Data-Driven Diagnosis of Model–Plant Mismatch in MIMO Closed-Loop Control System

Abstract

Control performance of a multivariable closed-loop control system greatly depends on the quality of the process model used in the controller design. Thus, it is highly desirable that the discrepancy between the process model and the plant can be measured. In this study, a novel methodology is proposed for model–plant mismatch diagnosis based on an internal model control framework. The outputs are first whitened, and the external disturbance model is identified. A process model residual measuring the process model–plant discrepancy is put forward based on the identified disturbance model and the control model. Three model quality indices are defined to monitor the model qualities for the overall process, individual outputs, and specific input–output channels. This presented methodology can identify the mismatched subchannels and isolate the model–plant mismatch from the changes in controller parameters. The capability of the presented methodology is demonstrated using a simulated column process and a Tennessee Eastman benchmark problem.