Critical heat flux prediction in water-cooled vertical tubes using fuzzy-clustering-assisted AdaBoost and random forest models

Accurate prediction of critical heat flux (CHF) in water-cooled vertical tubes is essential for thermal–hydraulic safety, yet many empirical and mechanistic methods are valid only within restricted operating ranges. This study evaluates whether fuzzy c-means (FCM) clustering improves data-driven CHF prediction using 4,384 experimental observations from vertical round tubes. Six predictors were considered: tube diameter, heated length, pressure, mass flux, inlet subcooling, and exit quality. Elastic Net, k-nearest neighbors, random forests (RF), AdaBoost, FCM-assisted RF, and FCM-assisted AdaBoost were evaluated using a common fixed 70:30 training-test partition. Hyperparameters of the standalone models were optimized by five-fold cross-validation within the training set, whereas the FCM-assisted models used an internal fitting-validation split for joint optimization of the clustering and regression components. FCM membership values were used to decompose the CHF target without including CHF in the clustering inputs. FCM-assisted AdaBoost achieved the best test performance, with a relative root-mean-square error (rRMSE) of 18.1 % and a root-mean-square relative error (RMSRE) of 11.7 %. FCM markedly reduced the RMSRE of RF, from 65.9 % to 16.4 %, whereas the improvement over standalone AdaBoost was smaller (RMSRE: 16.1 % to 11.7 %). These results demonstrate that fuzzy pre-clustering can improve predictive performance for the present heterogeneous CHF dataset, although the physical interpretation of the resulting fuzzy partitions requires further investigation.