Data Analysis and Graphical Presentation System for Control Loop Performance Assessment, 1998 Process Control, Electrical & Info. Conference Proceedings

A data analysis and graphical presentation software package for control loop performance assessment was developed. The package runs on a personal computer that obtains process data from a data server (mini-computer) connected to a distributed control system. Approximately 300 control loops are monitored daily collecting process data every 2 seconds for a 45 minute period. A normalized performance index (NPI), which was proposed by T. Harris [?], is calculated for each loop. Although the NPI provides a convenient measure for loop performance, it was felt that more information is necessary to grasp loop performance. The package presents the following data in graphical form: 1. time plot of setpoint and error, 2. histogram of error, 3. sample autocorrelation function, 4. sample partial autocorrelation function, 5. power spectrum estimates, 6. the NPI versus dead-time plot, 7. daily plot of the NPI and root mean square error. The NPIs are estimated by calculating autoregressive coefficients using the Durbin-Levinson algorithm, which gives the estimate of the partial autocorrelation functions as byproducts. It was found that most of the loops can be approximated well by high order AR models. This makes the NPI estimation calculation fast, and the package can plot the NPIs for various dead-times, on-the-fly, as each data set is selected by the user. A few examples from paper mill process data are presented. The mill data shows wide variety not only in the NPI, but also other statistics. We should utilize many statistical and time-series tools in diagnosing control loops. The package allows the user to observe the analysis results in graphical form quickly.