Deep learning based real-time energy extraction system modeling for flapping foil

Considering the increasing energy consumption and greenhouse gas emissions, the promising energy extraction system via flapping foil for flow and wind energy has attracted more and more attention. Due to the expensive computation resource and time cost, the CFD method impedes the realization of real-time modeling for flapping foil. The surrogate model by machine learning is a promising alternative, but it only focuses on the objective functions and ignores the importance of physical fields. Aiming at providing a comprehensive model to predict the aerodynamic characteristics as well as the physical fields, a deep learning based real-time model containing two modular convolutional neural networks are devised in this paper. With the numerical simulations as training dataset, a well-trained model can accurately predict the pressure and velocity fields as well as the lift and moment coefficients in millisecond. Moreover, the global sensitivity analysis and the optimizations are conducted based on this model. By leveraging the automatic differential mechanics in deep learning method, the time consumption for kinematic optimization is accelerated into a minute, which further demonstrates the real-time capability. Overall, the presented deep learning model can provide a reliable and competitive choice for the digital twin of flapping foil energy extraction system.