Identification of a line break by a neural network methodology

Abstract

Early failure detection and identification play a fundamental role in the operation of plants for which safety and reliability are of paramount importance. Therefore many efforts have always been, and still are, devoted to the attainment of this goal. An emerging technique, nowadays widely applied in many different scientific and industrial fields, which seems worthwhile exploring is the artificial neural network methodology. In this paper we consider its application in the nuclear reactor field. We explore the feasibility of using an artificial neural network for the quantitative identification of a line break in a simulated auxiliary feedwater system. The system is coupled to four steam generators and the signals consist of thermal-hydraulic data. Here we employ a supervised, feedforward neural network trained by the back-propagation algorithm. The results confirm the validity of the approach. Indeed, not only the method allows to correctly forecast the location of the break, but also it is capable of performing the much more difficult task of estimating the break dimension, with very good accuracy. Moreover the technique appears to be robust against noisy signals.