Automated autocorrelation function analysis for detection, diagnosis and correction of underperforming controllers

Date of Completion

January 2009


Engineering, Chemical




This research focuses on industrially relevant controller performance assessment (CPA) metrics. CPA is an area of process control research dedicated to developing automated means to analyze how well control systems are operating. While many CPA methods have been established, few are capable of classifying controller behavior across the extremes from overly aggressive to very sluggish. In this dissertation, a new method based on automating the analysis of the autocorrelation function (ACF) provides a means to classify control performance across such a broad range. The method is run on stochastic data and does not require a priori knowledge of the process or deliberate process upsets. Thus it can be readily applied in industrial applications. ^ A second-order underdamped model is applied as a novel pattern recognition tool for classifying controller disturbance rejection responses. In the literature, often only exponential decay is considered in defining optimal control, however non-self regulating (integrating) processes cannot generally achieve this standard. The use of the underdamped model allows a full range of controller behaviors to be defined as optimal and characterized through a damping factor that represents both oscillations and exponential decay. The damping factor is then compared to desired performance through a Relative Damping Index ( RDI).^ The natural period of oscillation computed from the underdamped model is used in a unique solution to the problem of selecting the appropriate length of the ACF for analysis. An additional use of a second-order model is presented in the development of the Howard-Cooper Index that builds upon the minimum variance standard of the Harris Index. ^ The work goes beyond the detection and diagnosis of controller performance to the problem of selecting corrective action. To guide the user in retuning the controller to regain desired performance, visual tuning maps are presented for both self-regulating and integrating processes. The potential for an analytical solution to retuning through the development of tuning correlations based on the underdamped model parameter is also presented. The work was tested in various applications in the cogeneration power plant at the University of Connecticut. ^