8.25. As the power of computers increased over the years, their use to accomplish so-called “thinking” processes which simulate certain human activities became the subject of a subdiscipline known as artificial intelli­gence (AI). Applications have varied from simple game playing (chess) to speech recognition. Normally required are a bank of stored knowledge and a search procedure in accordance with specified rules that lead to a decision. The boundaries of these efforts are not well defined. For example, many optimization methods follow a similar approach, although optimi­zation is not normally considered an AI procedure. Our initial attention will be focused on “expert systems,” a so-called single-purpose AI pro­cedure that is now widely used commercially [5].

8.26. An expert system is a computer program that incorporates the knowledge and reasoning of human experts as a decision tool. A typical expert system includes a knowledge base, and inference engine, and a system — human interface. The knowledge base includes available relevant infor­mation from data bases as well as various rules-of-thumb facts developed from experience by so-called “heuristic” processes. The inference engine manipulates the knowledge base information by programmed rules of logic to reach the desired conclusions. Translation of input into computer lan­guage and then from the computed results to output is accomplished by the interface feature.

8.27. The major advantage of this type of program is the ability to apply the wisdom of experts systematically in a repeatable manner to the solution of a problem without the experts being present. Prescribed logic paths for decision making are incorporated in the inference engine. “Reasoning” techniques may be forward “chaining” starting from a set of known facts or backward chaining in which the system works backward from goals and tries to establish needed supporting evidence. Common features are “if — then,” “and,” and “or” decision gates. Inprecise or “fuzzy” information may be incorporated into the program through the use of “certainty fac­tors,” which represent judgments of the validity of a fact on a numerical scale of 0 to 1, where 1 means complete certainty.

8.28. In nuclear reactor control rooms, expert systems can provide the operator with guidance regarding measures to be taken in the event of unanticipated incidents, thus reducing operator error. However, the ap­plicability of an expert system is limited by the scope of information in the knowledge base. Therefore, development of expert systems to cover a very wide range of emergency situations is taking place. Other uses include the monitoring of the performance of various plant components and recom­mending operating changes; various training situations, including the analy­sis of trainee performance; and many management-related applications.

8.29. Neural network procedures utilize a type of computer architecture in which the processing elements are interconnected so that a great many calculations can be carried out in parallel, imitating the way that neuron cells in the brain process information. The theory and descriptions of some neural network applications are available in a number of references [6]. In general, neural networks consist of processing units, arranged in layers, with connections between units. These connections have “weights” which have a memory capability and may be “trained” to adjust to changing conditions. Information is processed in a distributed manner rather than by a conventional computer serial approach. Hence a general characteristic is the ability of a network quickly to recognize the various conditions or states of a complex system once it has been suitably “trained.” Thus, applications in pattern recognition, language translation, and speech under­standing are typical.

8.30. Applications to the operation of nuclear power plants are being developed. For example, efficient pattern recognition can be a useful di­agnostic tool in interpreting control room data. Another area is the mod­eling of nonlinear systems with application to process dynamics [7].