Non-stationary process monitoring based on common trends model was proposed

because conventional multivariate statistical process control methods with stationary data assumption were inapplicable to non-stationary process monitoring.The new common trends model was capable of identifying common factors from co-integrated non-stationary multiple variables and decomposing each non-stationary process variable into summation of a non-stationary common trends component and a stationary counterpart.Contrary to existing non-stationary process monitoring technique from cointegration model

the common trends model captured the stationary component of each non-stationary process variable

eliminated effects of non-stationary common factors and unveiled overall dynamic equilibrium relationships among variables.Hence

non-stationary process monitoring was transformed to an application of common trends model

which involved obtaining stationary component of each process variable

creating estimation for the stationary components by conventional multivariate statistical methods and setting up monitoring on corresponding control limits.A case study of monitoring petroleum distillation process showed that the proposed approach possessed more reliable process monitoring performance than the method of cointegration model.