New void fraction equations for two-phase flow in helical heat exchangers using artificial neural networks

In this research, two new empirical equations based on Artificial Neural Network (ANN) were developed to determine the new void fraction in two-phase flow inside helical vertical coils with water as work fluid. The first model included vapor fraction (xg), density ratio ρgρl, viscosity ratio μlμg, and curvature ratio dD, as input variables, and 2 neurons in the hidden layer to predict satisfactorily the void fraction. In order to simplify the model, a second model of ANN was proposed without curvature ratio. The best architecture to the second model, with 3 input variables, was also with 2 neurons in the hidden layer. The coefficients of determination were R2 > 0.9 to both models. The ANN models of void fraction satisfied the interval condition of 0-1. Therefore, both models have been considered to be satisfactory for predicting the behavior of void fraction of a two-phase flow. To validate these new void fraction equations, three different helical heat exchangers described in previous works reported, were applied in two ways: first, experimental and simulated heat fluxes were compared using steady state test data from two helical double-pipe vertical evaporators integrated into two absorption heat transformers; second, experimental and simulated heat fluxes were also compared in an innovative design prototype full-scale helically coil steam generator in which, numerical results for pressure along the tube reveal a better way to represent the two-phase flow. The second evaluation also provided evidence on the successful extrapolation of simple ANN equations of void fraction in function of dimensionless numbers. The analyses of the contribution of input variables in the ANN model showed that the curvature ratio could not impact the simulative accuracy of void fraction under the experimental conditions worked.