Nonstationarity and cointegration tests for fault detection of dynamic processes

Abstract

As continuous industrial processes often operate around a desirable region of profitability, the measurement series for most process variables act as stationary series. However, there are inevitably some observed time series which are nonstationary caused by unexpected disturbances. Some series grow slowly for a long time with the equipment aging, and others appear to wander around as if they have no fixed population mean. For these series, traditional dynamic PCA or other statistical modeling methods are not applicable because the statistical properties of variables are time variant. In this paper, nonstationarity test is adopted to distinguish nonstationary series from stationary series. After that, cointegration analysis is used to describe the stochastic common trends and equilibrium error, which can be used to construct monitoring indices. Case study on Tennessee Eastman process shows that the proposed nonstationary process monitoring can efficiently detect faults in the nonstationary dynamic process.