A Study On Statistical Diagnosis Of Autocorrelated Processes

The most important of the assumptions made concerning control charts is that of independence of the observations. In recent years, with the inprovement of data collection and measuring, the assumptions of data independence been questioned. In fact, the assumptions of data independence can't be satisfied in many cases. The existance of data relevant, even very slight relevance, will seriously affect the process control. This has been highly concerned in academic and business, and a great deal of discussion of the autocorrelation procedure has turned up. It has been discovered that, autocorrelation not only affect the statistical process control as an established mode, but also affect the statistical process diagnosis with the transition of autocorrelation mode. Mode exception mixed with the exception of factors and reasons, has enhanced the complexity to the statistical process diagnosis. While there are few papers discussed about the diagnosis of autocorrelation process. This paper aims at discussing diagnosis of autocorrelation process.The paper proposed a method for diagnosing the change of autocorrelation mode, which is based on autoregressive T~2 statistic. The key idea is to change the autocorrelation to inter-correlation. The method is discussed in detail, and a table for the diagnosis of first order autocorrelation process is got. The paper analyses the effect of the method through simulation. When the autocorrelation of formal mode is positive and strong (a>0.6), and the autocorrelation of exceptional mode change to negative, the method is effective. If the autocorrelation is strong and negative, and the autocorrelation of exceptional mode change to the other direct, the method is also effective.This paper also extend the method based on autoregressive T~2 statistic, then it can be suitable to diagnosis various autocorrelation problems. Furthermore, the diagnosis method witch is based on autoregressive T~2 statistic is extended to multivariate autocorrelation processes. A diagnosis table is got, witch is used to diagnosis univariate signal and mode signal, including autocorrelation and correlation.