Prediction of nuclear power valve faults using sample expansion method, multi-domain feature-optimal screening method and GA-SVM: (Prediction of nuclear power valve faults)

Abstract

Valves are extensively applied in the nuclear power field. They commonly serve under high temperature, radiation, corrosion and other harsh environments for a long time, and once a failure occurs, it will lead to serious accidents, and thus it is of great significance for the prediction of nuclear power valve faults. The typical fault data relevant to nuclear power valves is usually limited, i.e., small sample. In addition, multi-domain features including time domain and frequency domain are commonly employed to predict nuclear power valve faults, in which redundant features may reduce the prediction accuracy. A sample expansion method is first proposed to overcome the difficulty of the small sample, and then a feature-optimal screening method is employed to address the issue relevant to redundant features in multi-domain features. Further, Genetic Algorithm Support Vector Machines (GA-SVM) is employed to predict nuclear power valve faults. The results demonstrate that the proposed method can obtain good prediction accuracy.