柔性仿生非线性流固耦合系统深度学习降阶模型研究

The intelligent flexible bio-inspired aircraft that actively utilizes the fluid-structure interaction effects is one of the important development directions of the future disruptive conceptual aircraft. The existing system identification based reduced-order model and flow-feature decomposition based reduced-order model are difficult to accurately predict the nonlinear dynamic behavior of the flexible bio-inspired structure in the complex unsteady flow field under large deformation or large motion. This project aims at quickly and accurately predicting the nonlinear dynamic behaviors of flexible bio-inspired fluid-structure coupled systems with multi-scale flow phenomena such as the flow separation,vorticity movement and etc. The deep neural network based reduced-order model for unsteady flow fields will be explored.The modeling strategy of the deep learning reduced-order model (DLROM) for nonlinear fluid-structure coupled system is investigated. The deep learning reduced-order models suitable for two-dimensional and three-dimensional nonlinear fluid-structure coupling system of flexible bio-inspired structure are proposed. The DLROM will provide a new powerful method and tool for the rapid prediction of the nonlinear dynamic behaviors, control modeling and optimization design of flexible bio-inspired nonlinear fluid-structure coupling systems.The innovations of this project are: fluid-structure coupled deep learning reduced-order model with combined neural network based on the traditional deep convolution neural network and non-convolution network; the construction method of 3D deep convolution network based reduced-order model for nonlinear Bio-inspired fluid-structure coupling system.

柔性翼翅和柔性身体所蕴藏的丰富流固耦合机理和气动弹性效应是昆虫、鸟类和鱼类获得非凡飞行游走性能的关键因素之一。主动利用流固耦合效应的智能柔性仿生飞行器是未来颠覆性新概念飞行器的重要发展方向。现有系统辨识和特征分解流固耦合降阶模型难以对柔性仿生结构在复杂非定常流场中的大幅大变形运动非线性动力学行为进行准确预测。本项目针对含有流动分离和漩涡等多尺度流动现象的柔性仿生流固耦合系统非线性动力学行为快速准确预测难题，研究非定常流场深度神经网络降阶模型建模方法，探索非线性流固耦合系统深度学习降阶模型建模策略与框架，提出适用二维和三维柔性仿生流固耦合系统的深度学习降阶模型，为智能仿生变形飞行器等柔性仿生非线性流固耦合系统动力学行为的快速预测、控制建模和优化设计提供新方法和工具支撑。本项目创新点为：深度卷积网络和非卷积网络复合的流固耦合深度学习降阶模型；非线性流固耦合系统三维深度卷积网络降阶模型构造方法。

本项目针对柔性仿生流固耦合系统非线性动力学行为快速准确预测难题，研究了非定常流场深度神经网络降阶模型建模方法。开展了深度神经网络流场降阶及高维重构的研究，对不同时刻的非定常流场进行降阶和重构，并采用该方法对低速非定常流场进行时空预测。通过构造并改进混合深度神经网络，还实现了对高速跨声速抖振这一强非线性复杂非定常流场的快速准确预测。同时，发展了一种新的混合深度神经网络，实现了对运动边界周围非定常流场的快速准确预测，研究了神经网络参数对预测结果的影响。最终建立了非线性流固耦合系统深度学习降阶模型建模策略与框架，提出适用柔性仿生流固耦合系统的深度学习降阶模型，实现了对不同结构参数流固耦合系统响应的快速预测。此外，本研究还发展了一种仿生柔性结构非定常流固耦合运动的可并行高精度IB-LB-FEM数值模拟方法，并开展了仿生柔性襟翼流动控制机理研究以及气动噪声机理研究。研究成果为智能仿生变形飞行器等柔性仿生非线性流固耦合系统动力学行为的快速预测、控制建模和优化设计提供新方法和工具支撑。