Prediction of the heat transfer coefficient and frictional pressure drop for flow boiling of environmentally friendly refrigerants based on an artificial neural network

Abstract

Flow boiling heat transfer can effectively satisfy the heat dissipation requirements of integrated electronic devices. Environmentally friendly refrigerants have become reliable substitutes for traditional refrigerants due to their low ozone depression potential and/or global warming potential. In this paper, a series of experiments on the flow boiling heat transfer coefficient (HTC) and frictional pressure drop (FPD) of R1234yf and R1234ze(E) in a 1.88 mm circular tube were conducted with mass fluxes of 400–870 kg·m^–2·s^–1, heat fluxes of 40–65 kW·m^–2, and saturation pressures of 0.6–0.8 MPa. Artificial neural network (ANN) models were built and trained based on the flow boiling HTC and FPD databases on environmentally friendly refrigerants. ANN models followed the trend of flow boiling HTC and FPD well, with minimum mean absolute deviations (MADs) of both 8.3 %. Different combinations of dimensionless parameters as the input layer of ANN models significantly affected the prediction accuracy. Based on the compiled databases, ANN models were compared with several empirical correlations on two-phase HTC and FPD. The comparison results show that the minimum MADs of ANN models for HTC and FPD databases are 15.2 % and 12.5 %, respectively, which are much smaller than the minimum MADs of the empirical correlations, indicating that ANN models are superior to empirical correlations and can obtain more satisfactory prediction results.