基于浸入边界法的复杂流动多尺度模拟

发布时间：2018-04-17 14:51

本文选题：浸入边界法 + 大涡模拟　；参考：《中国科学院研究生院(过程工程研究所)》2015年博士论文

【摘要】：含有复杂运动边界的流固耦合(fluid-solid interaction, FSI)问题广泛存在于化工生产过程之中,如搅拌反应釜以及颗粒流体两相流等。尽管计算流体力学(computational fluid dynamics, CFD)在近几十年来有了迅速的发展,但是精确并高效地模拟含复杂运动边界的流动和传递现象仍然是巨大的挑战。近年来,基于欧拉正交网格的浸入边界法(immersed boundary method, IBM)逐渐成为研究此类问题的主要数值方法之一。本论文在有限体积／差分方法框架下改进了浸入边界法及其并行算法,并以此实现了实验室以至工业尺度的搅拌反应釜中流动、传热以及颗粒悬浮等过程的模拟,证明了该方法的有效性并展现了广阔的应用前景。论文的主要内容与创新点如下：在传统的IBM模拟中,无滑移边界条件与速度零散度条件通常不能同时满足。本论文对此提出了改进方法,即：压力修正方程先于体积力源项求解,体积力仅仅散布于浸入边界的内部,同时采用多重直接力方法使无滑移边界条件得到更好的满足,从而避免了上述问题。在标准测试算例中,尤其是在运动边界附近,改进算法能够得到比传统方法更为精确的计算结果。在IBM的实际应用中经常同时遇到固定与运动边界,传统方法对此均采用相同的边界处理方式,难以充分发挥其优势。本论文根据不同壁面的运动特性选取不同的IBM分别处理静止壁面与运动壁面,建立了混合浸入边界法。对比算例表明该方法能显著提高计算效率并保持计算精度。本论文借助于CPU (central processing unit) - GPU (graphic processing unit)并行计算突破了复杂流动模拟中计算量大的瓶颈,GPU计算相对于CPU单核获得了7～20倍的加速。传统的IBM在处理运动边界时欧拉网格与拉格朗日网格变量间的传递算法并行性较差。对此,本论文提出了部分速度插值与体积力散布的方式,在不增加通信量的情况下简化了并行算法。基于这些改进工作并结合大涡模拟(large eddy simulation, LES),高雷诺数复杂流固耦合问题的大规模模拟成为了可能。应用上述改进的IBM,本论文结合大涡模拟研究了Rushton搅拌釜内的湍流流动,并与实验以及其他模拟结果进行了对比,获得了合理的速度场和湍流参数分布,证明该方法是研究搅拌釜中湍流流动的有效方法。该方法还进一步应用于工业尺度搅拌反应釜内高粘度流体的流动与传热模拟,有效预测了反应釜内的流场与温度场信息,为其优化设计提供了依据。论文还将改进的IBM与离散单元法(discrete element method, DEM)结合,并应用CPU-GPU异构超级计算实现了搅拌釜内液固两相流的大规模颗粒解析模拟,阐明了桨型和挡板等几何结构对颗粒分布与运动方式的影响。通过本论文的改进,IBM结合超级计算已能够实现工业规模的含有复杂运动边界系统的有效模拟,并能深入揭示其流动与传递过程的细节,而这方面的进一步发展将为搅拌反应釜等重要工业反应器的优化设计提供强有力的工具。
[Abstract]:Solid coupling containing complex moving boundary flow (fluid-solid interaction FSI) problem exists widely in the chemical production process, such as stirring reactor and particle fluid two-phase flow. Although the computational fluid dynamics (computational fluid, dynamics, CFD) has developed rapidly in recent decades, but the accurate and efficient simulation of complex moving boundary flow and transport phenomena is still a great challenge. In recent years, the immersed boundary method based on Euler orthogonal grid (immersed boundary method, IBM) has gradually become one of the main numerical methods of this problem. In this paper, finite volume / finite difference method under the framework of the improvement of the immersed boundary method and its parallel algorithm, and then the flow stirred tank reactor laboratory and industrial scale, and the particle simulation of heat transfer process, proves that the method is effective and fair Now has a broad application prospects. The main contents and innovations are as follows: in the traditional IBM model, no slip boundary condition and speed zero divergence condition usually cannot be satisfied at the same time. This paper puts forward an improved method, namely: the pressure correction equation before the volume source terms of solution, the internal volume load only scattered in the immersion the border, while using multiple direct force method to non slip boundary condition is better satisfied, so as to avoid the above problems. In the standard test examples, especially in movement near the border, the improved algorithm can get more accurate calculation results than the traditional method. In practical application of IBM often encountered with the moving boundary, the traditional method which adopts boundary treatment in the same way, it is difficult to give full play to its advantages. This paper according to the different characteristics of different wall movement were treated with IBM A stationary wall and wall motion, builds the mixed immersed boundary method. Comparative examples show that this method can significantly improve the computational efficiency and precision. This paper based on the CPU (central processing unit) - GPU (graphic processing unit) parallel computing through complex flow simulation in a large amount of computational bottleneck, GPU compared with the single core CPU obtained a speedup of 7~20 times. The traditional IBM in dealing with moving boundary grid Euler and Lagrange mesh variables transfer between parallel algorithm is poor. Therefore, this thesis proposes some scatter velocity interpolation and volume force, the increase in traffic under the condition of simplified parallel algorithm. These the improvement work combined with the large eddy simulation (large eddy simulation, LES), possible large-scale simulation of high Reynolds number of complex FSI problems. The application of the improved IBM, this paper. Large eddy simulation of turbulent flow in Rushton stirred reactor, and compared with the experimental and other simulation results, obtained the reasonable velocity field and turbulence parameters distribution, prove that the method is effective method of mixing kettle in turbulent flow. The method also should be used to simulate the flow and heat transfer of high viscosity fluid reaction the reactor of industrial scale mixing, effective prediction of reactor flow field and temperature field of information, provides the basis for the optimization design. The paper will also improve the IBM and the discrete element method (discrete element method, DEM and CPU-GPU) with application of heterogeneous supercomputing simulate large-scale particle analysis of liquid-solid two-phase reactor flow mixing, illustrates the influence of impeller type and baffle geometry on the particle distribution and movement patterns. Through the improvement of this paper, combined with the IBM supercomputer has been able to achieve industrial scale The effective simulation of complex motion boundary system and the details of its flow and transfer process can be further revealed. The further development of this aspect will provide a powerful tool for the optimization design of important industrial reactors such as stirred tank reactor.

【学位授予单位】：中国科学院研究生院(过程工程研究所)
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TQ019

【相似文献】