On Automation of the PID Tuning Procedure

Detta är en avhandling från Department of Automatic Control, Lund Institute of Technology, Lund University

Sammanfattning: Within process industry, and in many other areas, the PID controller is responsible for handling regulatory control. An educated guess is that the number of executing PID control loops lies in the billions (2011) and there are no signs indicating a decrease of this number. Properly tuning the PID controller, i.e., setting its parameter values based on characteristics of the process it controls together with robustness criteria, is commonly both timely and costly. Hence, the tuning is often overseen, resulting in numerous poorly tuned loops. These result in unnecessary lack of performance, which might be both hazardous and uneconomic. If a linear time invariant model of the process is given, there exists numerous feasible tuning methods. However, automatically obtaining even a low complexity model is far from trivial in the absence of a priori process information. This thesis addresses system identification to be used in the automatic PID tuning procedure. A method for generating the identification input signal is proposed. Its objective is to yield higher model accuracy in the frequency range where it is most needed for robust tuning. Subsequently, methods for obtaining process models from input and output data pairs are proposed and discussed. All methods are presented using numerous simulations and laboratory experiments. Finally, a simulation study of closed-loop anesthesia in human patients, based on clinically obtained model parameters, is presented. The novelty lies in that the depth of hypnosis PID controller is individualized based on data collected during the induction phase of anesthesia. It is demonstrated that updating the controller, using a herein proposed method, significantly improves performance.