The plant control system concept for the International Reactor Innovative and Secure (IRIS) will make use of integrated control, diagnostic, and decision modules to provide a highly automated intelligent control capability. The plant control system development approach established for IRIS involves determination and verification of control strategies based on whole-plant simulation; identification of measurement, control, and diagnostic needs; development of an architectural framework in which to integrate an intelligent plant control system; and design of the necessary control and diagnostic elements for implementation and validation. This paper describes key elements of the plant control system development approach established for IRIS and presents some of the strategies and methods investigated to support the desired control capabilities.

An interactive research facility for advanced controls has been developed at Oak Ridge National Laboratory by the Advanced Controls Program. A description of this facility and results of its use to develop and test supervisory control concepts for multimodular advanced liquid metal reactors are presented. 15 refs., 3 figs.

As advanced computing technology becomes part of the control system for power plants, the opportunity arises to address the real goals of plant control. Digital control systems are able to monitor more information and to accomplish more simultaneous tasks than human operators. In future nuclear plants, intelligent supervisory control systems should be responsible for maneuvering the plant in a fashion to minimize the component stress damage. The control system should generate strategies based on traditional operational objectives and on the current plant state and the stress history of various components and transients. In this paper, the authors elaborate on the desirability of including component mechanical stress information in digital control systems. Explicit consideration of stress constraints in the control strategy can significantly reduce the impact of transients on critical components, providing a significant contribution towards meeting current lifetime design goals of approximately 60 years. For illustration, one of the Advanced Liquid Metal Reactor design duty cycles events is discussed from this perspective for three hypothetical response scenarios. 7 refs., 4 figs.

This paper describes the development and implementation of a digital model-based reactivity control system that incorporates a knowledge of the plant physics into the control algorithm to improve system performance. This controller is composed of a model-based module and modified proportional-integral-derivative (PID) module. The model-based module has an estimation component to synthesize unmeasurable process variables that are necessary for the control action computation. These estimated variables, besides being used within the control algorithm, will be used for diagnostic purposes by a supervisory control system under development. The PID module compensates for inaccuracies in model coefficients by supplementing the model-based output with a correction term that eliminates any demand tracking or steady state errors. This control algorithm has been applied to develop controllers for a simulation of liquid metal reactors in a multimodular plant. It has shown its capability to track demands in neutron power much more accurately than conventional controllers, reducing overshoots to almost negligible value while providing a good degree of robustness to unmodeled dynamics. 10 refs., 4 figs.

This paper describes the directions and present status of research in supervisory control for multimodular nuclear plants at ORNL as part of DOE's advanced controls program ACTO. The hierarchical supervisory structure envisioned for a PRISM-like supervisor closest to the process actuators and how it has actually been implemented for demonstration in a network of CPU's is presented next. Two demonstrations of supervisory control with an expert system are also described, one for control of a plant with a single reactor and turbine, the other for control of a plant with three reactors and one turbine. An appendix contains the mathematical basis for the novel approach to large scale system decomposition we have used in the demonstrations of supervisory distributed control of the single reactor plant. 6 refs., 5 figs.

The objective of this study was development of a generalized learning algorithm that can learn to predict a particular feature of a process by observation of a set of representative input examples. The algorithm uses pattern matching and statistical analysis techniques to find a functional relationship between descriptive attributes of the input examples and the feature to be predicted. The algorithm was tested by applying it to a set of examples consisting of performance descriptions for 277 fuel cycles of Oak Ridge National Laboratory's High Flux Isotope Reactor (HFIR). The program learned to predict the critical rod position for the HFIR from core configuration data prior to reactor startup. The functional relationship bases its predictions on initial core reactivity, the number of certain targets placed in the center of the reactor, and the total exposure of the control plates. Twelve characteristic fuel cycle clusters were identified. Nine fuel cycles were diagnosed as having noisy data, and one could not be predicted by the functional relationship. 13 refs., 6 figs.

Good agreement is shown between a centralized and a hierarchical implementation of a controller for a hypothetical nuclear power plant subject to multiple demands. The performance of the hierarchical distributed system in the presence of localized subsystem failures is analyzed. The proposed approach to the hierarchical control of large systems eliminates the need for the typical iterative computations to account for the coupling effects between subsystems. In addition, the computational independence between the unknown terms generated by each subsystem's controller facilitates both its implementation in a distributed network of CPUs and the isolation and diagnosis of sensor and system component failures. The role of the coordinator is transformed from that of an inflexible black-box-like numerical procedure to that of an intelligent supervisor of subsystem performance. By incorporating symbolic and numerical recipes, the supervisor issues the appropriate set of demands for each subsystem required for the specific goal. It is the task of the individual subsystem controllers to fulfil those demands in an optimal fashion

A technique for the hierarchical decomposition of large-scale systems is proposed. It results in a noniterative scheme for near-optimal control of both linear and nonlinear large-scale systems. In this approach, the role of the coordinator module is that of a supervisory controller whose task is to establish the strategy for distributing the overall system demands among each of the subsystems according to their performance and plant status. Each subsystem's controller relays in optimal control algorithms based on uncertain dynamics methods. The solution to each subsystem's optimal control problem is based on Pontryagin's maximum principle and the time-reversal paradigm for free-terminal-time problems. This approach transforms the two-point boundary value problem into an initial value problem which permits integration of both the state and adjoint equations forward in time. For those control problems for which the two-point boundary value formulation can be transformed into an initial value case, this approach will permit online implementation of nonlinear hierarchical controllers for large-scale systems