Abstract
In industrial processes, process data often exhibit complex characteristics, such as nonstationarity and nonlinearity, which brings challenges to process monitoring. In this study, a monitoring strategy for nonstationary processes is proposed based on cointegration theory and multiple order moments. Considering the nonstationarity presented in some variables, cointegration analysis (CA) is applied to obtain long-term equilibrium relationships among these nonstationary variables, which are then combined with stationary variables to form a new stationary dataset. For the purpose of process monitoring, a new monitoring index that contains multiple order moments is proposed to capture different statistical features of a previously obtained stationary data set. Moving windows are applied to capture changes of local statistical characteristics to implement online monitoring. Case studies on simulation data and an industrial dataset are presented to illustrate the effectiveness of the proposed method for nonstationary process monitoring. Comparing with the PCA and common CA-based monitoring methods, the proposed method has better performance with a lower false alarm rate and earlier alarm time.
Highlights
With increasing attention to industrial safety, process monitoring and control have attracted broad public attention
Considering the complex characteristics presented in industrial process data, a monitoring strategy based on cointegration analysis and comprehensive statistics is proposed
For nonstationary processes that cannot be monitored by traditional multivariate statistical methods, a monitoring strategy based on cointegration theory and multiple order moments is proposed
Summary
With increasing attention to industrial safety, process monitoring and control have attracted broad public attention. Yu et al [10] proposed a recursive cointegration analysis (RCA) for nonstationary industrial processes, in which the monitoring model was updated when the long-term equilibrium relationship of process variables extracted by cointegration analysis changed. The method can avoid frequent model updating compared with common adaptive methods On this basis, Zhang et al [11] applied recursive principal component analysis (RPCA) to capture remaining short-term dynamic information after extracting the long-term equilibrium relationship by RCA. The residual of long-term equilibrium relationship of multivariate nonstationary time series will contain dynamic components and exhibit non-Gaussian traits, which will lead to incorrect monitoring results. Considering the complex characteristics presented in industrial process data, a monitoring strategy based on cointegration analysis and comprehensive statistics is proposed.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
