Convolution neural network for fluid flow simulations in cascade with oscillating blades

Abstract

This paper aims to design a computational model for simulating the unsteady flow field in a cascade of oscillating blades. The core of the new model is a convolutional neural network, which is trained on a simplified cascade consisting of three blades. The primary advantage lies in significantly reducing the computational cost, as the new model is several orders of magnitude faster than traditional CFD methods for evaluations, though training the model remains computationally intensive. The convolutional neural network can accurately predict the unsteady flow field, as demonstrated in validation examples. In the next step, a composition algorithm is proposed to combine several simplified cases, enabling the solution of a cascade with any number of blades.