Neural network flow optimization using an oscillating cylinder

Abstract

Flow behaviors of a downstream object can be affected significantly by an upstream object in close proximity. Combined with the neural network algorithms, this concept is used for flow control in this study to optimize the aerodynamic performance of a downstream object. Flow with an oscillating cylinder placed upstream is systematically studied because there are multiple control parameters that influence the flow dynamics around the downstream object. These control parameters are used as the input factors of a back-propagation neural network, and then a revised genetic algorithm is applied to find the optimal set of control parameters. In the current study, we use an airfoil in a low-Reynolds-number flow as an example to investigate the proposed neural network flow optimization concept. The datasets used to train the neural network are from the computational simulation with a previously validated immerse-boundary method to accommodate the motion of the cylinder. The results show that by optimally placing an upstream moving cylinder, it is possible to enhance the aerodynamic performance of the downstream object. Compared to the reference case, the optimized lift/drag ratio of the downstream airfoil can be achieved 2.4 times of its reference value, while maintaining a relatively high lift coefficient.