Process monitoring based on canonical variate analysis

A new process monitoring strategy is proposed based on canonical variate analysis (CVA). State-space models are identified from input/output data collected over different time periods, and then an optimal test of hypothesis concerning process change is computed by comparing the respective model fits, using the Akaike Information Criterion. The new process monitoring strategy is illustrated in a simulation example, a continuous tank reactor (CSTR) problem.