GPU-Based Workflow Accelerates Hypersonic Flow Physics Emulators

Accurately modeling complex physical phenomena at high fidelity and low computational cost is a significant challenge in modern engineering, particularly for hypersonic flows where precise prediction of shock waves is critical. Traditional models and neural emulators often struggle to capture the steep gradients and maintain physical consistency in these extreme conditions. To overcome these limitations, researchers developed a comprehensive GPU-based workflow. This workflow integrates rapid data generation using a differentiable high-fidelity solver (JAX-Fluids) with the training of neural emulators. It also incorporates uncertainty quantification and physics-aware refinement techniques. The framework allows for residual-based refinement, enabling training even when only mesh and input parameters are available, significantly reducing errors and enhancing physical consistency. This approach yields physics emulators that remain reliable and accurate even when applied to conditions outside their initial training distribution, which is a crucial requirement for their deployment in real-world engineering design and analysis.