• Citer

Laminar to turbulent transition prediction at hypersonic regimes with metamodels

par Danvin, Florian
Président du jury Coussement, Axel
Promoteur Parente, Alessandro
Co-Promoteur Deconinck, Herman
Publication Non publié, 2021-10-08

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

Accessing space has and will become more and more frequent, especially with the advent of reusable space vehicles. During atmospheric reentry at hypersonic speeds, a spacecraft will undergo severe aero-heating. Boundary layer transition from a laminar to a turbulent state can dramatically increase thermal load, compromise flight stability, justifying the need for thermal protection systems.The ability to predict boundary layer stability is therefore required to design efficient and safe spacecrafts. This problem is currently still addressed in a two-tier manner. Researchers model the intrinsic physics of the problem by using computationally expensive numerical methods, such as DNS, LES, modal or non-modal stability theory coupled with the semi-empirical e^N method. In the engineering community, boundary layer transition at hypersonic regimes is usually accounted for with correlated models coupled to a RANS simulation. The applicability of such correlated models to other geometries and or flow conditions is not always straightforward.The aim of the thesis is to reconcile the research and engineering methodology by proposing a metamodel of modal linear stability theory suitable for hypersonic transition investigations on slender conical spacecrafts.This was achieved by teaching a database of LST computations to a neural network based regression method. Our study reveals how a limited amount of meaningful flow quantities can be sampled and fed to an artificial neural networks to predict Mack’s second mode growth rate. It has been demonstrated that CFD solutions were required to accurately predict the flow stability on moderately blunted cones.Furthermore it has been shown that ensemble methods, such as neural network committees or mixture of experts, can improve the metamodel accuracy and robustness.The proposed metamodel delivers fast and accurate stability predictions on sharp and blunted cones, allowing for cone angle and nose bluntness variations. This model can easily be integrated within a RANS code, and bears no restrictions regarding the turbulence model used. If required, the model applicability can be directly extended to other geometries, flow conditions, or instabilities.