A data-mechanism-integrated framework for rapid thermo-hydraulic simulation and intelligent design of zigzag printed circuit heat exchangers

To meet the cross-domain and wide-speed operational requirements of hypersonic vehicles, efficient and compact cooling technologies are urgently required. Zigzag printed circuit heat exchangers (PCHE) using Ga-In-Sn alloy and supercritical N₂ as working fluids exhibit superior heat transfer performance, compactness and operational safety, demonstrating strong potential for the propulsion system. However, the dramatic thermophysical-property variations of transcritical fluids coupled with periodic geometric disturbances induced by zigzag channels, result in highly nonlinear thermo-hydraulic behaviors, posing significant challenges to reliable prediction. To overcome these limitations, a data-mechanism-integrated framework is proposed for zigzag PCHEs simulation and design, which systematically unifies physical mechanism analysis with data-driven modeling. A three-dimensional numerical model is first developed, and a micro-element-based analysis is conducted to elucidate local thermo-hydraulic characteristics and evolution mechanisms, from which a physically informed input feature system is established. Artificial neural network (ANN) models are subsequently trained to rapidly and accurately predict the heat transfer coefficients, pressure drops and wall conductive thermal resistance, with explainable analysis employed to achieve data-physics mutual validation. Finally, the ANN models are embedded into a one-dimensional energy conservation solver to establish an AI-driven rapid simulation and intelligent design method, and is demonstrated through a design case. The results indicate that the proposed approach enables compact zigzag PCHE designs under specific thermo-hydraulic constraints, accurately capturing N₂ transcritical effects and geometric disturbances. Compared to three-dimensional simulations, the average streamwise temperature and pressure errors are below 1%, while the design efficiency improves by about 6 orders of magnitude, demonstrating strong engineering applicability.