| Resumo : |
The critical heat flux (CHF) is the heat flux value in the boiling or cooling process in which the heat transfer decreases, and the heated surface temperature rises rapidly due to factors such as vapor films and bubbles. Due to the difficulties in elaborate experiments and disagreements about measurement and evaluation techniques related to CHF, some methods, like lookup tables, physical correlations, and machine learning, aim to predict the CHF in subcooled flow boiling, which can be related to the operation situation in nuclear power plants. Additive models are regression models that use an additive structure to define the relationship between the input variables and the CHF to make predictions. This study compared four types of additive models on different configurations with consolidated predictive methods in critical heat flux prediction literature. The results showed that the generalized additive models and their quantile version obtained better or closer predictive performance than their traditional counterparts. In addition, this study analyzed tools in generalized additive models to interpret and understand the underlying relations between the inputs and the CHF to obtain its predictions. Furthermore, a brief study of the benefits of quantile generalized additive models when focused on predicting extreme values of the CHF was conducted, showing the possibility of developing safety systems in nuclear power plants. |