A novel method of synergistic multi-scale neural operator architecture for accurate aerothermodynamic prediction of thermal protection systems at high speeds

Accurate and efficient prediction of aerodynamic heating over complex configurations along practical flight trajectories is a crucial prerequisite for the refined design of thermal protection systems in high-Mach-number vehicles. However, conventional computational fluid dynamics (CFD) simulations often result in extremely high computational costs and low efficiency when applied to complex vehicle configurations across the entire mission profile or full lifecycle. To enhance the performance of full-configuration coupled field prediction for complex geometries under diverse service conditions, this paper proposes a synergistic multi-scale neural operator architecture, called SynoPhys, an end-to-end neural partial differential equation (PDE) solver built upon a Transformer architecture with adaptive fusion and multi-scale clustering. By integrating multi-scale geometric encoding with an adaptive fusion mechanism, this framework effectively captures both global and local flow-field characteristics. It learns the nonlinear mapping from flight parameters to the full-field heat flux, eliminating the need for manual meshing or intermediate preprocessing. The proposed model has been evaluated on four representative configurations: blunt-cone, double-cone, double-ellipsoid, and lifting-body. It demonstrates high-fidelity generalization, achieving mean relative errors of 1.13%, 1.00%, 0.97%, and 0.47%, respectively. The model performs well in high-gradient stagnation regions, maintaining relative errors below 1% compared to CFD benchmarks. These results confirm that SynoPhys offers a unified and efficient framework for rapid and accurate aerodynamic heating prediction across diverse geometries and flight conditions.