Predicting drag on rough surfaces by transfer learning of empirical correlations

Recent developments in neural networks have shown the potential of estimating drag on irregular rough surfaces. Nevertheless, difficulty obtaining a large high-fidelity dataset to train is deterring their use practical applications. In this study, we propose transfer learning framework model surfaces even with limited amount direct numerical simulations. We show that empirical correlations, reported literature, can significantly improve performance for prediction. This because correlations include ‘approximate knowledge’ dependency physics. The allows learn surface statistics known affect more efficiently. developed be applied applications where acquiring difficult but been reported.