PART I
FUZZY SYSTEMS

Chapter 1
Why Fuzzy Process Control?

1.1 Introduction

Fuzzy logic can serve as a bridge between mathematics and language. Humans readily and efficiently manipulate quantities, concepts, and abstract linguistic terms. Computers, on the other hand, manipulate mathematical expressions with lightning speed. In fuzzy systems, traditional production rules are combined with fuzzy membership functions to produce systems that are easily comprehended by humans and effectively applied by computers. The fuzzy sets provide meaning to the linguistic terms used in the rules. There is nothing fuzzy about the logic, rather the sets used to define the linguistic terms are fuzzy.

Fuzzy systems are becoming increasingly popular as tools for solving difficult problems, and many products employing fuzzy logic have recently been marketed. Numerous types of problems have been solved effectively in a wide range of application domains. Process control is no exception. Fuzzy logic is being embraced by the controls community because it greatly simplifies the development of efficient and comprehensible systems.

When this research effort began, the standard for industrial process control was the proportional-integral-derivative (PID) controller. The setpoints for individual PID control loops were determined by human operators based on their experience with some particular system; different operators developed their own impressions of how the system should be controlled and adjusted the setpoints accordingly. The authors have witnessed situations in which an operator spends his or her entire eight-hour shift tweaking setpoints to get the system “running smoothly.” The next operator on duty immediately changes these setpoints. When questioned as to why he changed the setpoints the previous operator spent an entire shift establishing, the operator generally responds with a more or less polite version of, “He’s an idiot.” The point is that there are differences from operator to operator in selecting setpoints. These differences are often so apparent that simple pattern recognition investigations of plant operating data identify when shift changes occur.

Because of the limitations of PID control, scientists and engineers in a wide variety of industries began to investigate alternative control strategies. When the current research effort began, researchers in the mineral processing industry were beginning to investigate artificial intelligence techniques for increasing the efficiency and improving the consistency of process control systems. In one research area they were beginning to investigate the use of expert systems for the control of mineral processing systems (Melema, et al, 1987).

Expert systems are rule-based systems that utilize the if-then production rules used to capture human decision-making abilities. These computer based systems have been used successfully in a number of problem domains (Tanimoto, 1987) including medical diagnosis, mineral exploration, and chemical compound analysis. The success they had achieved made them potential tools for attaining consistent, efficient process control. However, as will be pointed out in this chapter, these expert systems have shortcomings that make them poor vehicles for precise control.

This chapter will provide insight into why expert systems were not used as the basis for the control systems developed by the Bureau of Mines’ research effort into adaptive process control. In keeping with the flavor of the book, this current chapter does not provide theoretical evidence, nor does it present convergence proofs regarding conventional control algorithms. Rather, it presents the qualitative arguments used by the authors in directing the research effort. Later in the book, fuzzy expert systems are introduced as extensions of traditional expert systems, and a link between fuzzy models and fuzzy control systems is forged. Finally, the modeling ability of a traditional expert system is compared with that of a fuzzy expert system where each is used to model a small data set.