Variation in system dynamics and modeling uncertainty (due to unmodeled system behavior and/or presence of disturbances), have posed significant challenges to the effective luminosity and orbit control in accelerators used in nuclear physics research. Adaptive control has long been pursued as a possible solution, but difficulties with online model identification, and robust implementation of the adaptive control algorithms has prevented their widespread application. In addition, the performance of the control system is contingent on the responsiveness of the control algorithm to the inevitable deviations of the model from the actual system. This project will use neural networks to detect significant changes in system behavior and develop the methodology for online identification of new empirical models. Furthermore, an optimal model-predictive-based adaptive control algorithm will be developed, which enables the robust implementation of an effective control strategy. In Phase I, simulations were conducted to clearly demonstrate the feasibility and benefits of implementing model predictive control technology in accelerator control problems. In addition, a prototype for the optimal model-predictive-based adaptive control algorithm was developed for a well-known nonlinear temperature control problem for gas-phase reactors. In Phase II, a classification algorithm for dynamic data will be developed to enable the detection of significant changes in system behavior. Algorithms for efficient handling of variable dynamics in the nonlinear model predictive control system will be developed, and the machinery that allows the implementation of the optimal adaptive schemes will be put in place. Prototypes to implement the above-mentioned features in commercially available software will be developed.
Commercial Applications and Other Benefits as described by the awardee: The online system identification and optimal model-predictive-based adaptive control software should have applicability in process industries, power systems, and financial systems. In particular, the day-to-day operation of accelerators should immediately benefit from the findings in this project.