A review of surrogate solving and model closure for multiphysics fields in aero-engine combustors

As one of the critical parts of an aero-engine， the effective organization of complex turbulent combustion within the combustor is essential for enhancing the performance of the combustor， the aero-engine， and even the entire aircraft. To achieve high-fidelity analysis of this turbulent combustion process， methods such as experimental measurements， numerical simulations， and neural networks have been increasingly applied in the field of aero-engine combustion. Neural networks， which effectively integrate multi-fidelity data and physical information， introduce a data-centric fourth paradigm of scientific research， alongside the empirical， theoretical， and simulation-based paradigms. Consequently， neural networks hold the potential to assist traditional experimental measurements and numerical simulations in obtaining higher-fidelity combustion field data at a lower cost. This paper focuses on the challenges and bottlenecks in analyzing combustion reaction flows， introduces representative neural network methods with potential applications in this domain， and reviews the current status and future prospects of neural network applications in combustion and related fields.