Data-driven prediction of second-mode instability in non-similar hypersonic flows: a comparative study
Year: 2026
Type: article
Abstract: Accurate prediction of the laminar–turbulent transition is critical for the design of hypersonic vehicles. However, established physics-based methods, such as linear stability theory (LST), are computationally intensive. The development of efficient surrogate models to overcome these limitations is particularly challenging for blunt-nosed vehicles. Strong non-similar flow effects caused by the entropy layer render simplified profile parameterizations inadequate in the case of these vehicles. To address this, the present paper proposes a methodology for creating surrogate models trained on a comprehensive database of laminar Navier–Stokes simulations and LST computations for a 7 $$^\circ $$ ∘ blunt cone. A pivotal element is the utilization of the boundary layer impedance inflection point for feature extraction, which provides a compact parameterization of the acoustic waveguide and reduces the reliance on boundary layer edge quantities, which are ambiguous in the presence of an entropy layer. Two modeling frameworks are employed and compared in this study: radial basis function (RBF) interpolation and eXtreme gradient boosting (XGBoost). Both models agree closely with the LST benchmark at the freestream conditions of the MF-1 flight experiment. In the implementations examined here, XGBoost can be trained on the complete high-resolution database, whereas the partitioned RBF approach requires downsampling. Feature analysis and a wall temperature study show that the XGBoost model relies strongly on the wall cooling ratio and reproduces the destabilizing effect of wall cooling obtained with LST. Quantitatively, the XGBoost surrogate reduces the time to solution of the instability analysis by a factor of about 600 compared to direct LST, reproducing the benchmark N-factor envelopes with a root-mean-square error (RMSE) of less than 0.06.
Source: CEAS Space Journal
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,390
OpenAlex ID: W7237359883
