Prediction of Cooling Heat Transfer Coefficient of Supercritical CO2 With Small Amount of Entraned Lubricating Oil Entrained by Neural Network Method

The modeling of the heat transfer performance of supercritical carbon dioxide with a small amount of entrained PAG-type lubricating oil is presented in this paper. Due to the complexity of the heat transfer mechanism, including the complexity of the changes in the thermophysical properties, changes in the solubility of CO2 in oil, and the flow pattern at different temperatures and pressures, a neural network method based on a large amount of experimental data is used to construct a semiempirical prediction model. The proposed approach involves a feedforward threelayer neural network with the tube diameter, Prandtl number, Reynolds number, heat flux, thermal conductivity, and oil concentration as the input parameters and the heat transfer coefficient as the output parameter. The number of neurons in the hidden layer is 16, and all the weight matrices are determined by employing the backpropagation algorithm using a total of 1313 experimental data elements. The experimental data used corresponds to a large number of experimental conditions with the following variations: tube diameter from 1 to 6 mm, oil concentration from 0% to 5%, pressure from 8 to 10 MPa, mass flux from 200 to 1200 kg/ms, and heat flux from 12 to 24 kW/m. The proposed model is found to agree well with the experimental results, with a deviation within ±20% for 87.3% of the valid data; 96.1% (1262/1313) of the valid data are found to be within a deviation of ±30%. In addition, the prediction results under different pressures, temperatures, and mass flux conditions are also presented.