功能性造型截面参数化过程的约束方法研究

传统的功能性造型截面参数化方法存在压缩率较低、难以实现近似无损压缩等诸多不足，大大降低了造型优化算法的寻优效率。针对上述问题，本项目选择垂直轴风轮为载体，从风轮截面图像的二值性、高分辨率、稀疏性、强规律性四大特点出发，研究探索功能性造型截面的参数化方法，其主要研究内容是：1）基于拉普拉斯特征图的多尺度自适应近似无损约束方法，可实现高压缩比近似无损重建；2）基于随机扰动的新颖性约束方法，可提高参数化方法所构造流形上小邻域内造型间的相似性，增加新颖解的占比，提高风轮寻优算法的效率；3）基于空间解耦及时间分治的稳定性约束方法，从网络结构以及计算顺序进行统筹规划，以便解决8个损失函数及5大功能模块间的冲突问题。本项目的开展将最终研究实现一套具备风轮新颖性约束的近似无损的多尺度的参数化方法，通过替换风轮专用数据库可用于其他功能性造型优化，所提出的约束方法具备较强的创新性及通用性，值得广泛深入研究。

It is hard to further improve the efficiency of functional shapes optimization algorithms because the compression rate of parametrization methods is too low to reach the near lossless image compression level currently. In this project, we try to improve parametrization methods of the cross-section images of functional shapes based on features of Vertical Axis Wind Turbine (VAWT) blades, a typical kind of functional shapes. More specifically, cross-section images of VAWT blades are binary, high resolution, sparse and following regular patterns all the time. We will develop 1) a multiscale near lossless constraint method with high compression rate based on Laplace feature maps, 2) a novelty constraint method based on stochastic disturbance to increase the similarity of neighboring shapes on the manifold which is created by the parametrization methods, to add more novelty shapes on the manifold and to increase the efficiency of the blade optimization algorithm, 3) a stability constraint method based on decoupled structures and sectionalized train sequences to solve the confliction between 8 loss functions and 5 modules. We will develop a VAWT blades parametrization method which is constrained to generate novelty and near lossless shapes at the end of this project. The developed method can be transplanted to other functional shapes by simply replacing its VAWT blades database. The constraint methods proposed in this project are strong in novelty and generality. They are worthy to be further studied.