Application Of GPU-based Acoustic Parallel Computing In Automobile Body

In the numerical simulation of acoustics,it is a widely used method to use the finite element method to discretize the acoustic Helmholtz equation to solve the acoustic problem.However,the numerical dispersion error leads to the low accuracy of the acoustic finite element and is only suitable for low frequency acoustic problems.To this end,many methods for improving finite elements have been successively proposed and applied to acoustics.Among them,the typical ones are smooth finite elements and gradient weighted finite elements.They can effectively improve the accuracy of acoustic numerical simulation,but they also exist the problem of low computational efficiency.In this paper,we study the parallel calculation method based on graphics processor(GPU)for the internal sound field problem and the sound-solid coupling problem,and realize the high-speed solution of large-scale acoustic problems using acoustic smooth finite elements on personal computers.The results are:(1)Combining the theoretical basis of acoustic smooth finite element and elementby-element(Element by Element,EBE),a parallel strategy of smooth domain by smooth domain(SBS)is developed.GPU thread and smooth finite element calculation are fused by establishing an abstract mapping mode of thread and smooth domain.Accomplished fine-grained parallel calculation of acoustic smooth finite element;and developed semi-parallel construction method of smooth domain and degree of freedom(DOF)pre-indexing method,realized high-efficiency construction of smooth domain,avoided thread writing problem,and calculated matrix for smooth domain,provided index information to equation solving and data storage,which effectively improves the efficiency of GPU parallel computing.(2)The matrix-by-matrix(MBM)parallel strategy was developed on the basis of the element-by-element parallel strategy,which realized the fine-grained parallel calculation of the structural unit,the smooth field of the sound field,and the coupling boundary in the acoustic-solid coupling system.Corresponding matrix semi-parallel construction method and asymmetric degrees of freedom and pre-indexing method.(3)On the basis of the matrix-free method,the parallel Jacobian-PCG conjugate gradient algorithm(Jacobi-PCG)on the GPU is developed based on the smoothdomain-by-smooth-domain parallel strategy.Since the acoustic-solid coupled system equation is an asymmetric equation,based on the matrix-free method,combined with the matrix-by-matrix parallel strategy,the parallel bi-conjugate gradient(Bi CG)algorithm applied to the GPU was developed,which greatly improved the system equations Solution efficiency.(4)The full-process fine-grained parallel GPU internal sound field and sound-solid coupling parallel computing solution was developed,and the efficiency was verified on the lower cost CPU / GPU heterogeneous computing platform.In the 1.5 millionelement two-dimensional internal sound field problem,an acceleration effect of 7.42 times can be obtained,and in the 2 million-element three-dimensional internal sound field problem,the acceleration ratio can reach 10.54.In the sound-solid coupling problem of 200 million-element,the acceleration ratio can be obtained 16.34.