Hypersonic flow control: Generalized deep reinforcement learning for hypersonic intake unstart control under uncertainty

The hypersonic unstart phenomenon poses a major challenge to reliable air-breathing propulsion at Mach 5 and above, where strong shock-boundary-layer interactions and rapid pressure fluctuations can destabilize inlet operation. Here, we demonstrate a deep reinforcement learning (DRL)-based active flow control strategy to actively suppress unstart in a canonical two-dimensional hypersonic inlet at Mach 5 and a Reynolds number of 5×106. The in-house computational fluid dynamics solver enables high-fidelity simulations with a fifth-order spectral discontinuous Galerkin scheme and adaptive mesh refinement, resolving key flow features essential for learning physically consistent control policies. The DRL controller, utilizing a Soft Actor-Critic algorithm, robustly stabilizes the inlet over a wide range of back pressures. Notably, a policy trained exclusively at a throttling ratio of 40% (TR40) demonstrates strong zero-shot generalization to previously unseen scenarios, successfully preventing unstart at lower (TR30) and substantially higher (TR50) back-pressure conditions. Furthermore, control remains robust even when the state representation is reduced to a minimal, optimally selected set of 15 pressure sensors subjected to 10% measurement noise. It further generalizes to unseen Reynolds numbers of 10×106 and 15×106 without any retraining. These quantitative results establish a highly resilient, data-driven approach for real-time hypersonic flow control under realistic operational uncertainties, moving beyond traditional linear controllers by exploiting non-intuitive, multiscale aerodynamic interactions.