Over the past decades, the design of evaporators has experienced improvement due to accruing experimental and numerical research on flow boiling. For micro-finned geometries that are deemed promising in augmentation of thermal performance, several empirical models have been developed. Recently, to attain higher accuracies in modeling, authors also resorted to Artificial Intelligence (AI) techniques while encouraging their further investigations. In this case, a database comprised of 1358 experimental heat transfer coefficients (HTC) and frictional pressure drops per unit length (FPD) has been considered for a holistic assessment of flow boiling modeling methods. The wide range of geometric features within the database, which includes tubes with outside diameters from 3 mm to 7 mm, helps to acquire a more reliable evaluation of the models, since the flow boiling mechanism is strongly dependent on the geometric parameters.After evaluating three empirical models for the HTC, it was confirmed that, depending on the diameter, the flow boiling mechanism undergoes alterations pertinent to a balance between convective and nucleate boiling, and the models must be modified to account for these conditions.The mean average deviation (MAD) for the models was recorded to be 11.7 %, 22.2 %, and 21.5 % for Diani et al., Mehendale, and Tang and Li, respectively. Higher accuracies were obtained after modification of the empirical models, of which the most accurate one provides a MAD of 10.13 %. Moreover, by the implementation of a novel approach, for the first time, a power function correlation has been established among dimensionless parameters for a machine learning-powered model. The MAD of 10.9 % and 15.8 % was reported for the Nusselt number and the two-phase multiplier respectively. Artificial Neural Network (ANN) was also considered for modeling, and MADs of 4.6 % and 4.2 % were recorded for the Nusselt number and two-phase multiplier, respectively.
Comprehensive study of flow boiling modeling inside helical micro-finned tubes: Empirical, non-convex optimization and deep learning predictive models
Diani A.
2024
Abstract
Over the past decades, the design of evaporators has experienced improvement due to accruing experimental and numerical research on flow boiling. For micro-finned geometries that are deemed promising in augmentation of thermal performance, several empirical models have been developed. Recently, to attain higher accuracies in modeling, authors also resorted to Artificial Intelligence (AI) techniques while encouraging their further investigations. In this case, a database comprised of 1358 experimental heat transfer coefficients (HTC) and frictional pressure drops per unit length (FPD) has been considered for a holistic assessment of flow boiling modeling methods. The wide range of geometric features within the database, which includes tubes with outside diameters from 3 mm to 7 mm, helps to acquire a more reliable evaluation of the models, since the flow boiling mechanism is strongly dependent on the geometric parameters.After evaluating three empirical models for the HTC, it was confirmed that, depending on the diameter, the flow boiling mechanism undergoes alterations pertinent to a balance between convective and nucleate boiling, and the models must be modified to account for these conditions.The mean average deviation (MAD) for the models was recorded to be 11.7 %, 22.2 %, and 21.5 % for Diani et al., Mehendale, and Tang and Li, respectively. Higher accuracies were obtained after modification of the empirical models, of which the most accurate one provides a MAD of 10.13 %. Moreover, by the implementation of a novel approach, for the first time, a power function correlation has been established among dimensionless parameters for a machine learning-powered model. The MAD of 10.9 % and 15.8 % was reported for the Nusselt number and the two-phase multiplier respectively. Artificial Neural Network (ANN) was also considered for modeling, and MADs of 4.6 % and 4.2 % were recorded for the Nusselt number and two-phase multiplier, respectively.
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
