A Study on Automatic Diagnosis of Abnormality in Nuclear Reactor Systems

抄録

This paper deals with a method to detect the so-called abnormal phenomena during nuclear operations. The new method consists of the following ideas; an index defined as the sum of ratios of the square of a factor score to the contribution weight of the factor, which is evaluated by applying the multi-factor analysis technique to the data of the state of nuclear reactor systems like neutron flux, temperature, flow rate and so on. The excess of the index over some given threshold shows the reactor system would be in an abnormal state. Then a theory of optimal filtering by Kalman with aid of the stochastic approximation is applied to estimate the neutron flux distribution at its abnormal state and subsequently the squared sum of difference between desirable and estimated flux distributions shows the spot at which the abnormal phenomena would have occurred in terms of the peak of its distribution.