Atsuo Kohsaka

Development of Reactor Accident Diagnostic System DISKET Using Knowledge Engineering Technique

An accident diagnostic system DISKET has been developed to identify the cause and the type of an abnormal transient of a nuclear power plant. The system is based on the knowledge engineering (KE) and consists of an inference engine IERIAS and a knowledge base. The main features of DISKET are the following: 1. Time-varying characteristics of transients can be treated. 2. Knowledge base can be divided into several knowledge units to handle a lot of rules effectively. 3. Programming language UTILISP, which is a dialect of LISP, is used to manipulate symbolic data effectively. For the verification of DISKET, performance tests have been conducted for several types of accidents. The knowledge base used in the tests was generated from the data of various types of transients produced by a PWR plant simulator. The results of verification studies showed a good applicability of DISKET to reactor accident diagnosis.

An Approach to Build a Knowledge Base for Reactor Accident Diagnostic Expert System

In the development of a rule based expert system, one of key issues is how to acquire knowledge and to build knowledge base (KB). On building the KB of DISKET 1),2), which is an expert system for nuclear reactor accident diagnosis developed in JAERI, several problems have been experienced as follows. To write rules is a time consuming task, and it is difficult to keep the objectivity and consistency of rules as the number of rules increase. Further, certainty factors (CFs) must be often determined according to engineering judgement, i.e. empirically or intuitively. A systematic approach was attempted to handle these difficulties and to build an objective KB efficiently.

RELAP4 Modeling for Reverse Flow Property of BWR Jet Pump under Large Break LOCA Condition

It is very important to identify the reverse loss coefficient of BWR jet pump in the evaluation of core inlet flow at the beginning phase of BWR LBLOCA (Large Break Loss-of-Coolant Accident) analyses. Hence, the reverse flow property of jet pump was investigated in relation between the momentum equation, pressure loss coefficient and RELAP4 noding, and a new modeling has been proposed. In the proposed modeling, an equivalent pressure loss coefficient is used to take into account of the effect of accellerating pressure loss by the continuous flow area reduction from the tale pipe to the throat. The effectiveness of this model was studied by analyses for the LOFT 1/6 scale jet pump experiment and typical BWR LBLOCA. It has been, consequently, shown that this proposed model gives better jet pump property than a previous model which is used in the WREM sample problem and which gives very conservative result in core inlet flow and in the peak cladding temperature through whole transient.