Prediction of the heat transfer coefficient during condensation within spirally coiled tubes of various orientations: artificial intelligence modeling and correlation development

• New models for the condensation HTC in spirally coiled tubes were presented. • The intelligent techniques of MLP-NN and ANFIS were used for the modeling. • The accuracy of the literature correlations was examined in details. • The most important factors governing the HTC in spirally coiled tubes were identified. • A simple correlation for predicting the HTC in spirally coiled tubes was developed. Spirally coiled tube heat exchangers are increasingly utilized in refrigeration, air conditioning, and other thermal applications, where accurate prediction of the condensation heat transfer coefficient (HTC) is vital for design optimization. However, traditional correlations are often limited in scope and accuracy. This study developed generalized predictive models for condensation HTC within spirally coiled tubes over a broad spectrum of fluids, operating situations, and geometrical configurations. A database of 563 experimental data points from 10 studies was compiled, and seven dimensionless parameters were selected as model inputs. Two machine learning techniques, Adaptive Neuro-Fuzzy Inference System (ANFIS) and Multilayer Perceptron Neural Network (MLP-NN), were implemented. The MLP-NN achieved the best performance with testing mean absolute percentage error (MAPE) of 4.60% and relative root mean squared error (RRMSE) of 5.78%, while ANFIS yielded MAPE of 6.80%. Rigorous grouped five-fold cross-validation confirmed the generalization ability of both models. In addition, more than 85% of predictions from both models fell within a ± 10% error band. Literature correlations exhibited much higher deviations with MAPE values exceeding 25%. Trend analysis showed that the proposed models correctly reproduced the physical influence of key parameters such as mass flux, vapor quality, coil curvature, and orientation on HTC. The reliability of the dataset and absence of outliers were confirmed using William’s plot. Finally, a new dimensionless correlation was derived, which achieved a reasonable overall MAPE of 14.69% and provided a practical tool for engineering use. The results demonstrated that machine learning algorithms, when combined with physically grounded inputs, offer robust and accurate predictive tools for condensation HTC in spirally coiled heat exchangers, with broad applicability across operating regimes.