A comprehensive review on numerical approaches to simulate heat transfer of turbulent supercritical CO2 flows

Extensive computational investigations have been performed to obtain more detailed information about the peculiar phenomena of turbulent supercritical carbon dioxide (sCO(2)) flow as ideal heat transfer fluids in various thermal engineering applications. This paper reviews the simulation techniques used and discusses their advantages, shortcomings and applicability. Not only is a comprehensive inspection on various computational approaches provided, but also the model refinements are suggested. Direct Numerical Simulations (DNS) provides valuable and reliable information about the thermohydraulics of turbulent sCO(2) flows, in particular within the near-wall region, which well interprets the observed heat transfer enhancement and deterioration with property variations, flow acceleration and buoyancy discussed. However, DNS is not feasible when it comes to high Reynolds number flows with complex geometries encountered in practical applications because of the drastically increasing computational cost. Reynolds-Averaged Navier-Stokes (RANS) modeling is able to fill the gap with acceptable accuracy and becomes the mainstream for turbulent sCO(2) heat transfer simulations. The flow and heat transfer behaviors of turbulent sCO(2) can be simulated using RANS modeling leading to acceptable predictions. However, the performance variation is considerable for different models and for the same model of changing operating conditions, model generality is not reached. In addition, some treatments implemented into the RANS models for constant property fluids are not appropriate for variable-property sCO(2) flows, causing inconsistency on the mixed convection predictions. Variable turbulent Prandtl number and more advanced calculation schemes for buoyancy production of turbulent kinetic energy are strongly recommended. Also, more appropriate treatments for damping functions are demanded to enable the model properly respond to the local property changes, particularly near the wall. Much simpler models with far less computational cost based upon the two-layer theory are being developed to achieve the generality. While this is promising, the examinations are still limited to the certain conditions and some model parameters need to be calibrated against the DNS data, which definitely reduces the model universality since DNS only covers a limited range of operating conditions. Developing more generic and reliable RANS models is still the main focus of simulation techniques used for turbulent sCO(2) heat transfer.