A data-driven deep learning approach for predicting separation-induced transition of submarines

Abstract

Separation-induced transitions can affect the hydrodynamic performance of submarines significantly. The onset of this transition plays a critical role in the design of aircrafts and underwater vehicles. Since the transition is affected by various factors, its prediction is a challenging task. Convolutional neural networks (CNNs) in the field of machine learning can extract features from high-dimensional data automatically and have good generalization performance. In this paper, we propose an end-to-end CNN-based model to automatically extract the features of separation-induced transitions by learning from image-expressed flow field data. The proposed data-driven deep learning model employs a high-resolution network, which is widely used in key-point detection in the field of computer vision, to extract the underlying features in the separation-induced transition under only a few empirical assumptions. A novel representation of separation-induced transition onset in the form of a heatmap is especially proposed to indicate the probability of transition onset. We use implicit-large-eddy-simulation data generated by the second-order discontinuous Galerkin method over Lyu's-1 model line submarine to verify the proposed method, and the results demonstrate that the new method is able to predict separation-induced transition onsets with quantified uncertainty. The proposed model can be used as an auxiliary tool for aerodynamic and hydrodynamic designs.