On performance assessment of feedback control loops

A reliable and profitable operation of chemical processes requires monitoring and diagnosis of control loop performance. Performance assessment technique, which gained increasing attention in recent years, is a suitable tool for addressing those issues. When the performance is measured by variability of key process variables, which is the case with most chemical process industries, the best achievable performance with feedback control is given by that of minimum variance control. The minimum variance performance bound can thus be used as a first-level benchmark performance when assessing performance of feedback control loops. In this regard, a simple method to estimate the minimum achievable variance in a cascade control loop is developed for the performance monitoring of cascade control system. Then a methodology for the estimation of multivariable minimum-variance performance bounds from the first few Markov parameters of the process and routine closed-loop output data is developed that does not require any knowledge of interactor matrix.; The use of the minimum variance performance bound as a benchmark is not suitable for processes where constraints on process variables are important, because it does not account for the performance limitation due to constraints on process variables. To obtain a meaningful minimum variance performance bound, a performance assessment method is developed for model predictive control systems that explicitly accounts for hard constraints on process variables.; When installed controller performance does not match that of minimum variance control, application of more realistic performance measures is desirable. A requirement for a certain controller structure such as PID control restricts achievable performance further than the performance limitation due to the process time delay. Thus, a controller-specific performance bound can be useful as a benchmark performance in the higher-level performance assessment. Numerical schemes are developed in this dissertation for the estimation of achievable performance with PID, DMC, and multiloop PID controllers.; In assessing control loop performance, the robustness of the installed controller is important for a proper assessment. The robustness, in this case, can be assessed by computing the variance of manipulated variable. For this purpose, a new design method for LQG controller is developed that allows direct calculation of variances of both controlled and manipulated variables.