Diagnosis of vessel failure provides for operators and TSC personnel very important information to manage the severe accident in nuclear power plant. However, operators can not diagnose the reactor vessel failure by watching the temporal trends of some parameters because they never have experienced the severe accident. Therefore, this study proposes a method on the diagnosis of the PWR vessel failure using a Spatiotemporal Neural Network (STN).
STNs can deal directly with both the spatial and the temporal aspects of input signals and can well identify a time-varying problem. The target patterns are generated from MAAP code. Vessel failure diagnosis has been performed for 8 accidents and the developed STNs have been verified for untrained three severe accidents. STNs identifies the vessel failure time and the initiating events. For example, when large break LOCA (break size = 0.16 ㎡) is used for input accident scenario, only the output value for the target pattern of LBLOCA is activated greater than the threshold value near the real vessel failure.
To validate vessel failure diagnosis system and to train severe accident to operators, extensive severe accident simulator is to be an absolute necessity. Therefore, a simplified severe accident simulator, SIMAAP (severe accident Simulator based on MAAP), has been developed. SIMAAP simulates the various severe accident progress through on-line communication with MAAP.