Development Of A Parallel Algorithm And A Customized Software For Response Simulation Of Ceramic Composites Based On Domain Decomposition Methods

Continuous fiber reinforced ceramic matrix composites(CFRCMC)are widely applied in key components such as engine hot end parts and casing parts and so forth,in the aerospace industry due to their low density,long life,and excellent mechanical and chemical properties in extremely high temperature environments.For a long time,research and development and performance tuning of such materials have mostly adopted the trial-and-error method,which has problems such as long research and development cycle,high economic cost,and great contingency,and it is difficult to quantitative optimization,these factors have led to the slow progress of research in this discipline.In recent years,with the improvement of computer performance and the continuous improvement of theories such as micro-mechanics and computational-mechanics,the research of CFRCMC has gradually changed from the trial-and-error method to the method of combining numerical simulation to design microstructure and predict macroscopic properties with experimental verification results,which effectively improves the efficiency of composite material research and development and greatly shortens the period from material development to actual service.CFRCMC has a variety of cross-sectional shapes and braiding methods of fiber bundles at the mesoscale,and there are a large number of natural cracks and pores,resulting in an extremely complex microstructure.In order to obtain accurate numerical results,the scale of the simulation model needs to be increased,and the performance of the computer processor is relatively high.In the past,the numerical simulation of the performance of such materials usually adopted serial program calculation,relying on the improvement of single-core performance to improve the solution efficiency,but in the past 15 years,the clock speed of CPU has been increased by less than 30%,resulting in limited single-core performance improvement.The way to do this was to increase the number of cores,and now every computer is a parallel machine.Therefore,focusing on parallel algorithms can effectively utilize all computing resources to solve large-scale problems.The domain decomposition methods perform large-scale computing domain decomposition and parallel computing in the form of domain blocks,and have the characteristics that different sub-domains can be solved independently.They are very suitable for the current CFRCMC large-scale numerical simulation problem.Aiming at how to mobilize the existing computing resources to improve the solution efficiency of large-scale numerical simulation problems of CFRCMC,this paper proposes a solution that applies the domain decomposition methods to the effective modulus prediction and transient analysis of CFRCMC.Make the numerical implementation of the algorithm,verify its accuracy,and analyze its parallel evaluation indicators.The research of the dissertation mainly includes the following four aspects:(1)Theoretical derivation of a domain decomposition method and its linear algebra iterative operator improving: Based on domain decomposition theory,a parallel solution strategy for governing equations of 3D continuum finite element quasi-static and dynamic problems is established.The existing iterative process algebraic operator solution methods are improving according to the types of governing equations,so that the algorithm can be better combined with the current large-scale linear equation solvers to improve the efficiency of solving the Dirichlet boundary problems.(2)Apply the established domain decomposition parallel strategy to the prediction of macroscopic mechanical properties of CFRCMC: Firstly,write the parametric modeling program of the minimum period unit of CFRCMC.Secondly,in the establishment of the finite element model,the multi-thread parallel calculation of the element stiffness matrix is accomplished by combining with Open MP.After that,on the basis of the displacement response serial solver developed and verified by the research group,the domain decomposition tool METIS is adopted to block the minimum period unit model,and the Restricted Additive Schwarz(RAS)algorithm is adopted to transfer the equation solver to solve the displacement response of the minimum period unit of CFRCMC in parallel.Then,compare the serial solution results to verify the correctness and accuracy of the parallel solution.Finally,the parallel solution performance is quantitatively evaluated.(3)Apply the established domain decomposition parallel strategy to the transient response simulation of the CFRCMC wing model: Firstly,the wing model is established according to the selection of the wing section.Secondly,the wing model is meshed and its finite element model is established.After that,After that,on the basis of the serial solution program for transient analysis developed and verified by the research group,METIS is adopted to block the wing model,and the RAS algorithm is adopted to call the equation solver to solve the displacement response of the CFRCMC wing model in parallel.Then,the serial solution results are compared to verify the correctness and accuracy of the parallel solution,and the performance of the parallel solution is quantitatively evaluated.Finally,it is compared with the different parallel efficiencies in the effective modulus prediction,and the reasons are analyzed.(4)Development of simulation software for CFRCMC: Considering the scalability of program functions,cross-platform development and solution efficiency,this paper selects C++ language,Qt development platform,and VTK visualization toolkit to develop a simulation software CFRCMC(V2.0)with parametric modeling,effective modulus prediction and transient analysis functions.Finally,the digital design and rapid performance prediction of CFRCMC microstructure are accomplished.