Research On Key Technology Of The Integrated LBM-FVM Scheme For Massive Parallel Simulation

With the development of high performance computing,the scientific computing aimed at addressing the large-scale scientific and engineering problems becomes one of the third powerful tools for exploring natural world after the theoretical analysis and the experimental study.As an important branch of scientific computing,computing fluid dynamics(CFD)faces not only enormous opportunities brought about by the promotion of high performance computing platforms,but also grate challenges from the demanding of the computing resources and efficiency when solving large-scale three dimensional industrial CFD applications.The numerical scheme and the application framework are two influencing factors in the development of CFD applications: on one hand,the numerical scheme is the key factor for high efficient parallel simulation,however,a single numerical scheme could hardly give consideration to effectiveness,efficiency and scalability as a whole.on the other hand,the application framework is a significant infrastructure for high efficient parallel simulation,however,most of the simulation frameworks now can not meet the needs of the large-scale,cross-domain,and multiple numeral schemes application.Taking the simulation of viscoelastic fluid flows as a cutting point,we carry out our research on key technology of the integrated LBM-FVM method for massive parallel simulation for the first time.Based on some well-established and promising numeral schemes,the integrated LBM and FVM method for the simulation of viscoelastic fluid flows is proposed.Then the integrated LBM-FVM framework is designed upon the integrated numeral scheme and some open source CFD frameworks.The parallel decomposition and the coupled simulation modules in the framework are optimized with the adaptive sparse block graph decomposition method for the coupled simulation and the hybrid multi-level parallel scheme to improve the efficiency and scalability.Finally,compressive simulations for typical benchmark flows are performed upon the integrated framework to validate the integrated numerical scheme.The integrated LBM-FVM numerical scheme can introduce the efficiency and the stability of LBM scheme,while maintaining the effectiveness and generality of FVM scheme.The hierarchy integrated LBM-FVM framework can shield the underlying platform implementation details,facilitate the cross-domain cooperation and the efficient integration of two numerical schemes,and achieve better performance as consequence.The main idea and its implementation of the integrated LBM-FVM scheme can also be extended to other applications involving integrating other numerical schemes.Therefore,the research on the key technology of the integrated LBM-FVM simulation is of great significance for promoting the development of CFD parallel frameworks and other coupled simulation technics.The main work and innovation of the thesis are summarized as follows:· The integrated LBM-FVM numerical scheme is proposedThe main idea of integrated LBM-FVM numerical scheme is proposed based upon the viscoelastic fluid flows application,and the LBM and FVM scheme,in which the LBM and FVM scheme are integrated to simulate the incompressible NavierStokes equations and the constitutive equation respectively: The parameter conversion process and conversion equations between LBM and FVM process are strictly defined under the same Reynolds Number condition;The spatial mapping scheme for LBM and FVM meshes are constructed for different spatial dimensions and boundary conditions;Synchronous and asynchronous time-stepping schemes are constructed for the simulation of viscoelastic fluid flows;Finally,a generally numerical algorithm framework for the integrated LBM-FVM scheme is established for the simulation of viscoelastic fluid flows.With the integrated numerical scheme, researchers can give precise prediction to the viscoelastic fluid flows,and improve the efficiency and scalability of the simulation.The integrated scheme can also be extended to other applications involving LBM scheme and other macroscopic numerical scheme.· The integrated LBM-FVM simulation framework is designedBased upon the integrated LBM-FVM numerical scheme and the well-established open source CFD frameworks,the integrated LBM-FVM simulation framework is designed: The high-cohesion loose-coupling hierarchial structure is introduced in to assemble the core functional modules of LBM calculation,FVM calculation,spatial mapping,and parameter conversion,with the well-defined module programming interfaces and the module configuration file formats.The SPMD parallel scheme based on mesh block decomposition and data field format is designed for massive parallel integrated simulation,and the parallel partition toolkits for the integrated simulation and the general IO scheme are also constructed for the framework.The integrated LBM-FVM simulation framework can simplify the developing of the integrated viscoelastic fluid flows solvers,achieve cross-domain cooperation,and facilitate massive parallel simulation and related optimization.· The adaptive sparse block graph decomposition scheme for the integrated framework is constructedA adaptive sparse block graph decomposition scheme(ASBMetis)is designed according the characteristics of the integrated LBM-FVM simulation process by overall consideration of the integrated numerical scheme,the load balance,and the communication overhead.ASBMetis adaptively decomposes the mesh of the simulation domain into atomic blocks,and assigns the blocks to different processors with two consecutive steps of the preliminary decomposition and the graph decomposition in an iterative way,in which the computation overhead in block and the communication overhead between blocks are abstracted as a graph format.The blocking and decomposition operation are organized as an iterative process to meet the preset decomposition quality requirement.Experiments show that ASBMetis always converges at the preset decomposition quality requirement under different geometry, and its load balance and communication overhead is greatly superior to that of the original multi-block or sparseblock scheme.· The multilevel hybrid parallelization scheme for the integrated framework is formulatedThe time overhead of FVM calculation,LBM calculation and the coupling operation involved in the integrated LBM-FVM simulation is analyzed to explore the finergrained parallelism after the top-level parallelism based upon mesh sub-domains,including: the task-level parallelism in the partial differential equation discretization and the data-level parallelism when solving the linear system in FVM calculation,the data-level parallelism based upon atomic blocks in LBM calculation,and the data-level parallelism in coupling operation.Then the multilevel hybrid parallelization is implemented for different modules with the hybrid MPI and OpenMP programming model,in which different parallelism degree could be set for different OpenMP code block.The hybrid optimization scheme can further improve the resource utilization,the parallel efficiency,and the scalability for the integrated LBM-FVM simulation.· the integrated LBM-FVM viscoelastic solvers are designed and comprehensive validations are performedTaking the simulation of isothermal incompressible viscoelastic fluid flows as a cutting point,two types of algorithms and solvers are designed over the integrated LBM-FVM numerical scheme and the framework,including the ILFVE algorithm and its solver with spatial interpolation and synchronous time-stepping scheme,and the IBLF-dts algorithm and its solver without spatial interpolation and with asynchronous time-stepping scheme.Based on four typical types of viscoelastic benchmark of Poiseuille flow,Taylor-Green vortex,Cavity flow and 4:1 contraction flow,a compressive validation process by comparing the solution of the integrated solver with the analytical solution,the semi-analytical solution,and the numerical solution of FVM PISO algorithm is implemented to validate the effectiveness and efficiency of the integrated numerical scheme.