Geometric Convolutional Neural Networks – A Journey to Surrogate Modelling of Maritime CFD

Abstract

. Computational Fluid Dynamics (CFD) has become an indispensable tool in the field of engineering design evaluation and optimisation. Existing numerical simulation methods are computationally expensive, memory demanding and time-consuming, thus limiting design space exploration and forbid generative design. In order to overcome these challenges, we propose a deep learning based surrogate modeling in-lieu of CFD simulations. Our proposed framework can predict flow fields (e.g pressure field) on the surface of the geometry as well as any overall scalar parameters (e.g drag force) given a three-dimensional shape input. It can also provide uncertainty quantification over predictions. Finally, we demonstrate that our proposed surrogate modelling does not require pre-processing of the input geometry and also outperforms state-of-the-art models in prediction accuracy. When comparing a dataset on aerodynamic drag of car geometries, we show that our model reduced the error standard deviation by a factor of ≈ 2 . 5 compared to a Gaussian Process-based surrogate model.