Benchmarking and tuning PID controllers

Abstract

It is important to be able to assess the quality of control provided by PID controllers or indeed by any classical or advanced control method. Since many PID controllers are set up using intuition or empirical tuning rules, it is particularly important that PID designs can be benchmarked and the quality of control assessed. Furthermore, benchmarking methods can provide guidance for tuning controllers. The most common performance measures used for PID control include step-response tracking performance, disturbance-rejection performance properties and the stochastic performance in regulating or tracking processes. With respect to the regulating performance, assessment and benchmarking methods can mostly be considered part of the minimum variance or generalised minimum variance family of techniques. There are commercial tools which use such methods, but these are often based on rather simplistic strategies. One of the advances made in the last decade has been the development of so-called restricted structure benchmarking which provides a figure of merit which is much more representative of what might be achievable if the controller is tuned optimally. With respect to step-input and disturbance rejection performance, the benchmarking should ideally determine how well the output of the system follows a step change in the system inputs (either a set-point or a disturbance change). This suggests a comparison with optimum tracking controllers, of which the most representative would be a model-based predictive controller. Thus, in the second part of this chapter the performance of a PID controller will be compared with the performance of a predictive controller.