Enhancing Computational Fluid Dynamics through Unsupervised Machine Learning
Please login to view abstract download link
Unsupervised Machine Learning efficiently identifies patterns in data without guidance, proving invaluable in Computational Fluid Dynamics (CFD) to identify diverse flow regions. Our research showcases Gaussian Mixture Models’ (GMMs) proficiency in segregating viscous and inviscid areas, as well as identifying shock-containing zones. We highlight the utility of isolated viscous and inviscid regions for mesh adaptation, and the generation of shock-capturing schemes within areas containing shocks, obviating the need for case-specific thresholds. Implemented within the Horses3D framework, our methodology is validated across simple geometries with and without shocks. Additionally, we extend these findings to practical applications like incompressible adaptive mesh refinement in wind turbine cases, emphasizing the method’s practicality and robustness.
