数据驱动下流体结构相互作用的卷积神经网络预测

Data-driven prediction of fluid-structure interaction based on CNN model

卷积神经网络(CNN:convolutional neural network)是人工智能领域中具有代表性的深度学习算法之一,在图像识别和音频处理等问题上具有良好表现。该文将CNN用于计算流体力学(CFD:computational fluid dynamics)中流体结构相互作用问题的求解,基于数据驱动(Data-Driven)的理念,跳过传统流体控制方程的计算步骤,实现流场结果的快速预测。文中以Navier-Stokes方程的计算结果作为训练集,使CNN预测模型自主学习流体结构相互作用规律与特征关系,并对带有结构物的顶盖驱动方腔流以及河道沙洲绕流等问题开展预测模拟。结果显示,基于卷积神经网络的计算模型在流速、压强的预测以及流场形态的预测方面都有良好的表现。该理念能够在基本保障精度的前提下大幅提高效率,可为流场求解提供一种新思路。

Convolutional neural network(CNN)is one of the representative deep learning algorithms in the field of artificial intelligence,which has good performance in image recognition,audio processing and other issues.In this paper,CNN is used to solve the fluid-structure interaction problems in the computational fluid dynamics(CFD)field.Based on the concept of data-driven,the calculation steps of traditional fluid control equations is skipped and the fast prediction of flow field results is realized.Some results of Navier Stokes equations are used as the training set,and CNN prediction model can learn the fluid structure interaction laws and characteristics independently.Several cases are simulated including lid-driven cavity flow with structures and flow past channel sandbanks.The results show that the convolution neural network-based model has a good performance in the prediction of velocity and pressure field.Present work provides an alternative way to solve fluid-structure interaction problems,which can greatly improve the calculation efficiency under the premise of basic guarantee accuracy.