Comparison of Standalone and Hybrid Machine Learning Models for Prediction of Critical Heat Flux in Vertical Tubes

Critical heat flux (CHF) is an essential parameter that plays a significant role in ensuring the safety and economic efficiency of nuclear power facilities. It imposes design and operational restrictions on nuclear power plants due to safety concerns. Therefore, accurate prediction of CHF using a hybrid framework can assist researchers in optimizing system performance, mitigating risk of equipment failure, and enhancing safety measures. Despite the existence of numerous prediction methods, there remains a lack of agreement regarding the underlying mechanism that gives rise to CHF. Hence, developing a precise and reliable CHF model is a crucial and challenging task. In this study, we proposed a hybrid model based on an artificial neural network (ANN) to improve the prediction accuracy of CHF. Our model leverages the available knowledge from a lookup table (LUT) and then employs ANN to further reduce the gap between actual and predicted outcomes. To develop and assess the accuracy of our model, we compiled a dataset of around 5877 data points from various sources in the literature. This dataset encompasses a diverse range of operating parameters for two-phase flow in vertical tubes. The results of this study demonstrate that the proposed hybrid model performs better than standalone machine learning models such as ANN, random forest, support vector machine, and data-driven lookup tables, with a relative root-mean-square error (rRMSE) of only 9.3%. We also evaluated the performance of the proposed hybrid model using holdout and cross-validation techniques, which demonstrated its robustness. Moreover, the proposed approach offers valuable insights into the significance of various input parameters in predicting CHF. Our proposed system can be utilized as a real-time monitoring tool for predicting extreme conditions in nuclear reactors, ensuring their safe and efficient operation.