A buoyancy-adaptive reduced-order surrogate for heated rod-bundle transients using gauge-consistent deep orthogonal decomposition

Heated rod-bundle transients couple changes in flow rate, temperature, and buoyancy-driven secondary motion, making them difficult to represent with a fixed global reduced basis. Proper orthogonal decomposition (POD) works well for nearly linear forced-convection hydraulics, but the transverse structures that develop under mixed convection vary with the buoyancy state and require a basis that adapts to the operating condition. This study develops a buoyancy-adaptive reduced-order surrogate for transient rod-bundle thermal-hydraulics using gauge-consistent deep orthogonal decomposition (DOD). The surrogate reconstructs the three velocity components and the temperature field, and is coupled with a coolant-residence-time-based thermal-lag model with a data-calibrated prefactor and a quasi-steady hydraulic closure. It is evaluated against processed steady operating states and transient CFD simulations. Compared with a global POD using the same coefficient mapper, the DOD surrogate gives more accurate reconstructions of the buoyancy-sensitive transverse velocity components. Heated flow-decrease and flow-increase tests show that the surrogate captures the main axial-flow response and temperature evolution under both uniform and skewed heating conditions. These results demonstrate the benefit of buoyancy-adaptive reduced bases for reconstructing nonlinear transverse flow structures in heated rod-bundle transients and establish a practical route for extending real-time rod-bundle digital twins from isothermal hydraulics to heated buoyant transients.