Use of a neural network in an expert system to enhance nuclear power plant safety

Nuclear power plants have many redundant systems and can continue to operate with one or more of these systems disabled. Indeed, it is the standard procedure to remove systems from service to maintain or test them to meet regulatory requirements. Often, there is a time limit for any particular redundant plant component or systems to be out of service. These limits are set by the U.S. Nuclear Regulatory Commission (NRC) as part of the technical specifications or the limiting conditions of operation based on perceived risk to the public. In previous research, the PRISIM computer code was used to calculate the increase in instantaneous core melt probability (called risk factor) when the specified set of components is out of service. The results from PRISIM are used for the expert system to provide advice to plant personnel. The expert system and PRISIM are operated in two separate computers. The values of the risk factor R are calculated in PRISIM. The value R is the ratio of the current probability of core melt risk with the corresponding individual component removed from service to the probability with the whole system operational. A neural network is used to partially replace the large code PRISIM to simplify the expert system. This simplification allows both the expert system and the neural network to operate on one computer. Therefore, the consultation process is much faster than with the previous system