Quantifying the impact of model-plant mismatch on controller performance

In model predictive control (MPC) of processes, the model fidelity plays an important role. The performance of MPC depends on the quality of the model and hence on the model-plant mismatch (MPM). A poor model may not necessarily lead to degradation in the performance. Moreover, performance degradation may also be a consequence of poor tuning and/or disturbances. Hence, it is highly desirable to be able to isolate the role of MPM in poor control and quantify its impact. This work seeks to address these issues and two rules that aid in the diagnosis of poor performance are proposed. The methodology described here is based on the analysis of closed-loop data from the process. In this work, it is also shown that the impact of MPM on control quality depends on the setpoint directions. This directionality aspect of MPM is discussed by drawing analogies with the concept of gain directionality in multivariable systems. The ability of the proposed methodology to diagnose poor controller performance is demonstrated via representative simulation examples and an industrial case study.