Carl R. Stern

Designing a portable architecture for intelligent particle accelerator control

We present a portable system for intelligent control of particle accelerators. This system is based on a hierarchical distributed architecture. At the lowest level, a physical access layer provides an object-oriented abstraction of the target system. A series of intermediate layers implement general algorithms for control, optimization, data interpretation, and diagnosis. Decision making and planning are organized by knowledge-based components that utilize knowledge acquired from human experts to appropriately direct and configure lower level services. The general nature of the representations and algorithms at lower levels gives this architecture a high degree of potential portability. The knowledge-based decision-making and planning at higher levels gives this system an adaptive capability as well as making it readily configurable to new environments.

Problem Solving as Model Refinement: Towards a Constructivist Epistemology

In recent years the Artificial Intelligence research group at the University of New Mexico have considered several areas of problem solving in interesting and complex domains. These areas have ranged from the low level explorations of a robot tasked to explore, map, and use a new environment to the development of very sophisticated control algorithms for the optimal use of particle beam accelerators. Although the results of our research have been reflected in computer-based problem solvers, such as the robot discovering and mapping out its world, these computational tasks are in many ways similar to expert human performance in similar tasks. In fact, in many important ways our computer-based approach mimics human expert performance in such domains. This paper describes three of these task domains as well as the software algorithms that have achieved competent performances therein. We conclude this paper with some comments on how software models of a domain can elucidate aspects of intellectual performance within that context. Furthermore, we demonstrate how exploratory problem solving along with model refinement algorithms can support a constructivist epistemology.

Tuning and optimization at Brookhaven and Argonne: results of recent experiments

Vista Control Systems Inc. is developing a portable system for intelligent accelerator control. Our system is general purpose and has been designed to be reused at multiple accelerator facilities. This portability arises from the hierarchical object-oriented nature of the architecture. The control system employs a multi-level organization in which knowledge based inferencing is used to dynamically configure a variety of optimization and control algorithms. We discuss results from recent beamline tuning tests at the Brookhaven National Laboratory ATF and the ATLAS facility at Argonne. Results are analyzed along a number of dimensions, including portability, performance as benchmarked against human tuning, adaptive behavior, noise handling, integration of control subsystems, and support for rapid knowledge capture and utilization.

Abductive model refinement for accelerator control

Many aspects of accelerator control require a complex procedure that includes planning, control, and re-evaluation of the process model. As control actions are performed new information is obtained from the system which allows the model to be adjusted. In many cases, observed errors in the model suggest certain control actions for gathering new information used for further refining the model. The process of comparing predicted with observed behavior to produce testable hypotheses for adjusting the predictive model is called abductive model refinement. This paper describes our ideas for applying abductive model refinement to beamline tuning tasks, including minimum steering through a set of quadrupole lenses and developing a waist at a specified location in a beamline.