Critical heat flux for downward flows in vertical round pipes

Abstract

Pressurized thermal shock (PTS) is one of the causes of pressure vessel failure during nuclear reactor accidents. Critical heat flux (CHF) for downward flow is an essential thermohydraulic parameter for predicting PTS using nuclear reactor system codes. While many experimental data have been accumulated and many correlations have been proposed for CHF for upward flow, only a few studies exist on downward flow. This study reviewed the literature that measured CHF for downward flow in round pipes and arranged the proposed correlations. Then, we extracted the parameter values that determine experimental conditions and CHF values from the literature and compared the CHF values predicted from the correlation in each study and the measured CHF values under those conditions across the literature. Each correlation shows relatively good prediction accuracy (average prediction accuracy of 10%–30%) for experimental data from their literature, but the accuracies sometimes decrease for experimental data from other literature. No correlation accurately predicts all the experimental data of the literature, indicating an issue in extrapolating existing correlations. Therefore, we developed a correlation that can accurately predict the experimental data of the collected literature. First, we used a neural network to select the essential dimensionless quantities that comprise the correlation. Then, we regarded the prediction accuracy when all candidate dimensionless quantities extracted from the literature were used for the input variables of the network as the achievable limit prediction accuracy and searched for the minimum combination of dimensionless quantities required to achieve it. The results showed that only the dimensionless mass flux (G*) and the ratio of the heating length to the channel diameter (L/D) are the essential parameters to achieve it. We developed a correlation equation using these two dimensionless quantities and achieved 17.6% of the average prediction accuracy. This result considerably improved existing correlation equations with 25%–40% average prediction accuracy for the same experimental data. Furthermore, unlike many existing correlations, it has excellent practical features, such as applicability over wide parameter ranges and no need to separate cases.