Foreword

Dr. Donald A. Stanley was one of my mentors, my business associate, and my close friend. We began this book project together and had hoped to complete it together. Unfortunately, Don succumbed to an illness and died before we could complete the project.

Near the end, Don Stanley graciously asked to have his name removed from the project; he felt strongly that the work was basically mine, and that I should be the sole author. That is the way Don was - unselfish, unassuming, and always striving to do what is fair. Out of respect for Don, I have agreed to his request. And I thought strongly about re-writing the Preface of the book since the Preface was co-authored. However, I believe this would be a mistake. Thus, I have left the Preface unchanged; it appears as Don and I originally wrote it together.

This book is about research conducted at the U.S. Bureau of Mines during my and Don’s tenure there. The research described is not the work of one or even two individuals, rather it was accomplished through the effort, drive, and insightful nature of numerous people - one of whom was Donald A. Stanley. It is my hope that the reader benefits from the efforts of our research group, including the mistakes we made along the way. Don Stanley was a big part of our success.

Donald A. Stanley was extremely bright. When he set his mind to something he was like a bulldog - he just would not let it go until “we got it right.” I owe a lot of my success to the lessons he taught me. I miss my good friend Don dearly.

Charles L. Karr
The University of Alabama

Preface

This book is about intelligent, adaptive process control. It brings together ideas from the field of computational intelligence, a part of the larger field of artificial intelligence, including fuzzy mathematics, genetic algorithms, and neural networks, and uses these ideas to develop a generic architecture for accomplishing adaptive process control. In the development of this architecture, the requisite tools are described and then demonstrated on a number of problems. Moreover, most of the examples provided are of interest in industrial settings (although some simple examples are provided in the beginning so that the reader can focus on technique and not be overburdened with the complexity of the problems being solved).

The focus of the book is on practical applications. It is meant to provide practicing engineers and scientists with the information they need to solve control problems in industry and in academia. Worldwide research is providing a vast body of literature on each of our tools, but most of this information is not needed for solving process control problems. In fact, it is amazing what a small portion of fuzzy logic, genetic algorithm, and neural network theory is required to develop simple, yet extremely effective, intelligent, adaptive systems. If the reader is interested in solving difficult control problems, then this book is a good place to start.

The intended audience for this book is the practicing professional who has a technical background. However, do not let the term “technical background” intimidate you. We do not presuppose an understanding of computational intelligence, control systems, or even calculus. Since our intent is to reach a broad spectrum of people, we present examples from a number of scientific fields. To successfully reach such a diverse audience, the tone of the book is intentionally casual; terseness is almost always sacrificed in the interest of clarity.

Now that we have a rough indication of what this book is, let’s discuss for a moment what it is not. We feel it is important to begin by building a relationship based on trust with the reader. Thus, if this book is not for you, let us save you time and expense: don’t buy it.

What This Book is Not

This book is not a text book on process control theory. There are already many good texts on control theory — linear, nonlinear, and adaptive. Besides, such textbooks require far more high-level mathematics than we need for our purposes. Complex numbers, Nyquist plots, convergence theorems — we cringe at the thought of the time it takes to gain an appreciation (much less an understanding) of these concepts. Besides, we have to wonder: are these concepts really necessary for developing effective, comprehensible software and for attaining excellent control of even the most complex physical systems? Our experience suggests that they are not.

This book is not a textbook on the theoretical analysis of fuzzy logic, nor of genetic algorithms, nor of neural networks. There are a number of excellent books on each of these three techniques, and they are used in college courses across the world. Books such as these provide great detail on many aspects of the various techniques. In fact, in our humble opinion, they cover many more aspects of the techniques than you really need. Although such theoretical texts are certainly necessary, this is not one of them.

This book is not a collection of works by various experts in the field. In fact, there are no experts involved, only us. Edited volumes in which numerous famous folks write about their earthshaking theoretical research efforts are great to have on one’s bookshelf. However, when you have a problem that must be solved and solved quickly, these books are usually the last reference you want to reach for. Such books are often quite thought provoking and useful on that infamous “higher plain,” but they are rarely consulted when the successful solution rubber meets the difficult real-world problem road.

This book is not a collection of case studies in which complex solution techniques are applied to toy problems. There is nothing more frustrating than reading a journal paper in which difficult-to-understand algorithms are applied to trivial problems. More often than not, when the algorithms are extended to problems of any consequence, they fail. Thus, this book presents only simple solution techniques that have been used to solve problems that are complex and typical of problems faced every day in industry.

A Book Based on Experience

This book is based on a seven-year project using computational intelligence techniques for solving mineral processing problems, conducted at the U.S. Bureau of Mines, Tuscaloosa Research Center. It is an approximate chronological recap of the research project. The authors initiated the project, directed the research effort, and were the key players in the successes (and failures) experienced. Because of the Bureau’s interest in industrial cooperation and in technology transfer, we were able to solve a number of difficult problems in and out of the minerals industry. For instance, this book contains example applications from the fields of aerospace engineering, chemical engineering, geology, and mineral processing; areas that superficially are not of direct interest to the Bureau of Mines, but areas that provided a forum for us to discover and test the ideas we were developing for the mineral processing industry.

The authors have very different backgrounds that actually served as a plus over the life of the project. Don Stanley was trained as a mineralogist. He worked for the Bureau of Mines for over 30 years. During his career he supervised projects in mineral surface physics, ceramics, and fine particle technology. He has worked with countless professionals from government, academia, and industry. Thus, Dr. Stanley brought some traditional ideas to the project. Chuck Karr, on the other hand, had never worked in the field of mineral processing when the project began. His education was in the fields of computational intelligence and fluid dynamics; he knew little of the traditional approaches to process control commonly used in industry. Thus, Dr. Karr brought some very nontraditional ideas to the project. Although this strange mix made for interesting and sometimes heated discussions over the years, it was not all bad. For instance, because neither of us had worked with process control problems, we had no preconceived notions of how process control should be done. Thus, we were forced to forge our own way in this area.

This book is, in a sense, a history of the research effort conducted at the Bureau of Mines. When the project began, non-adaptive expert systems, including fuzzy expert systems, were available to solve a number of difficult problems. However, we wondered at the time if such expert systems were all there was to computational intelligence. Asking this simple question began a seven-year journey that has been enlightening, exciting, satisfying, and often confusing. We discovered that the answer to the question, in a word, is no; there is far more to the field of computational intelligence than non-adaptive expert systems.

On this journey, in which we went from a prototypical non-adaptive expert system for phosphate flotation to an architecture for adaptive, intelligent control that has been applied to chemical processing, to a column flotation system, and to helicopter flight—we made some interesting discoveries, ran into some brick walls, and had a lot of fun. If you have aspirations for applying computational intelligence techniques to real-world problems, you will undoubtedly face some tough questions. We hope to help you address the issues you will face by relating to you some of the insights we gained and some of the technology we developed along our path.

The choices we made along the way were not the only ones possible at the time. However, they were, we feel, acceptable because they worked. We do not feel that the path we took to developing adaptive, intelligent control systems is the only one, or even necessarily the best one. We do feel that sharing our results has some potential benefits. Some will read what we did and immediately come to the same conclusions. Others will read what we did and immediately decide that there are better ways. Hopefully, the second class of readers will reciprocate our effort and tell us about techniques that can produce better control systems.

Structure of the Book

The book is divided into four parts, and these parts basically follow the order in which we implemented various techniques along the way.

Part 1 consists of four chapters dedicated to building fuzzy controllers. There is no detailed description of high-level fuzzy operators nor is there an attempt to prove convergence theorems. There is, however, a brief overview of why we chose fuzzy rule-based systems as our main tool for developing control systems. There is also a step-by-step procedure for building fuzzy systems, and this procedure is applied to three different problems of varying complexity.

Part 2 consists of five chapters on genetic algorithms. After a gentle introduction to genetic algorithms, these search algorithms, based on the mechanics of biologic evolution, are used to re-design the control systems presented in Part 1. Also, the importance of developing computer models is introduced. Computer models are needed in the realm of process control, especially in adaptive control, because they allow for rapid, inexpensive testing of alternative control strategies and for the training of adaptive control systems. In fact, the idea of modeling is so important to the systems presented in this book, there is an entire section devoted to it.

Part 3 focuses on developing computer models that can be used in adaptive control systems. This section presents three important topics: (1) genetic algorithms for tuning empirical computer models, (2) neural networks for process modeling, and (3) fuzzy mathematics for process modeling. Each chapter contains examples from industry.

Part 4 ties everything together. Adaptive, intelligent control systems are not problem specific. We will demonstrate their generality by solving problems from a number of different domains. The problems addressed include pH manipulation, chemical processing, column flotation, and helicopter flight control. Additionally, a problem from chaotic dynamics is solved. Since these systems can be large-scale, nonlinear, and coupled, they are frequently difficult to control. Yet, in each example the ideas presented earlier in the book permit the development of effective control systems. Additionally, each application chapter has a little different twist that is specific to the problem being addressed. Each problem has been selected to address an issue we faced over the life of the project.