A new insight into two-phase flow pressure-drop estimation and optimization of the refrigerant R1234yf

The overall aim of this study is to predict and optimize the two-phase pressure drop (2φΔP) of a refrigerant R1234yf for a wide range of testing conditions. At first, feature engineering was carried out to choose the most impactful features to estimate the considered output. The correlation matrix analysis revealed the heat flux (Q), mass flux (G), saturation temperature (T_sat), quality (x), and hydraulic diameter (Dh) as the most influential features. Based on the Bayesian surrogate (Gaussian process) model, an optimal set of hyper-parameters was found and employed to develop a deep neural network (DNN) with a structure of 12–271 × 9–1 for predicting the ΔP (pressure drop). Finally, genetic algorithm (GA) was used to optimize the Q, G, T_sat, x, and Dh for ΔP reduction. The developed DNN yielded an accuracy of 0.995 in terms of R² (correlation coefficient). A minimum ΔP (162.22 Pa/m) was achieved by setting the Dh, G, Q, x, and T_sat equal to 10.84 mm, 110 kg/m²s, 1.59 × 10^-4 kW/m², 0.052, and 19.96 ℃, respectively. This methodology can be helpful in design and optimization of many engineering systems concerning the 2φΔP of different refrigerants.