This review explores the transformative role of artificial intelligence in surrogate modelling for nozzle flow simulations, addressing the computational bottlenecks associated with high-fidelity computational fluid dynamics. Focusing on post-2020 advancements, the paper categorizes key methodologies into physics-based surrogates (such as Kriging and Radial Basis Functions), pure data-driven AI methods (including CNN-LSTM and Gaussian Processes), and AI-enhanced reduced-order modelling, such as Physics-Informed Neural Networks. Prominent applications are examined, ranging from high-precision flow rate prediction in sonic nozzles, achieving root-mean-square errors as low as 0.17%, to thrust optimization in aerospace components and real-time control of cold gas propulsion systems. While specific industrial cases have demonstrated significant speedup factors, significant challenges remain. These include data scarcity for specialized geometries, the difficulty of accurately resolving shock-wave discontinuities, and the inherent “black-box” nature of deep learning models, which complicates uncertainty quantification. The review concludes by identifying critical gaps in real-time industrial deployment and advocating for hybrid architectures that bridge the gap between data-driven flexibility and physical rigour to enhance model reliability in aerospace engineering.
Keywords: Artificial Intelligence, Computational fluid dynamics, aerospace propulsion, flow prediction, nozzle flow simulation, reduced-order modelling, surrogate modelling
