An Artificial neural network assisted approach for developing onset of significant void correlations in subcooled flow boiling

Nguyen, T.-B.; 廣瀬 意育  ; 佐藤 聡 ; 安部 諭  ; 柴本 泰照 ; 大川 富雄*

Nguyen, T.-B.; Hirose, Yoshiyasu; Satou, Akira; Abe, Satoshi; Shibamoto, Yasuteru; Okawa, Tomio*

The onset of significant void (OSV) is a critical parameter in the thermal-hydraulic design and safety analysis of nuclear reactors. However, existing OSV prediction models suffer from a fundamental lack of consensus regarding the underlying trigger mechanism, with various mechanistic models relying on conflicting assumptions such as bubble detachment, thermal-hydrodynamic limits, or global bubble coalescence. This study develops a novel, simplified prediction model by leveraging artificial neural network (ANN) to resolve these uncertainties and identify the most influential physical drivers. Initially, an ANN framework was employed as a robust analytical tool to evaluate the complex interdependencies among various thermal-hydraulic parameters across an extensive experimental database. Through this systematic feature selection, the complexity of OSV prediction was distilled from numerous variables down to only two primary dimensionless parameters that capture the core physics of the phenomenon. The resulting model provides a direct prediction that transcends the limitations of conflicting theoretical assumptions while maintaining high physical consistency. Benchmarking results demonstrate that the proposed model achieves a substantial improvement over established classical correlation. Furthermore, the model exhibits exceptional geometric robustness, maintaining high predictive accuracy across round tubes, annular, and rectangular channels. By providing a highly accurate and physically grounded tool, this research offers a superior alternative for estimating axial void fraction profiles and enhancing safety margins in light water reactors.