Modeling flow boiling heat transfer of pure fluids through artificial neural networks

The heat transfer modeling of flow boiling inside horizontal tubes at saturation conditions is studied. The conventional models in the literature tend to modify an elementary historical correlation trying to empirically follow the heat flux trends as evidenced by the experimental data. The problem basically constitutes a highly non-linear phenomenon for which it is advisable to use a very flexible function approximator to draw case-specific heat transfer coefficient correlations from the experimental data alone. Artificial neural networks (ANN) were here used for this purpose, proving able to precisely represent conventional heat transfer surfaces. The most synthetic correlation architecture, based on the directly-accessible physical quantities controlling the phenomenon as independent variables, was considered for the ANN function. Eight pure fluids and a constant composition ternary mixture were studied and for each of them an individual heat transfer equation was obtained in the same ranges of the respective data set. With respect to data these ANN equations reach individual absolute average deviations of few parts per cent with biases virtually null. The obtained accuracy is lower than the claimed experimental uncertainties, but the heuristic technique also evidenced for the data both a frequent incoherence among the sets and a rather questionable reliability posing serious limits to the drawing of precise fluid specific heat transfer surfaces. A great improvement of performance with respect to five conventional correlations was found for the present modeling. (C) 2005 Elsevier SAS. All rights reserved.