Improving RANS Modeling of Heat Transfer in Hypersonic Boundary Layers Using Learning and Inference Assisted by Feature-space Engineering

To address the stagnation in the accuracy of RANS models, data-driven model augmentation frameworks have been developed over the past few years with varying levels of success. This work presents an augmentation procedure which enforces consistency between the learning and prediction environments by simultaneously inferring and learning the model discrepancy during the training process. The approach, termed Learning and Inference assisted by Feature-space Engineering (LIFE), emphasizes a careful introduction of the augmentation, a meticulous design of the features and feature space, and the novel notion of localized learning, improving the generalizability and robustness of the augmentation. This work applies the LIFE framework to augment the Wilcox-2006 k-ω turbulence model to improve heat transfer predictions for hypersonic boundary layers. The structural form of a transport equation related to the Turbulent Prandtl number is inferred and learned. The generalization capabilities of the present approach are evaluated, and the impact of several modeling choices is examined.