Monitoring and fault detection: a comparative study using computational intelligence techniques

Abstract

In this work, two different Computational Intelligence Techniques were used to provide a comparative study by using Neural Networks and Neuro-fuzzy applied in Monitoring and Fault Detection in sensor of an experimental reactor. Both methodologies were developed and tested using a model composed by 9 variables: N2 (% power), T3 (pool water temperature), T4 (decay tank inlet temperature), T7 (primary loop outlet temperature), T8 (secondary loop inlet temperature), T9 (secondary loop outlet temperature), F1M3 (primary loop flowrate), F2M3 (secondary loop flow rate) and R1M3 (nuclear dose rate). It was used data from the first week experimental reactor IEA-R1 operation from October 2012 in both Monitoring Systems, which was divided in subsets in a following way: 60% for training, 20% for tens and 20% for validation. The results obtained using Neuro-fuzzy were better than the ones with Neural Networks.