A novel machine-learning schemes to predict heat transfer coefficient during condensation of CO2 in porous media

Accurate heat transfer predictions during CO2 condensation in porous media are needed to improve two-phase flow devices like refrigeration and air-conditioning systems. However, phase-change condensation system behavior makes predicting of the heat transfer coefficient very difficult. A reference dataset is used by gathering information from previous experimental investigations on heat transfer during condensation of CO2 as refrigerant in porous media at subcritical pressures. This database comprises four key factors: porosities of 39.8–44.5%, mass flow rates of CO2 8 × 10–5–16 × 10–5 kg⋅s−1, input pressures 3450–4300 kPa, and volume flow rates of 3–6 L⋅min−1. In the current study, four machine learning methods—two neural network models and two enhanced adaptive neural fuzzy inference systems—are developed for the prediction of internal heat transfer coefficient. The models are applied as machine learning regression models. For validating proposed models, the internal heat transfer coefficient is predicted and then examined against the corresponding actual values. The output data showed acceptable results for predicting and the proposed models successfully decreased the level of uncertainty in predicting the porous condenser’s internal heat transfer coefficient. The error analysis for the proposed methods is reported and discussed by using four statistical criteria. The BPNN, ANFIS-ANN, ANFIS, and ANN-3 models had RMSEs of 2.2036, 2.4413, 3.5532, and 2.7471, respectively.