Machine learning prediction of critical heat flux on downward facing surfaces

Abstract

Listen

Translate article

Application of external reactor vessel cooling (ERVC) in-vessel retention (IVR) was an important measure to ensure the integrity of the lower head of a reactor pressure vessel (RPV). As a typical boiling heat transfer process, the critical heat flux (CHF) plays a crucial role in the safety margin of a RPV's IVR-ERVC strategy. Although there have been a lot of correlations and experiments about the CHF of pool boiling on downward facing surfaces, their application range was relatively limited. To further expand the usability, machine learning was introduced in this research after collecting most accessible CHF data of pool boiling on downward facing surfaces. Considering the small amount of these experimental data, some pseudo data obtained by fitting the existed experimental data were added. Three machine learning methods, the ε-support vector machine (ε-SVM), back propagation neural network (BPNN) and random forest were used to predict CHF. Among the three methods, ε-SVM provided the best accuracy in the prediction. The effects of orientation, surface dimensions, heating surface material and pressure on the downward facing surface boiling crisis were predicted by ε-SVM, the results showed that in general, the CHF increased with the increase of orientation angle, and increased with the pressure from 1 bar to 10 bar. Under atmospheric pressure, the CHF decreased with increasing the width of the heating surface, but there was a width value that could eliminate the influence of width. In addition, the CHF seemed to increase with the increase of thermal effusivity. However, there are still some inexplicable phenomena that need to be revealed by further research. Overall, this method is expected to be widely used in predicting the CHF of pool boiling on downward facing surfaces.