Machine learning based fault prediction system for the primary heat transport system of CANDU type pressurized heavy water reactor

In nuclear power reactor, temperature of the core has to be maintained within the safety limits. This can be achieved by monitoring and controlling the various system's parameter of nuclear reactor. One of the important system of nuclear reactor is primary heat transfer (PHT) system. Therefore,...

Full description

Saved in:
Bibliographic Details
Published in2013 International Conference on Open Source Systems and Technologies pp. 68 - 74
Main Authors Ahsan, Syed Nadeem, Hassan, Syed Anwar
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.12.2013
Subjects
Artificial neural networks
Coolants
Data mining
formatting
Fuels
Inductors
insert
style
styling
Temperature measurement
Transient analysis
Online AccessGet full text

Cover

More Information
Summary:In nuclear power reactor, temperature of the core has to be maintained within the safety limits. This can be achieved by monitoring and controlling the various system's parameter of nuclear reactor. One of the important system of nuclear reactor is primary heat transfer (PHT) system. Therefore, any fault in the PHT system may lead to the state where PHT parameters cross the safety limits, and reactor becomes unsafe for operation. To avoid such conditions various fault monitoring and controlling systems have been used in nuclear power reactors. In the recent years, machine learning techniques have been used to build automatic fault prediction system which can be used as a fault monitoring system of nuclear power plant. In this paper, we propose our approach to build machine learning based fault prediction system for the PHT system of CANDU (Canada Deuterium Uranium) type reactor. The proposed approach is based on the classification techniques of supervised machine learning. Whereas, to validate our approach, we performed an experiment by extracting the historical data of the following reactor's parameters: coolant flow rate, coolant header temperature, and neutron power rate. After extracting the parameter's data, we labeled the data with the following three plant statuses: running, transient and shutdown. Finally, we used the binary tree and artificial neural network techniques of machine learning and built models which successfully classified the three statuses of the plant. In our experiment, the maximum obtained accuracy of our model is 99%. It shows that our proposed system can be used to predict fault in PHT loop.
DOI:10.1109/ICOSST.2013.6720608