Searching historical data segments for process identification in feedback control loops

Mathematical models of dynamic processes are often required for assessment, diagnosis and improvement of control loop performances in process industries. Data samples collected in daily operations of feedback control loops may enclose data segments suitable for process identification to obtain the mathematical models. This paper proposes a new method to search such data segments. The searching criterion is that the reference and process output in a feedback control loop should experience significant magnitude changes. Hypothesis tests are exploited to find changing positions of data segments with different probability distributions and to verify whether the reference and process output make significant magnitude changes inside one data segment or between two adjacent segments. Simulation and industrial examples are provided to illustrate the effectiveness of the proposed method.