NEW DEVELOPMENTS IN IDENTIFICATION AND CONTROLLER DESIGN FOR AUTOTUNING OF LINEAR PROCESSES

Abstract

The need for an increasing degree of automation in process industries has led to the requirements for higher standard autotuning control procedure. This thesis reports on new developments in identification and controller designs that contribute to autotuning of linear processes. In particular, a robust algorithm for the estimation of frequency response of SISO processes in the existence of noise and disturbance is devised. New tuning methods for PID controllers using exact gain and phase margins and model reduction PID tuning are proposed. A novel independent multiloop controller design for multivariable processes is also developed. The proposed identification method is designed to work in the relay feedback setting. Effects of random noise can be suppressed in the estimated frequency response by applying the averaging technique on input and output samples that spread over an excessive number of limit cycles. Complete elimination of linear disturbance is possible by performing multiple relay tests. Verification of the hypothesis has been carried out through extensive simulations and it is found that by involving more samples in the averaging operation, a reduced relative error in the resulting estimation can be achieved given a constant noise-to-signal ratio (NSR). Alternatively, the relative error can be maintained in the case of increasing NSR if more limit cycles of samples are collected. For the PI and PID controllers which are designed to tune to user-defined gain and phase margins, if the specifications can be precisely met regardless of the process order and damping nature, the response of the closed-loop system is then more predictable than those using ordinary model-based method. The PI design is based on finding the intersections between two graphs that are plotted using the frequency response of the process. The given gain and phase specifications can be achieved if intersections can be found, and each intersection corresponds to one solution. Suggestions are provided on how to modify the specifications for achieving satisfactory responses for cases where there is no solution or when the specifications can be met but poor responses result. For the PID tuning, since the set of unknowns outnumbers the set of equations in the problem, an extra constraint is introduced by relating the closed-loop bandwidth to the process one. This also leads to a simple solution to the gain and phase margin problem. Simulation results are included to demonstrate the effectiveness of the method. The model-based PID tuning method is developed using a second-order plus dead time modeling technique and a closed-loop pole allocation strategy through the use of root-locus. Satisfactory responses can be expected for processes with various dynamics, including those with low- and high-order, small and large dead time, and monotonic and oscillatory responses. Simulation examples are given to show the competence and flexibility of the controller in handling processes of different characteristics. Real time experimental results are included. An extension to the Internal Model Control strategy is also presented. The new independent design method for multi-loop controllers exploits process interactions for the improvement of loop performance. Unlike many other methods that emphasize on the suppression of interactions for decoupling purpose, our method is developed to channel the effect of interactions to individual loops for speeding up loop responses. This is achieved by regarding each loop together with its corresponding interactions from all other loops as an equivalent SISO plant, and designing for it an independent SISO controller. Once an objective transfer function is specified for each of these equivalent processes, a set of simultaneous equations is formed and separated into independent ones, each of which contains one controller element only. They are then solved to obtain exact solutions, which are usually irrational. The exact solutions can be well approximated by rational functions. The popular multi-loop PID controllers can naturally be obtained as a special case of rational approximation and they give reasonable trade-off between loop and decoupling performance. Simulation examples are provided to show the effectiveness of the proposed method and improvements over the BLT method are obtained.