Study On The Vibration Condition Monitoring And Forecasting Technique For Machinery In Submarine

To realize the vibration condition monitoring and forecasting of machinery in submarine, the research work as being showed below has been implemented: the characteristic selection, the techniques including cluster analysis, discriminant analysis, Hotelling T 2 statistic process control, AR(n) time series analysis and GM(1,1) grey forecasting how to been applied in the classification and identification of machinery conditions, in the threshold setting, and in the machinery condition forecasting.The general principle of the characteristic selection was demonstrated at first. By comparing the hardware arithmetic capability demands of measures to achieve the characteristic of vibration signal, the 1/3oct acceleration spectrum was been selected to be the characteristic as a result, considering the reaction time of the system.The cluster and discriminate analysis techniques was investigated that how to apply in the classification and identification of machinery conditions. It's pointed out that the Hierarchical Clustering Method and the Distance Discrimination Method have advantages. The following investigations were expounded also: the methods to define the amount in the cluster process, the methods to estimate the error probability, the methods to verify the identification effect.The research for the threshold setting utilizing the Multivariate Statistics Analysis is carried out in this paper. And it's introduced that how to use T 2 Control Chart to practice the condition monitoring of machinery, and how to judged sub-vectors which should answer for the abnormity, providing the frequency band information for the consequent diagnosis.In the aspect of condition forecasting, the modeling methods of AR(n) model have been discussed, including the Least Square Estimation Method, the Levison Parameter Recursive Estimation Method,and in succession, the Gradually Vanishing Memory Method and the Limited Memory Method which are real estimation methods to solve the data saturation based on the Least Square Recursive Estimation Method. The modeling process of the GM(1,1) grey model is demonstrated, along with the measure to improve the forecast precision using the residual GM(1,1) grey model.The data disposal methods is presented in detail, which are data transform, data normality test, outliers verification, time series stationary test, and sample capacity determination which considering the result of cluster and significance level.The author finished the experiment design, and set down the experiment scheme and content.