INTELLIGENT CONTROL



What we are doing...

This project is a collaborative effort betweenVista Control Systems, Inc. of Los Alamos, NM and researchers at UNM. The goal of this project is to create a portable, intelligent control system that successfully tunes the beam of a linear particle accelerator. To accomplish this task, we have built a large scale, distributed control system that simulates the actions of a human operator to steer, focus, and adjust the components of an accelerator beamline. The system is organized as a hierarchy of program elements that dynamically coordinate and delegate control actions. These program elements, known as controllers and solvers, incorporate knowledge-based control, genetic algorithms, neural networks, and other techniques from artificial intelligence.

At the lowest level of the hierarchy there is a Physical Access Layer (PAL) that abstracts the elements of a particular beamline into object models. These abstractions allow the control system to use high level concepts to manipulate beamline control elements. The PAL is built on an efficient software data bus called Vsystem, developed by Vista. Abstractions in the PAL combine Vsystem data channels into objects that encapsulate the functionality of a typical control element, such as a magnet or monitor. The abstraction and encapsulation of individual beamline elements allows our system to be easily ported between different accelerator facilities.

As a high level control mechanism, we use a planning technique known as Teleo-Reactive programming developed by Nils Nilsson at Stanford. Teleo-reactive programming combines the responsiveness of an analog feedback loop, typical of classic control theory, and the goal-oriented behavior of a production system, to allow our control system to govern the execution of its tasks in a dynamic environment. TR mechanisms enable the system to either take advantage of felicitous events in the environment, or to appropriately fall back in its plan to a previous state when outside events interfere with the proper execution of a plan.

Our system remains under development, but has been successfully tested at both the Brookhaven National Laboratory ATF and at the ATLAS facility at Argonne National Laboratory.


PUBLICATIONS


[Top][Back][Home]