Developing a general purpose intelligent control system for particle accelerators

Tuning and controlling particle accelerators is time consuming and expensive. Inherently nonlinear, this control problem is one to which conventional methods have not satisfactorily been applied; the result is constant and expensive monitoring by human operators. In recent years, and with isolated successes, advanced information technologies such as expert systems and neural networks have been applied to the individual pieces of this problem. Most advanced information technology attempts are also very special purpose and built in a manner not at all generalizable to other accelerator installations. In this paper, we discuss preliminary results of our research combining various methodologies from the field of artificial intelligence into a general control system for accelerator tuning. We consider state space search and adaptive/learning algorithms including fuzzy logic, rule-based reasoning, neural networks, and genetic algorithms. We then propose a framework for applying these methods to a general purpose system for control. Finally, we discuss future plans for extending the system to include parallel distributed reasoning, an enhanced object structure, and additional heuristic control methods.