Nonstationary Process Monitoring And Fault Diagnosis Using Cointegration With Structrural Change Testing Method

Cointegration theory, originated in 1980s, has become a useful tool in establishing a dynamic model for a non-stationary system, which commences to analyze the non-stationarity of time series and seeks a long-run equilibrium of a set of non-staionary variables. It is in the process of ever-perfecting and developing. The past decade witnessed cointegration theory and its practical application widely researched in economics. However, in engineering area, few academic works can be found to utilize the cointegration testing method for engineering system monitoring and fault diagnosis. As an initiative research, the main purse of this major is, based on inheriting and digesting the exsisten research findings, to continually seek new strategies of applying the cointegration testing method in engineering area, which shows uncommon features in dealing with non-stationary time series.This paper presents systematically the developing course of cointegration and its related core algorithm. Specific examples are also given to demonstrate how the cointegration testing method performs in the condition monitoring and diagnosis of two non-stationary systems, a simulated Fluid Catalytic Cracking Unit (FCCU) and a real debutanizer distillation unit. In these examples, existing problems are disclosed and analyzed, and corresponding solutions are consequently proposed. For instance, Augmented Dickey-Fuller unit test is revealed to be not sensitive enough to monitor some system fault conditions, and the reduced cointegration model method formerly used in fault diagnosis is proved to be useless if there is many a variable involving fault information. To resolve such problems, unit root test with structural change, referred to as Zivot-Andrews unit root test and Perron unit root test, and Gregory-Hansen cointegration test with regime shift are initiatively introduced for system condition monitoring and fault diagnosis. The application results indicate that cointegration technology will show much more good performances in engineering area.