Assessment of Multifidelity Surrogate Approaches for Expedient Loads Prediction in High-Speed Flows

What is it about?

This study explores a method to quickly and effectively predict the airflow around hypersonic flight vehicles, which is crucial for designing these vehicles and controlling them in flight. We examine two types modeling techniques that learn from detailed simulations or experiments to make fast predictions. One is based on a statistical approach called Kriging, and the other uses a form of deep learning. Our goal is to see which method offers the best mix of speed, accuracy, and ease of use, especially when we have limited detailed data. We find that while the Kriging method is slightly more accurate and easier to manage on the presented data set, it becomes less practical as the amount of data increases. We discover that highly accurate predictions can be made about the hypersonic flow regime with limited data, and that incorporating data classical engineering methods can enhance their accuracy at minimal extra effort.

Featured Image