This study presents a new approach to flow regime classification specifically tailored for typical wire-wrapped fuel assemblies in sodium fast reactors. Historically, the definition and understanding of flow regime boundaries have been extensively researched. However, many of these models suffer inaccuracy due to a lack of comprehensive data. In particular, the limited data, with only 36 data points for the laminar-to-transition boundary and 145 data points for the transition-to-turbulent boundary, often result in suboptimal models.

Recognizing the critical data gap, this study classified flow regimes based on a robust data set of over 5000 data points. A diverse range of algorithms was used to find the optimal classification model. These included logistic regression, artificial neural networks, support vector classifiers, Naïve Bayes, Gaussian Naïve Bayes, K-Nearest Neighbors, random forest, AdaBoost, GradientBoost, and XGBoost. A comparative analysis of these algorithms provides valuable insights.

This study presents a comprehensive set of machine learning algorithms to improve the accuracy and reliability of flow regime classification, which is a critical step in predicting friction factors and the efficient operation of sodium fast reactors.