ENGINEERING AND CFD

Results and Methods

Two flow configurations have been studied, namely the compression-corner (or ramp) geometry, representative of control surfaces or engine intakes of a hypersonic vehicle, and an Apollo-like capsule geometry. In both cases, surface roughness was modeled at specific locations (upstream of the compression corner and on the shoulder of the capsule, respectively) to study its influence on the downstream evolution of the high-speed boundary layer. The Direct Numerical Simulations are performed with the code NSMB, which allows to make full use of the parallel-computing capabilities of SuperMUC-NG. In the first configuration, the roughness shows a dramatic impact on the ramp-induced Shock-Wave/Boundary-Layer Interaction (SWBLI). The roughness induces the formation of streamwise vortices in its wake, which are convected downstream in the boundary layer. Once the flow separates due to the corner-induced shock and then reattaches on the ramp, the vortices are strongly amplified and generate a mushroom-shaped structure characterized by large velocity gradients (Figure 1, taken from [1]). Underneath this vortical structure, hot streaks appear at the wall, where the surface temperature reaches values much higher than those achieved without roughness. As known from the literature [2,3,4], strong roughness-induced velocity gradients can lead to flow instabilities that can initiate the transition process. Hence, a similar flow evolution can be expected in the reattached flow on the ramp, once freestream disturbances are injected in the baseflow. For the space capsule configuration, our research focused on a restricted domain specifically targeted at the shoulder region of the Apollo-like spacecraft. This region exhibits a high concentration of observed turbulent spots, making it a critical area for analysis. An example of such a restricted domain can be found in Fig. 2 from [5]. It displays a random roughness patch and cross-flow-like vortex formed in the wake region of the roughness-patch. To explore the impact of surface roughness on flow transition, we generated a comprehensive database which consists of over 10,000 Direct Numerical Simulations (DNS) results. Each simulation featured unique, individually randomized roughness patches, crafted using a superposition of three distinct wave-numbers. This approach aimed at modeling surface imperfections, akin to those encountered during the ablation process of an actual re-entry flight. Our analysis delved into the crossflow-like vortices generated downstream of each roughness patch [6]. These vortices play an important role in flow transition, and under-standing their formation process was paramount. Furthermore, we investigated the potential of machine learning (ML) in quantifying the strength of these crossflow-like vortices. We evaluated various ML algorithms, ultimately achieving the most promising results with a multi-layer perceptrons and convolutional neural networks. This integration of ML offers exciting possibilities for future research and real-time monitoring of re-entry conditions. The complete DNS database as well as our ML models can be downloaded and used from our GIT repository [7].