A study of new model-based fuzzy control techniques and their verifications

Fuzzy logic control is the application of rule-based expert systems to process control. Traditional fuzzy logic controller design relies on the availability of heuristic rules. Knowledge acquisition is often considered to be the bottle-neck in expert system design. Due to the heuristic nature of these controllers, there exist few stability criteria and analytical tools; none of them are widely accepted.;In this research, fuzzy control techniques based on 'deep', non-heuristic process knowledge and classical control theory were investigated. This culminated in two methods; each in a different way addresses the tuning problems often encountered when applying fuzzy controllers. Both methods were verified, one based on simulation and the other one experimentally.;Most fuzzy logic controllers belong to a class known as fuzzy heuristic controllers (FHC's) which bear similarities to conventional PI controllers. By exploiting this property, the algebraic relations between their parameters were established, which enabled the adoption of PI tuning techniques to tune FHC's. A number of MIMO PI controller design/tuning methods were investigated. Based on step tests sufficient process information for proper tuning is obtained. The study of a simulated cement kiln process demonstrated the viability of this technique on MIMO systems with severe process interactions.;The model-based fuzzy gain scheduling (MFGS) technique was developed based on fuzzy gain scheduling (FGS). Simulation results demonstrated significant improvement in controller performance. Unlike FGS, MFGS utilizes process knowledge to enhance gain scheduling. Although FHC's can perform nonlinear control, they are designed for specific operating conditions. The FGS technique is not limited by that restriction. A process space is partitioned into a number of regions each described by a linear model and a corresponding linear controller. Fuzzy logic is used to interpolate controller parameters at region boundaries. A fuzzy gain scheduler is equivalent to an FHC with variable consequents. Since it uses linear controllers, it does not suffer from tuning problems often encountered in FHC implementation.;MFGS was applied to PID control of a laboratory-scale water-gas shift reactor. The closed-loop experimental results were compared with performance of traditional FGS, conventional gain scheduling, simple PID and nonlinear model predictive control (NMPC) strategies. The MFGS was comparable in performance to the NMPC method. It exhibited excellent control behavior over a wide operating range. The other three techniques were only adequate within a limited scope of the operating space. Due to the simple algorithm involved, the MFGS technique provides a low cost/effort alternative to other computationally intensive strategies.