| Computerization and automation of industrial and research institutions presents both a need and an opportunity for automatic control of complex processes. Past research in automatic control has concentrated primarily on analyzing the "black box" of a single-input, single-output (SISO) system. Techniques which apply to these systems are of limited value in more complex domains where multiple inputs and multiple outputs (MIMO) cannot be decomposed into a collection of independent SISO systems. Control of particle accelerator beam lines is one such domain. With a number of procedural, goal-oriented, and knowledge intensive activities required to tune accelerator beam lines, conventional control methods have not been successfully applied. This dissertation is a description of my thoughts and work in building an automatic control system for particle accelerators.; In order to successfully tune an accelerator beam line, automatic control must embrace a methodology different from that of conventional control systems. Rather than computing optimal control actions based on system state as in typical "fast control" systems, an intelligent system must emulate human reasoning processes in order to guide the beam line to a desired state. Good tuning can be accomplished through a combination of heuristic, search, and rule of thumb methods. Very high quality tunes, however, are usually only produced by physicists trained in accelerator theory, accelerator design, and the day-to-day operation of the accelerator facility. In either case, skills which have typically been easy for people but hard for machines are also required. These include pattern recognition, noise handling, diagnosis, and learning.; The control architecture presented here was designed with all of these problems in mind, as well as other problems which arise in control of parallel systems. The architecture is intended as a framework for assembling established techniques into a single system which builds on the strengths of the control methods used and makes up for some of their weaknesses. It is also intended to assist in building large control systems which need coordinated control of mostly local processes. The architecture includes components for interacting with the physical system, components for incorporating conventional control, components for incorporating knowledge-based methods, and structures which allow these components to work together in a distributed environment. |
