Monitoring the control loop on the basis of reduced process model-application to the DC motor-generator

A statistical test for detecting changes in the control loop based on the reduced order model is presented. The residuals are obtained by comparing the behaviour of the reduced order nominal model with the process behaviour. Any discrepancy might indicate a change in the process. However, there are two possible reasons for the discrepancy: process anomalies; and model error. The model error might cause false alarms, if it is not properly considered in the test statistics. Namely, in classical test statistics the residuals variance is invariant, while the residuals of the reduced order model depends on the process input. This paper describes the way of including unmodelled dynamics into the test statistics when the undermodelling is described by the stochastic embedding method proposed by Goodwin and co-workers (1992). The result of such a procedure is a decrease of false alarms. The performance of the proposed test statistics are tested on a DC motor-generator laboratory set-up.