Simulation study on nuclear reactor diagnosis by inverse sensitivity analysis

This report presents an approach to the diagnosis of nuclear reactors by means of sensitivity analysis. In a mathematical model of nuclear reactor, an abnormal situation in the reactor would correspond to variations of the parameters embodied in the model. This suggests the possibility of determining the abnormality by estimating the variations of the model parameters from the signals obtained from the actual reactor. A procedure is developed for estimating the parameter variations by means of inverse sensitivity analysis. We define a functional that represents the mean square of the errors existing between the observed disturbance and an estimated disturbance composed of the calculated sensitivity functions and the unknown variations of the parameter or parameters. The amounts of parameter variation are determined from the conditions requisite for minimizing this functional. As a numerical example, a lumped parameter model representing fuel and coolant temperatures of a sodium-cooled fast reactor is used to simulate up to five kinds of malfunctions. Based on this lumped model, sensitivity equations are also derived and sensitivity functions are calculated. Using the above mentioned procedure, the parameter variations are estimated from the change of the coolant temperature and the calculated sensitivity functions. The results are successful. This method possesses the prominent advantage of yielding the quantitative change of each parameter. (author)