数据与物理混合驱动的高速流体快速仿真与细节增强

Facing with the complex high-speed fluid scene, due to the complexity of solving physical equation, the simulation efficiency of physically based method is low. And data-driven simulation method do not need solve the physical equation, but can not show the complex animation . However, for high-speed fluid, it is a challenge to simultaneously describe the multiple coupling interaction with different media, physical states and multi-anisotropic materials with the simple physical model or simple data. The rich local details and the complex properties of physical collision also put forward new challenges for rapid rendering technology..Considering the characters of high-speed fluid with complex media, strong impact, rapid change, more details, etc, this project will focus on the interactive simulation technology of rapid simulation and detail enhancement driven by data and physical model of high-speed fluid. Based on the hybrid expression method integrating with the video data, anisotropic material properties and physical model, the hybrid modeling of fluid volume is performed based on time and space firstly. Based on the iteration of the video data and sampling data, the reverse seeking of animation parameters and the interactive computing for high speed fluid were carried out. Through effectively integrating the geometric constraints and physical characteristics, the preservation and enhancement of local details around fluid surface are built based on the framework of active learning and intelligent analysis. Finally, the implementation of efficient modeling and rapid simulation of typical high-speed fluid coupling and local details are performed based on the algorithm optimization and acceleration technologies, and the evaluation and verification method for fluid simulation also proposed. The research results will be integrated into a parametric simulation framework to expand these applications in the areas of fluid animation and virtual reality, etc.

面对高速流体这种更复杂的流体现象，基于物理的流体仿真方法计算量大难以稳定求解，数据驱动的方法可避免求解复杂物理方程，但动态性表现不够，因此统一的物理模型或者单一的数据无法表达高速流体的复杂运动与丰富细节，物理准确性与仿真快速性的矛盾异常突出。.项目旨在针对高速流体现象的介质杂、互动频、变化快、细节多等特性，基于物理模型与数据混合的思路，研究高速流体的建模与快速仿真技术。重点融合流体的表观数据、材质属性数据与物理模型，研究高速流体的多元混合建模方法；结合视频数据、采样数据与运动特性，进行高速流体的运动参数反求与迭代演化；有效地耦合表面细节与物理规律，实现基于学习和智能分析的局部细节增强；最后基于算法优化和加速，构建评测验证方法，实现典型高速流体现象的高效建模与快速仿真，可应用于虚拟现实和流体动画等领域。

高速流体现象常常蕴含着异构介质、高速变化和丰富细节，其快速真实感仿真是图形学领域极具挑战性的难点问题。传统基于物理的流体仿真方法计算量大、难以稳定求解，数据驱动的方法动态性表现不够，物理准确性和仿真快速性的矛盾十分突出。本项目针对高速流体现象仿真的特点，引入物理模型与多源数据混合驱动的思路，研究了高速流体现象的混合建模与细节增强方法。项目遵循混合表达、演化交互、细节增强、并行绘制的技术路线，结合计算机图形学及深度学习等领域最新发展趋势，重点研究了考虑复杂多相变化的流体物理建模表达、视频数据与物理规律融合的流体混合建模、基于多源数据学习的过程化虚拟场景生成、以及面向流体仿真的并行加速等关键技术与算法，实现数据与物理混合驱动的高速流体快速仿真与细节增强，并将研究成果集成到一个参数化仿真原型系统。.本项目拓展了传统的基于物理模型仿真与数据驱动重建的方法，构建了融合流体属性数据与物理模型的混合建模方法，拓展了物理模型引导的互动迭代演化仿真方法，实现了基于多源数据的流体现象细节重构与增强，最后提出的优化加速算法也保证了大尺度高速流体现象仿真的效率。相关技术有效地将数据驱动下的逆向学习重建思路与基于物理模型的正向图形绘制方法结合起来，可以拓展应用于虚拟现实、行业仿真、数字城市、文化娱乐等领域中虚拟场景的生成与交互仿真。