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. Significant successes of this work are reported in [1,2].
DESIGNING AN ARCHITECTURETo be most useful, control architectures should be flexible enough to incorporate the important tools necessary for robust control, and should include design features which support the application of those tools in a timely and transparent manner. Our control architecture attempts to meet the following six requirements:1. The use of conventional control techniques where appropriate. Conventional control is the best solution for a large class of problems, and consists of established, well developed, robust techniques for dealing with linear and approximately linear small systems.
The use of high-level control knowledge from experts in the field.More complex control systems include highlevel knowledge and information obtained through knowledge-engineering sessions with a human expert. This knowledge is preserved in the form of symbolic data sets (rules, relations, objects, etc.) and is manipulated through sophisticated, high-level reasoning mechanisms. This knowledge-based component encodes internal control information about how and when to use classes of control algorithms and heuristics and for storing configuration information for PID, neural network, fuzzy control, etc.
The use of supervisory control for dealing with macro state transitions.For instance, in beam line tuning, if a failure in an upstream monitor forces the use of stripline data in judging beam intensity and distribution, a downstream controller may need to use a control algorithm which is less precise, faster, and less sensitive to intermittent failure or noise. Supervisory control can also be used for control over different internal control subsystems.
Support for real time reactive control.In this case, "real time" means the ability to compute and perform "satisfactory" or "good enough" decisions that are not delayed due to control system response.
The control of complex systems through the use of a hierarchical distributed architecture.To perform intelligent control that optimizes behavior in complex systems, a control architecture should support the ability to coordinate the individual partitions of the...