Predicting neutron source reliability at ISIS Neutron and Muon Source

Abstract

ISIS Neutron and Muon Source is a neutron-source that produces neutrons through a complicated process and provide these neutrons to researchers to study properties of different materials. However, for different reasons, such as unplanned maintenance and failures, this process could be interrupted, causing lost time. The main aim of this project is to predict these interruptions before happening by using previously captured data. The data is cleaned and pre-processed. Then the dataset has been analyzed with statistical methods to uncover patterns and trends. For the prediction part, different ideas have been tested. Two of them based on regression and one is based on classification. Only the classification method has been proven to be practical with 83% accuracy to predict the failures within three hours before happening. Furthermore, an architecture has been proposed to implement a prediction system. A prototype has been developed and presented as a proof-of-concept.