Answers are grounded in this paper and cite their sources.
Supercritical carbon dioxide (sCO 2 ) has garnered significant interest in advanced thermal power cycles due to its high efficiency and compact system design. Nonetheless, under heating conditions, sCO 2 displays highly intricate thermo-fluid behavior, especially in the proximate wall region. As the fluid traverses the Widom line, abrupt variations in thermophysical properties induce pseudo-boiling and other transport pseudo-phenomena, thereby markedly modifying turbulence structures and heat transfer mechanisms. Precise modeling of these effects necessitates fine near-wall resolution and sophisticated turbulence models, which consequently elevate computational costs. For large-scale systems, high-fidelity simulations become time-consuming and often impractical. To address these limitations, the present study develops a supervised machine learning (ML) framework aimed at predicting near-wall characteristics of heated sCO 2 flow within horizontal tubes. A comprehensive numerical dataset was generated employing an Eulerian approach coupled with the Shear Stress Transport ( 𝑆 𝑆 𝑇 ) 𝑘 − 𝜔 turbulence model, encompassing heat fluxes of 25-100 kW/m 2 , mass fluxes of 200-1000 kg/(m 2 .s), pressures of 7.5-20 MPa, and tube inner diameters of 4-12 mm. Various learning algorithms, loss functions, and optimizers were systematically evaluated to identify the optimal model architecture. The results demonstrate that the proposed modified deep residual ANN, incorporating residual connections, SiLU activation, a combined Huber–MAE loss function, and AdamW optimization, achieves significantly improved statistical and physical performance compared to the other ML models. The model attains a coefficient of determination (R 2 ) exceeding 0.97 for both upper and lower wall temperatures, while accurately capturing localized thermal peaks associated with heat transfer deterioration in horizontal tubes with sCO 2 .
No takes yet. Share an insight, caveat, or question.
Dolatabadi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: