Closed-loop performance diagnosis for model predictive control systems

The time-varying dynamics of real systems often limit the life-time performance of model predictive control applications. A critical step for cost-effective maintenance of control systems is to distinguish between control-relevant plant changes and variations in disturbance characteristics in the event of an observed closed-loop performance drop. This paper addresses the latter performance diagnosis problem for a general class of model predictive control systems using prediction error identification. The performance diagnosis approach is demonstrated with a simulation case study.