| In recent years,with the rapid development of radar technology and the improvement of target detection and recognition technology,especially the role of aircraft carriers,known as "sea hegemons",in national-country confrontation has become increasingly important.Therefore,the requirements for the analysis and calculation of radar cross section are getting higher and higher.Experimental measurement and numerical calculation are the main ways to recognize the electromagnetic characteristics of complex systems.However,due to many problems such as test site,experimental targets and so on,it is not allowed and difficult to carry out accurate experimental measurement in many practical situations.For example,the electromagnetic environment of an aircraft carrier sailing on the sea cannot be simulated in a microwave darkroom.For this reason,electromagnetic numerical simulation analysis has become a modern necessary way to solve such problems,and it plays an increasingly important role in the design and analysis of equipment.Electromagnetic numerical simulation should aim at "high-grade,precision and advanced" to achieve "high precision,less memory,and fast speed".However,high precision and the huge computing resources required for the calculation of the radar cross section of a complex electrically large target are contradictory.Therefore,how to accurately and efficiently calculate the radar cross section of an electrically large target in complex environment under limited computing resources has become a very challenging research at present.Based on the above background,this dissertation aims to quickly and accurately calculate and analyze the electromagnetic scattering characteristics of electric large targets in halfspace with limited resources.We used the multilevel fast multipole algorithm(MLFMA)as the main research tool,combined with parallel computer technology and domain decomposition method,and propose a parallel domain decomposition algorithm based on the octree.This method can accurately and effectively solve a series of practical engineering scattering characteristics calculation problems of electrically large targets in a half-space environment under the condition of limited resources.This main research contents and achievements of this dissertation are as follow:(1)Aiming at the problem of the MLFMA in half-space is that the Green's function of the layered medium is a dyadic form,and each component is a complex Sommerfeld integral.A layering method and the real mirror method are introduced to deal with the half-space reflection,and the half-space condition is calculated by modifying the MLFMA in free space.(2)In order to expand the computational scale of the parallel MLFMA,we compare two domain decomposition algorithms in half-space.One is the non-overlapping domain decomposition method(IE-NDDM),which proposes an explicit boundary condition for the PEC problem to ensure continuous current between sub-domains,and the inter-domain coupling is calculated by field iteration to avoid storage mutual impedance matrix,thereby reducing memory requirements.However,this method needs to manually divide the domain and establish an artificial interface in the sub-domain to ensure the subdomain is closed.The operation is cumbersome and the boundary conditions imposed on the boundary will cause the current continuity between the subdomains to deteriorate and the calculation accuracy to decrease when calculating the low RCS carrier.Therefore,this dissertation proposes another domain decomposition method based on an octree(OT-DDM).This method does not need to divide the domain manually and establish an artificial interface.It can automatically divide the domain by MLFMA's own hierarchical grouping characteristics.In order to ensure the continuity of the current between the subdomains and improve the calculation accuracy,impedance calculation is used to replace the imposed boundary conditions on the boundary of adjacent subdomains.(3)In terms of OT-DDM parallel strategy,we propose a parallel strategy based on field group and source group partition to avoid load balancing problem in the field calculation caused by the traditional plane wave adaptive partitioning strategy.This strategy corrects the problem that the traditional strategy only considers the source group.In the process of parallel task partition,the source group and the field group are partitioned into tasks respectively to ensure that the tasks of each process are basically equal when calculating the field effect between sub-domains.Meanwhile,in order to accelerate the calculation efficiency of field action calculation and save computing resources,this dissertation proposes a near field and far field partitioning strategy.In this strategy,the method of moments is used to calculate the near-field self-action,MLFMA is used to calculate the interaction,and FAFFA is used to accelerate the calculation of the far-field interaction.This strategy breaks through the memory limitation when calculating the electrically large-scale model,which can accelerate the calculation process and reduce the memory consumption on the premise of ensuring the accuracy.(4)Aiming at the parallel computing process between domains in the OT-DDM algorithm,this dissertation studies different parallel modes for workstation windows system and cluster Linux system.The research found that there is no dependency relationship between the calculation of each subdomain,so a strategy of simultaneous calculation of multiple subdomains in multi processes can be adopted.For the workstation windows system,we present a parallel mode based on process groups.This method can automatically partition and calculate each subdomain through MPI.For the cluster Linux system,with the help of the cluster efficient task scheduling system,we present a task-level parallel mode.This method is similar to the task partitioning strategy of the process groups parallel mode.The task scheduling system is used to submit each subdomain computing task,and the parallel task is guaranteed to be performed uniformly by means of indicating file feedback.(5)This dissertation presents a subdomain preconditioning method for accelerated iteration of OT-DDM.This method accelerates the convergence speed of the iterative process by constructing preconditions in the subdomains themselves.And a MPI hybrid Open MP strategy is studied that accelerates the field calculation between subdomains to accelerate the calculation efficiency of OT-DDM.(6)This dissertation researches how to reduce the memory consumption of OT-DDM from the algorithm itself and the hardware architecture.In terms of algorithms,the functions of real-time calculation and interpolation calculation of half-space transfer factor are realized.The half space transfer factor,which needs a lot of memory in the calculation process,is reduced by the way of top-level interpolation and bottom-level real-time calculation,so that the memory consumption of OT-DDM is effectively reduced,and the goal of using a small amount of resources to calculate the electrical large-scale model is achieved.Based on the hardware architecture,we use the out-of-core technology and parallel I/O technology to use hard disk storage instead of memory to convert memory consumption into hard disk consumption.In summary,this dissertation summarizes and analyzes the research of domestic and foreign scholars,and the parallel MLFMA and domain decomposition method in half space are studied in depth and systematically.Finally,a novel domain decomposition algorithm based on octree is proposed.This algorithm can accurately and effectively solve the calculation of the scattering characteristics of large-scale targets in half-space under the condition of limited resources,expands the application range of the MLFMA in practical engineering,and improves the scale of the solution.It provides a calculation guarantee for practical projects in the relevant technology fields. |
PDF Full Text Request