58
Probabilistic
Methods for Model-Based Accelerator Control--Sandiaview Software, Inc., 1009 Bradbury Drive SE,
Suite 7, Albuquerque, NM 87106;
505-245-7400
Dr.
Carl R. Stern, Principal Investigator, stern@sandiaview.com
Dr.
Carl R. Stern, Business Official, stern@sandiaview.com
DOE
Grant No. DE-FG03-01ER83301
Amount:
$99,976
Despite
large prior investments in control hardware and software, particle accelerator
operations still rely to a great extent on human intervention to adapt and
respond to changing conditions, reducing both the efficiency and reliability of
operations. This project will develop probabilistic methods that will extend the
ability of model-based control systems to adapt and respond to changing
conditions. The effort will leverage recent advances in (1) intelligent control
and automated methods for error correction and (2) calibration of an accelerator
model, closing the loop between adaptive modeling and intelligent control. Phase I will design and prototype a system that integrates
intelligent control and adaptive modeling. This system will be designed towards
meeting the requirements for control of a high energy heavy ion accelerator
operations at Argonne National Laboratory. The prototype will demonstrate the
ability to adapt and improve the effectiveness of a control model over time
through the application of probabilistic learning methods.
Commercial Applications And Other Benefits as described by awardee: The application of new technologies in adaptive modeling and intelligent control should yield significant benefits to DOE accelerator facilities in reducing machine startup and tuning time, automating troubleshooting and maintenance, improving performance efficiency, and reducing operational costs and staffing requirements. These automation techniques also should address needs in the industrial control market, significantly enhancing the competitiveness of U.S. industry.