Optimization of regression model using principal component regression method in passive system reliability assessment

Abstract

The estimation of reliability of a passive system employed in innovative nuclear reactors has been a challenging assignment due to substantial involvement of natural convection and gravity in its function. This involvement warrants phenomenological failures to cast major influence on probabilistic safety assessment (PSA) of passive system. The phenomenological failures associated with passive systems cannot be evaluated directly by deterministic approach. Probabilistic method deploys thermal-hydraulic codes to evaluate availability of the system and generates simulated response surfaces comprising of large amount of data, which eventually consumes improbable computation time. Regression models can be used to generate large amount of data with limited number of thermal –hydraulic code runs. To make regression models more effective, it is necessary to evaluate effect of each input parameter that goes into them. Isolated condenser system (ICS) is one of the passive systems used in nuclear reactors. The performance of ICS is evaluated by measuring a parameter called integrated power ratio (IPR) with use of regression models. In present study, these regression models are optimized using principal component regression (PCR) method. PCR method eliminates multi-collinearity present within input parameters and thus retains only those parameters which significantly affect the output. Results obtained by PCR method are compared to the outcome of hypothesis testing. The exercise indicates that under given design conditions, at least one input parameter can be omitted from the regression model without any significant effect on IPR. This makes ICS more reliable due to reduction in its probability of failure owing to omission of a redundant input parameter.