Generative Machine-Learning Methods to Predict the Wake of a Distributed Roughness Patch in a Hypersonic Boundary-Layer Flow

This study presents a UNet-based conditional generative adversarial network (cGAN) to predict the flow field in the wake of random roughness patches. The model successfully captured the characteristic cross-flow-like vortex, but its fidelity needs to be improved for detailed transition calculations. However, the model proved effective as a screening tool for identifying roughness patches likely to induce or suppress cross-flow-like vortices. By estimating vortex size and location, the model aids in identifying stability-critical vortices. Two roughness patches with a strong and weak vortex were selected based on the model’s predictions and subjected to unsteady Direct Numerical Simulations (DNS). The interaction between the cross-flow-like vortex and unsteady perturbations leads to amplification in regions of high shear-flow and vortex breakdown. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.     «

This study presents a UNet-based conditional generative adversarial network (cGAN) to predict the flow field in the wake of random roughness patches. The model successfully captured the characteristic cross-flow-like vortex, but its fidelity needs to be improved for detailed transition calculations. However, the model proved effective as a screening tool for identifying roughness patches likely to induce or suppress cross-flow-like vortices. By estimating vortex size and location, the model aids...     »