Key Technology Research On Lagrangain Coherent Structures Visual Computing Of Unsteady Flow Field

Flow field visualization is an important means to analyze flow field data obtained by numerical simulation and experiment.With the development of high performance computer capability and experimental data production technology,the flow field data set continously enlarges.Especially in numerical simulation of unsteady flow,the generated data set can reach GB,TB and even PB bytes.In order to effectively carry out the analysis and research of the flow properties and mechanism,the main features and structures in the flow field must be extracted with the help of visual computing.Coherent structure is an important feature in flow field and also a common method for topology analysis,which has meaningful significance in transport properties and turbulence research.The methods based on Euler framework are well applicable in steady flow field.However,these methods have objective and physical flaws when applying to unsteady flow field.The theory and method of coherent structure based on Lagrangian framework becomes a hot topic in relative disciplines that draws more and more attention of visualization researchers.The Lagrangian Coherent Structure(LCS)research is still developing in mainly three aspects.First,the mathematical model accuracy needs theoretical improvement.Second,the role of LCS in specific problem analysis and research is being explored in application usage.Third,efficient computing algorithm and visual expression are developing for visualization.This dissertation researches on the computing precision,accuracy,adaptation and efficiency of LCS visual computing.The major contributions and innovations are summarized as follows.1.For the extraction precision problem of Lagrangian vortex based on Lagrangian Average Vorticity Deviation(LAVD),a high-precision Lagrangian vortex visual feature extraction algorithm is proposed.The dissertation systematically investigates the influence on computation result due to numerical simulation with different precision,and demonstrates the relationship among the visual computing result,numerical method and the flow field data accuracy.With these derivation,a high-precision principle is deduced.The Lagrangian vortex extraction precision is dramatically improved by using WENO-based grid point vorticity solving and stencil smoothness probing based off-grid point data interpolation strategy.The result of the experiment verifies the conclusion of precision analysis and high precision feature of proposed algorithm.By comparing with the extraction results of two Euler based methods,the proposed algorithm demonstrates considerably effectiveness and accuracy.2.For the adaptive and efficiency problem of existing LCS visual feature extraction,an adaptive and efficient algorithm is proposed for hyperbolical and elliptic LCS visual feature extraction.This algorithm unifies LCS visual computing on flow field of different grids and data types by using a gradient-based interpolation method.By adopting computing method based on interpolation and advection integration,the computation efficiency is improved.Moreover,different strategies are adopted during various LCS visual feature extraction steps to further reduce the computational cost.Experimental results show that the proposed algorithm can achieve 6.4~10.6 speedup ratio of single-thread execution compared with 4th-order Runge-Kutta(RK4)based method in unsteady flow field's LCS visual feature extraction,and is liable to OpenMP accleration.3.For the adaptive and efficiency problem of Finite Time Lyapunov Exponent(FTLE)sequence analysis on unsteady flow field,an efficient FTLE sequence visual computing algorithm is proposed based on preprocessing and composition policy.By using integration advection based time interval flow mapping and efficient flow mapping composition,the proposed algorithm acomplish FTLE sequence visual computing with controllable error and none extra memory consumption.The results show that the proposed algorithm can achieve 51.4~163.7 speedup ratio compared to existing RK4 based FTLE visual computation.It demonstrates that the proposed algorithm can generate forward and backward FTLE visualization images on flow field with million grid cells in real-time by GPU acceleration.4.To meet the scalability requirement of large-scale FTLE visual computing,a highly scalable heterogeneous parallel algorithm is proposed.A partial differential equation based CPU and Intel Many Integrated Core(MIC)collaborative parallel computing approach is adopted,which is suitable for partition-based parallel computing.The proposed algorithm can make full use of the computing power of various processors and data zones can be divided statically.It also has fixed communication mode and quantity,and is capable of improving the scalability and parallel visual computing performance.The experiment results show it can achieve more than 81% parallel computing efficiency in a single MIC calculation.The experimental result from 1~4 heterogeneous node with MIC and CPU on Tianhe-II super computer show that the proposed algorithm has good parallel efficiency and scalability.