy of the time series corresponding to the index of moisture and mechanical impurities tend to infinite,and the rate of information loss is too large,so that it can't be effectively predicted.The K entropy corresponding to the index value of viscosity,viscosity-index,flash point and total alkali is small,and the rate of information loss is low.These indexes can be effectively predicted.(3)Based on the GM(1,1)model and the statistical difference model,a complementary correction scheme is established to predict the physicochemical indexes,which can improve the accuracy of each index forecast.Based on the statistical difference model,the distribution interval of the predicted values of each physicochemical index in the 26th sample of the data set is reduced.On the premise of sacrificing 20%probability,the predicted value of physicochemical indexes will not exceed these reduction range.