Adaptive, online models to detect and estimate gross error in SPNDs

Abstract

In-core flux measurement is critical for monitoring and power regulation of a large core nuclear reactor. Self Powered Neutron Detectors (SPNDs) are used for measuring the neutron flux in a nuclear reactor. In this paper we propose an online method for SPND gross error detection, identification and estimation. The method uses linear models which are extracted from data and which adapt continuously in time. This adaption is made possible by use of recursive PCA. However, unlike existing recursive PCA approaches which make approximations to achieve recursion, our proposed approach does not make any approximation. We term our technique as ‘Recursive Principal Component Analysis: Exact Computation’ (RPCA-EC). Use of recursion ensures that the computational requirements of RPCA-EC are low thereby facilitating online implementation. Continuous adaption ensures that the model evolves to adequately capture the time varying relationships amongst the SPNDs. The relationships amongst the SPNDs vary with time due to significant variations in the neutron flux profiles in the reactor with varying operating power levels in the reactor. We apply our proposed method on data taken from an operating nuclear reactor in India and compare results with alternate implementations. Results show that the false alarm rate of our implementation is reasonable thereby indicating that the model adapts to time varying relationships. Performance in presence of gross errors is also satisfactory.