Numerical study and machine learning on local flow and heat transfer characteristics of supercritical carbon dioxide mixtures in a sinusoidal wavy channel PCHE

The ambient temperature has a great influence on the practical application of the supercritical carbon dioxide(SCO2) Brayton cycle. The introduction of mixtures is an effective way to change the critical characteristics of the working fluid, which allows the system to better match the ambient temperature. The mixtures not only affect the overall performance of the power system, but also affect the flow and heat transfer processes in various heat exchange equipment. In this investigation, the physical model of a sinusoidal wavy channel printed circuit heat exchangers (PCHE) is built, and numerical simulations are conducted firstly on the flow and heat transfer characteristics of three S-CO2 mixtures (CO2-He, CO2-Ke and CO2-Xe). Effects of the mole fraction and type of the additive gasses on the thermal-hydraulic performance are discussed, as well as the variations of the local heat transfer coefficients and pressure drop along the flow direction. It's found that transfer coefficient and pressure drop of CO2-He increase with the increase of the additive gas mole fraction n, with maximum heat increases of 396 % and 1147 % against pure S-CO2, respectively. With the increase of n, heat transfer coefficient and pressure drop of CO2-Kr and CO2-Xe decrease. Maximum decreases of 71 % and 38 % can be seen for the heat transfer coefficient and pressure drop against pure CO2 for CO2-Kr, and 80 % and 64 % against pure CO2 for CO2-Xe, respectively. CO2-Xe yields the best thermal-hydraulic performance of the PCHE among the three mixtures. Secondly, machine learning is adopted to predict the local flow and heat transfer characteristics of S-CO2 mixtures in response to the significant variation of S-CO2 mixtures along the flow direction within the PCHE. Four machine learning models including Support vector machine (SVR), Artificial neural network (ANN), Random forest (RF) and Extreme Boosting Tree (XGBoost) are used and the corresponding prediction performance are analysed. It is found that machine learning is efficient and accurate in the prediction. Among the four machine learning models, XGBoost has a strong fitting ability to the local Nu and f, with an R2 of 0.9992 and 0.9718 on the test set, respectively. The prediction results of the XGBoost model can reflect the variations of local Nusselt number and friction factor along the flow direction well. However, ANN has stronger generalization ability in the prediction under new working conditions. The use of machine learning can greatly help the design and optimization of PCHEs with S-CO2 mixtures as working fluids.