A hybrid AI framework for hypersonic flow physics: Bridging experiments and numerical simulations

The characterization of hypersonic flow physics in high-enthalpy regimes is one of the crucial challenges in fluid dynamics, where strong thermochemical non-equilibrium effects, steep gradients, and extreme flow conditions significantly limit both experimental and numerical approaches. Experimental campaigns in facilities such as SPES (Small Planetary Entry Simulator) provide valuable insights into these complex physical phenomena, but are inherently limited by sparse spatial resolution and measurement noise. On the other hand, high-fidelity Computational Fluid Dynamics (CFD) simulations can resolve detailed flow physics, yet their high computational cost severely restricts their use across wide parametric spaces. To address these limitations, this work introduces a hybrid Artificial Intelligence (AI) framework based on Deep Neural Networks (DNNs) for reconstructing key hypersonic flow field quantities from limited and heterogeneous datasets. The approach combines high-fidelity CFD data with experimental measurements obtained at the SPES facility of the University of Naples Federico II, enabling a physically consistent integration between numerical and experimental representations of the flow. Validation on unseen cases demonstrates that the proposed model accurately predicts the main thermochemical quantities, achieving errors within a ± 10% range. These results highlight the potential of physics-driven, data-integrated AI models as efficient tools for analyzing complex hypersonic flow physics, as well as for supporting the design and optimization of future experimental campaigns.